Line data Source code
1 : #define _GNU_SOURCE
2 : #include <linux/futex.h> /* FUTEX_WAKE */
3 : #include <sys/syscall.h> /* SYS_futex */
4 : #include <unistd.h> /* syscall(2) */
5 :
6 : #include "fd_replay_tile.h"
7 : #include "fd_replay_tile_private.h"
8 : #include "../../ballet/bls/fd_bls12_381.h"
9 : #include "fd_sched.h"
10 : #include "fd_execrp.h"
11 : #include "generated/fd_replay_tile_seccomp.h"
12 :
13 : #include "../admin/fd_adminctl.h"
14 : #include <errno.h>
15 : #include <fcntl.h>
16 : #include "../genesis/fd_genesi_tile.h"
17 : #include "../poh/fd_poh.h"
18 : #include "../poh/fd_poh_tile.h"
19 : #include "../votor/fd_votor_tile.h"
20 : #include "../resolv/fd_resolv_tile.h"
21 : #include "../restore/utils/fd_ssload.h"
22 : #include "../restore/utils/fd_wfs.h"
23 : #include "../../flamenco/gossip/fd_gossip_message.h"
24 :
25 : #include "../../disco/tiles.h"
26 : #include "../../disco/fd_txn_m.h"
27 : #include "../../disco/shred/fd_fec_set.h"
28 : #include "../../disco/shred/fd_shred_tile.h"
29 : #include "../../disco/pack/fd_pack.h"
30 : #include "../../disco/pack/fd_pack_cost.h"
31 : #include "../backup/fd_snapmk_tile.h"
32 : #include "../../disco/keyguard/fd_keyload.h"
33 : #include "../../disco/genesis/fd_genesis_cluster.h"
34 : #include "../../discof/genesis/genesis_hash.h"
35 : #include "../../util/pod/fd_pod.h"
36 : #include "../../util/fd_boolau.h"
37 : #include "../../flamenco/rewards/fd_rewards.h"
38 : #include "../../flamenco/leaders/fd_multi_epoch_leaders.h"
39 : #include "../../flamenco/stakes/fd_stakes.h"
40 : #include "../../flamenco/progcache/fd_progcache_admin.h"
41 : #include "../../flamenco/rewards/fd_rewards.h"
42 : #include "../../disco/metrics/fd_metrics.h"
43 : #include "../../flamenco/events/fd_event_runtime.h"
44 : #include "../../disco/events/generated/fd_event_gen.h"
45 : #include "../rotor/fd_rotor_tile.h"
46 : #include "../../flamenco/runtime/fd_runtime.h"
47 : #include "../../flamenco/runtime/fd_runtime_stack_tmpl.h"
48 :
49 : #include "../../flamenco/runtime/sysvar/fd_sysvar_cache.h"
50 : #include "../../flamenco/runtime/sysvar/fd_sysvar_stake_history.h"
51 : #include "../../flamenco/runtime/sysvar/fd_sysvar_clock.h"
52 : #include "../../flamenco/runtime/sysvar/fd_sysvar_epoch_schedule.h"
53 : #include "../../flamenco/runtime/sysvar/fd_sysvar_rent.h"
54 : #include "../../flamenco/runtime/program/fd_precompiles.h"
55 : #include "../../flamenco/runtime/program/vote/fd_vote_state_versioned.h"
56 : #include "../../flamenco/runtime/program/vote/fd_vote_codec_tmpl.h"
57 : #include "../../flamenco/runtime/tests/fd_dump_pb.h"
58 : #include "../../disco/events/fd_event_report.h"
59 : #include "../../disco/metrics/fd_metrics_sample.h"
60 :
61 : /* Replay concepts:
62 :
63 : - Blocks are aggregations of entries aka. microblocks which are
64 : groupings of txns and are constructed by the block producer (see
65 : fd_pack).
66 :
67 : - Entries are grouped into entry batches by the block producer (see
68 : fd_pack / fd_shredder).
69 :
70 : - Entry batches are divided into chunks known as shreds by the block
71 : producer (see fd_shredder).
72 :
73 : - Shreds are grouped into forward-error-correction sets (FEC sets) by
74 : the block producer (see fd_shredder).
75 :
76 : - Shreds are transmitted to the rest of the cluster via the Turbine
77 : protocol (see fd_shredder / fd_shred).
78 :
79 : - Once enough shreds within a FEC set are received to recover the
80 : entirety of the shred data encoded by that FEC set, the receiver
81 : can "complete" the FEC set (see fd_fec_resolver).
82 :
83 : - If shreds in the FEC set are missing such that it can't complete,
84 : the receiver can use the Repair protocol to request missing shreds
85 : in FEC set (see fd_repair).
86 :
87 : - The current Repair protocol does not support requesting coding
88 : shreds. As a result, some FEC sets might be actually complete
89 : (contain all data shreds). Repair currently hacks around this by
90 : forcing completion but the long-term solution is to add support for
91 : fec_repairing coding shreds via Repair.
92 :
93 : - FEC sets are delivered in partial-order to the Replay tile by the
94 : Repair tile. Currently Replay only supports replaying entry batches
95 : so FEC sets need to reassembled into an entry batch before they can
96 : be replayed. The new Dispatcher will change this by taking a FEC
97 : set as input instead. */
98 :
99 : /* Agave defines TIME_TO_COMPLETE_BLOCK_BROADCAST to adjust their slot
100 : end. Our own telemetry data of finishing pack => shredding last
101 : batch suggests roughly the same delay, so we mirror the constant. */
102 :
103 0 : #define AG_TIME_TO_COMPLETE_BROADCAST_NS (6000000L)
104 :
105 0 : #define IN_KIND_SNAP ( 0)
106 0 : #define IN_KIND_GENESIS ( 1)
107 0 : #define IN_KIND_IPECHO ( 2)
108 0 : #define IN_KIND_RESOLV ( 4)
109 0 : #define IN_KIND_POH ( 5)
110 0 : #define IN_KIND_EXECRP ( 6)
111 0 : #define IN_KIND_REPAIR ( 7)
112 0 : #define IN_KIND_TXSEND ( 8)
113 0 : #define IN_KIND_RPC ( 9)
114 0 : #define IN_KIND_GOSSIP_OUT (10)
115 0 : #define IN_KIND_SNAPMK (11)
116 0 : #define IN_KIND_ADMIN (12)
117 0 : #define IN_KIND_VOTOR (13)
118 :
119 : #define DEBUG_LOGGING 0
120 :
121 : /* The first bank that the replay tile produces either for genesis
122 : or the snapshot boot will always be at bank index 0. */
123 0 : #define FD_REPLAY_BOOT_BANK_IDX (0UL)
124 :
125 : /* wfs_resolve_leader_gate_auto resolves to true when
126 : wait_for_vote_to_start_leader is configured as auto, otherwise it
127 : passes the config value through. True keeps the leader gate armed,
128 : which is correct for a non-WFS boot and for NOOP. */
129 :
130 : static inline int
131 0 : wfs_resolve_leader_gate_auto( int wait_for_vote_to_start_leader ) {
132 0 : return fd_int_if( wait_for_vote_to_start_leader==FD_BOOLAU_AUTO,
133 0 : FD_BOOLAU_TRUE,
134 0 : wait_for_vote_to_start_leader );
135 0 : }
136 :
137 : static inline ulong
138 0 : fd_block_id_ele_get_idx( fd_block_id_ele_t * ele_arr, fd_block_id_ele_t * ele ) {
139 0 : return (ulong)(ele - ele_arr);
140 0 : }
141 :
142 : static inline fd_block_id_ele_t *
143 : fd_block_id_ele_query( fd_replay_tile_t * ctx,
144 : fd_hash_t const * block_id,
145 0 : ulong slot ) {
146 0 : ag_block_id_t key = ag_block_id( slot, block_id->uc );
147 0 : return fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &key, NULL, ctx->block_id_arr );
148 0 : }
149 :
150 : FD_FN_CONST static inline ulong
151 0 : scratch_align( void ) {
152 0 : return 128UL;
153 0 : }
154 : FD_FN_PURE static inline ulong
155 0 : scratch_footprint( fd_topo_tile_t const * tile ) {
156 0 : ulong chain_cnt = fd_ag_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
157 :
158 0 : ulong l = FD_LAYOUT_INIT;
159 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
160 0 : l = FD_LAYOUT_APPEND( l, fd_genesis_align(), fd_genesis_footprint( fd_genesis_account_max( tile->replay.genesis_max_message_size ) ) );
161 0 : l = FD_LAYOUT_APPEND( l, fd_runtime_stack_align(), fd_runtime_stack_footprint( FD_RUNTIME_MAX_VAT_VOTE_ACCOUNTS, FD_RUNTIME_MAX_STAKED_VOTE_ACCOUNTS, FD_RUNTIME_MAX_STAKE_ACCOUNTS ) );
162 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
163 0 : l = FD_LAYOUT_APPEND( l, alignof(ulong), sizeof(ulong) * tile->replay.max_live_slots );
164 0 : if( FD_UNLIKELY( tile->replay.report_runtime_diffs ) ) {
165 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_hash_t), sizeof(fd_hash_t) * FD_FEC_BLK_MAX * tile->replay.max_live_slots );
166 0 : l = FD_LAYOUT_APPEND( l, 8UL, FD_EVENT_RUNTIME_SLOT_DIFFS_FOOTPRINT * tile->replay.max_live_slots );
167 0 : }
168 0 : l = FD_LAYOUT_APPEND( l, fd_ag_block_id_map_align(), fd_ag_block_id_map_footprint( chain_cnt ) );
169 0 : l = FD_LAYOUT_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
170 0 : l = FD_LAYOUT_APPEND( l, fd_accdb_align(), fd_accdb_footprint( tile->replay.max_live_slots, 0 ) );
171 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_reception_stats_t), sizeof(fd_reception_stats_t)*tile->replay.max_live_slots );
172 0 : l = FD_LAYOUT_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots, tile->replay.max_shreds_per_block, tile->replay.max_txn_per_slot ) );
173 0 : l = FD_LAYOUT_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
174 0 : l = FD_LAYOUT_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
175 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
176 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_event_block_completed_t), sizeof(fd_event_block_completed_t) );
177 0 : l = FD_LAYOUT_APPEND( l, fd_timing_slot_pool_align(), fd_timing_slot_pool_footprint( FD_REPLAY_TXN_TIMING_SLOTS ) );
178 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_replay_txn_timing_t), FD_REPLAY_TXN_TIMING_SLOTS*tile->replay.max_txn_per_slot*sizeof(fd_replay_txn_timing_t) );
179 0 : l = FD_LAYOUT_APPEND( l, alignof(ulong), tile->replay.max_live_slots*sizeof(ulong) );
180 :
181 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
182 0 : l = FD_LAYOUT_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
183 0 : }
184 :
185 0 : l = FD_LAYOUT_FINI( l, scratch_align() );
186 :
187 0 : return l;
188 0 : }
189 :
190 : /* wait_info_healthy returns 1 if the replay tile is healthy: caught up
191 : and not falling behind turbine by more than 12 slots. */
192 :
193 : static inline int
194 : wait_info_healthy( int caught_up,
195 : ulong catch_up_max_fec_slot,
196 0 : ulong reset_slot ) {
197 0 : ulong turbine_slot = catch_up_max_fec_slot==ULONG_MAX ? 0UL : catch_up_max_fec_slot;
198 0 : ulong health_reset = reset_slot==ULONG_MAX ? 0UL : reset_slot;
199 0 : return caught_up && !( turbine_slot>health_reset && turbine_slot-health_reset>12UL );
200 0 : }
201 :
202 : /* query_next_leader_slot returns the next slot at or after from the
203 : one we lead, ULONG_MAX if the schedules hold none. */
204 :
205 : static ulong
206 : query_next_leader_slot( fd_replay_tile_t * ctx,
207 0 : ulong from ) {
208 0 : if( FD_UNLIKELY( ctx->next_leader_query_start==ULONG_MAX ||
209 0 : from<ctx->next_leader_query_start ||
210 0 : from>ctx->next_leader_query_slot ) ) {
211 0 : ctx->next_leader_query_start = from;
212 0 : ctx->next_leader_query_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, from, ctx->identity_pubkey );
213 0 : }
214 0 : return ctx->next_leader_query_slot;
215 0 : }
216 :
217 : /* refresh_epoch_info republishes the epoch fields wait_info exposes.
218 : Call wherever reset_slot moves. */
219 :
220 : static void
221 : refresh_epoch_info( fd_replay_tile_t * ctx,
222 : fd_bank_t const * bank,
223 0 : ulong slot ) {
224 0 : fd_epoch_schedule_t const * sched = &bank->f.epoch_schedule;
225 0 : ulong ep = fd_slot_to_epoch( sched, slot, NULL );
226 0 : ctx->epoch_end_slot = fd_epoch_slot0( sched, ep ) + fd_epoch_slot_cnt( sched, ep ) - 1UL;
227 0 : ctx->slots_per_epoch = sched->slots_per_epoch;
228 0 : ctx->ns_per_slot = bank->f.slot_params.ns_per_slot;
229 0 : }
230 :
231 : /* wait_info_publish writes the wait_info topology object so the wait
232 : command can read a consistent snapshot without touching metrics. */
233 :
234 : static void
235 0 : wait_info_publish( fd_replay_tile_t * ctx ) {
236 0 : if( FD_UNLIKELY( !ctx->wait_info ) ) return; /* not in this topology */
237 :
238 : /* Query the schedule, not ctx->next_leader_slot: there ULONG_MAX
239 : means already consumed, but the wire maps it to 0, which the gap
240 : check reads as no upcoming leader slot and passes. */
241 :
242 0 : ulong highwater = fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot );
243 0 : ulong from = fd_ulong_max( ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot+1UL,
244 0 : fd_ulong_if( !highwater, 0UL, highwater+1UL ) );
245 0 : int next_leader_known = !!fd_multi_epoch_leaders_get_leader_for_slot( ctx->mleaders, from );
246 0 : ulong nl = query_next_leader_slot( ctx, from );
247 0 : ulong next_leader = nl==ULONG_MAX ? 0UL : nl;
248 0 : int healthy = wait_info_healthy( ctx->caught_up, ctx->catch_up_max_fec_slot, ctx->reset_slot );
249 :
250 0 : fd_wait_info_write_begin( ctx->wait_info );
251 0 : ctx->wait_info->info.caught_up = healthy;
252 0 : ctx->wait_info->info.reset_slot = ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot;
253 0 : ctx->wait_info->info.tip_slot = ctx->catch_up_max_fec_slot==ULONG_MAX ? 0UL : ctx->catch_up_max_fec_slot;
254 0 : ctx->wait_info->info.leader_highwater_slot = highwater;
255 0 : ctx->wait_info->info.next_leader_known = next_leader_known;
256 0 : ctx->wait_info->info.next_leader_slot = ctx->leader_bank ? ctx->leader_bank->f.slot : next_leader;
257 0 : ctx->wait_info->info.leader_slot = ctx->leader_bank ? ctx->leader_bank->f.slot : 0UL;
258 0 : ctx->wait_info->info.epoch_end_slot = ctx->epoch_end_slot;
259 0 : ctx->wait_info->info.slots_per_epoch = ctx->slots_per_epoch;
260 0 : ctx->wait_info->info.ns_per_slot = ctx->ns_per_slot;
261 0 : ctx->wait_info->info.delinquent_known = ctx->delinquent_known;
262 0 : ctx->wait_info->info.delinquent_stake_lamports = ctx->delinquent_stake_lamports;
263 0 : ctx->wait_info->info.cluster_active_stake_lamports = ctx->cluster_active_stake_lamports;
264 0 : ctx->wait_info->info.snap_active = ctx->snapmk.active;
265 0 : ctx->wait_info->info.snap_finished_full = ctx->snapmk.snap_finished_full;
266 0 : ctx->wait_info->info.snap_finished_incr = ctx->snapmk.snap_finished_incr;
267 0 : ctx->wait_info->info.snap_full_interval_blocks = ctx->snapmk.full_interval_blocks;
268 0 : ctx->wait_info->info.snap_incr_interval_blocks = ctx->snapmk.incremental_interval_blocks;
269 0 : ctx->wait_info->info.snap_produced_incr_cnt = ctx->snapmk.snap_produced_incr_cnt;
270 0 : fd_wait_info_write_end( ctx->wait_info );
271 0 : }
272 :
273 : static inline void
274 0 : metrics_write( fd_replay_tile_t * ctx ) {
275 0 : fd_accdb_flush_metrics( ctx->accdb );
276 :
277 0 : ulong store_fec_spilled = 0UL;
278 0 : ulong store_fec_spill_bytes= 0UL;
279 0 : ulong store_fec_spill_reads= 0UL;
280 0 : ulong store_fec_spill_read_bytes = 0UL;
281 0 : ulong store_fec_cache_used = 0UL;
282 0 : ulong store_fec_cache_max = 0UL;
283 0 : ulong store_fec_cache_pinned = 0UL;
284 0 : if( FD_LIKELY( ctx->store ) ) {
285 0 : store_fec_spilled = FD_VOLATILE_CONST( ctx->store->fec_spill_cnt );
286 0 : store_fec_spill_bytes = FD_VOLATILE_CONST( ctx->store->fec_spill_bytes );
287 0 : store_fec_spill_reads = FD_VOLATILE_CONST( ctx->store->fec_spill_read_cnt );
288 0 : store_fec_spill_read_bytes = FD_VOLATILE_CONST( ctx->store->fec_spill_read_bytes );
289 0 : store_fec_cache_pinned = FD_VOLATILE_CONST( ctx->store->cache_pinned_cnt );
290 :
291 0 : store_fec_cache_max = FD_VOLATILE_CONST( ctx->store->cache_slot_cnt );
292 0 : ulong store_fec_cache_free = fd_ulong_min( FD_VOLATILE_CONST( ctx->store->cache_free_cnt ), store_fec_cache_max );
293 0 : store_fec_cache_used = store_fec_cache_max - store_fec_cache_free;
294 0 : }
295 :
296 0 : FD_MHIST_COPY( REPLAY, STORE_QUERY_DURATION_NANOS, ctx->metrics.store_query_work );
297 0 : FD_MCNT_SET ( REPLAY, STORE_FEC_QUERIED, ctx->metrics.store_query_cnt );
298 0 : FD_MCNT_SET ( REPLAY, STORE_FEC_QUERY_MISSED, ctx->metrics.store_query_missing_cnt );
299 0 : FD_MCNT_SET ( REPLAY, STORE_FEC_SPILLED, store_fec_spilled );
300 0 : FD_MCNT_SET ( REPLAY, STORE_FEC_SPILLED_BYTES, store_fec_spill_bytes );
301 0 : FD_MCNT_SET ( REPLAY, STORE_FEC_SPILL_READ, store_fec_spill_reads );
302 0 : FD_MCNT_SET ( REPLAY, STORE_FEC_SPILL_READ_BYTES, store_fec_spill_read_bytes );
303 0 : FD_MGAUGE_SET( REPLAY, STORE_FEC_CACHE_USED, store_fec_cache_used );
304 0 : FD_MGAUGE_SET( REPLAY, STORE_FEC_CACHE_CAPACITY, store_fec_cache_max );
305 0 : FD_MGAUGE_SET( REPLAY, STORE_FEC_CACHE_PINNED, store_fec_cache_pinned );
306 :
307 0 : FD_MGAUGE_SET( REPLAY, ROOT_SLOT, ctx->consensus_root_slot==ULONG_MAX ? 0UL : ctx->consensus_root_slot );
308 0 : ulong leader_slot = ctx->leader_bank ? ctx->leader_bank->f.slot : 0UL;
309 :
310 0 : if( FD_LIKELY( ctx->leader_bank ) ) {
311 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, leader_slot );
312 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
313 0 : } else {
314 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, ctx->next_leader_slot==ULONG_MAX ? 0UL : ctx->next_leader_slot );
315 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, 0UL );
316 0 : }
317 0 : FD_MGAUGE_SET( REPLAY, RESET_SLOT, ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot );
318 0 : if( FD_UNLIKELY( ctx->reset_slot!=ULONG_MAX && ctx->consensus_root_slot!=ULONG_MAX ) ) {
319 0 : FD_MGAUGE_SET( REPLAY, ROOT_SLOT_DISTANCE, fd_ulong_sat_sub( ctx->reset_slot, ctx->consensus_root_slot ) );
320 0 : }
321 0 : FD_MGAUGE_SET( REPLAY, LAST_REWARDED_VOTE_SLOT, fd_ulong_if( ctx->metrics.voted_slot==ULONG_MAX, 0UL, ctx->metrics.voted_slot ) );
322 :
323 0 : FD_MGAUGE_SET( REPLAY, BANK_USED, fd_banks_pool_used_cnt( ctx->banks ) );
324 :
325 0 : FD_MCNT_SET( REPLAY, BANK_FROZEN, ctx->metrics.slots_total );
326 0 : FD_MGAUGE_SET( REPLAY, CHAIN_TXN_HEIGHT, ctx->metrics.transactions_total );
327 :
328 0 : FD_MGAUGE_SET( REPLAY, LAST_REASSEMBLY_DEQUEUED_SLOT, ctx->metrics.reasm_latest_slot );
329 0 : FD_MGAUGE_SET( REPLAY, LAST_REASSEMBLY_DEQUEUED_FEC_INDEX, ctx->metrics.reasm_latest_fec_idx );
330 :
331 0 : fd_sched_metrics_write( ctx->sched, ctx->clock );
332 :
333 0 : FD_MCNT_SET( REPLAY, FEC_POLL_SKIPPED_SCHED_FULL, ctx->metrics.sched_full );
334 0 : FD_MCNT_SET( REPLAY, FEC_POLL_SKIPPED_LEADER_FREEZE_WAIT, ctx->metrics.leader_bid_wait );
335 0 : FD_MCNT_SET( REPLAY, FEC_POLL_SKIPPED_BANK_FULL, ctx->metrics.banks_full );
336 0 : FD_MCNT_SET( REPLAY, FEC_PARENT_MISSING_DROPPED, ctx->metrics.parent_unavailable );
337 0 : FD_MCNT_SET( REPLAY, LEADER_WINDOW_ALREADY_PRODUCED, ctx->metrics.leader_slot_already_produced );
338 0 : FD_MCNT_SET( REPLAY, LEADER_WINDOW_SUPERSEDED, ctx->metrics.leader_window_superseded );
339 0 : FD_MCNT_SET( REPLAY, FEC_ROOTED_PAST_SKIPPED, ctx->metrics.fec_rooted_past );
340 0 : FD_MCNT_SET( REPLAY, ROOT_ADVANCE_DEFERRED, ctx->metrics.storage_root_behind );
341 :
342 0 : fd_progcache_admin_metrics_t const * pcm = &fd_progcache_admin_metrics_g;
343 0 : FD_MCNT_SET( REPLAY, PROGCACHE_ROOTED, pcm->root_cnt );
344 :
345 : /* Shared metrics for the program cache */
346 0 : ulong pc_class_used[ FD_PROGCACHE_CACHE_CLASS_CNT ];
347 0 : ulong pc_class_max [ FD_PROGCACHE_CACHE_CLASS_CNT ];
348 0 : fd_progcache_cache_class_occupancy( ctx->progcache, pc_class_used, pc_class_max );
349 0 : FD_MGAUGE_ENUM_COPY( REPLAY, PROGCACHE_CLASS_USED, pc_class_used );
350 0 : FD_MGAUGE_ENUM_COPY( REPLAY, PROGCACHE_CLASS_CAPACITY, pc_class_max );
351 :
352 0 : FD_ACCDB_METRICS_WRITE( REPLAY, fd_accdb_metrics( ctx->accdb ) );
353 0 : }
354 :
355 : static void
356 : metrics_trigger( fd_replay_tile_t * ctx,
357 : uint reason,
358 0 : ulong slot ) {
359 0 : FD_COMPILER_MFENCE();
360 0 : metrics_write( ctx );
361 0 : FD_COMPILER_MFENCE();
362 0 : fd_event_metrics_trigger( reason, slot, ULONG_MAX );
363 0 : }
364 :
365 : /* Agave: delinquent once the last vote is this far behind.
366 : https://github.com/anza-xyz/agave/blob/v4.2.1/rpc-client-types/src/request.rs#L166 */
367 :
368 0 : #define DELINQUENT_VALIDATOR_SLOT_DISTANCE 128UL
369 :
370 : /* Sampling interval for the O(vote accounts) stake walk, a quarter of
371 : DELINQUENT_VALIDATOR_SLOT_DISTANCE so the figure cannot lag far
372 : behind the threshold it feeds. Stamping is not gated by it: a cert
373 : not stamped when it arrives is lost. */
374 :
375 0 : #define DELINQUENT_SAMPLE_SLOTS 32UL
376 :
377 : static ushort
378 : replay_voter_rank( fd_replay_tile_t * ctx,
379 : fd_bank_t * bank,
380 0 : ulong epoch ) {
381 0 : ulong fork_id = bank->vote_stakes_fork_id;
382 0 : ulong fork_epoch = fd_vote_stakes_fork_epoch( fork_id );
383 0 : int iter_kind = FD_VOTE_STAKES_ITER_T_2;
384 0 : if( FD_UNLIKELY( epoch!=fork_epoch ) ) {
385 0 : if( FD_UNLIKELY( !fork_epoch || epoch!=fork_epoch-1UL ) ) return USHORT_MAX;
386 0 : iter_kind = FD_VOTE_STAKES_ITER_T_3;
387 0 : }
388 :
389 0 : fd_vote_stakes_t * vote_stakes = fd_bank_vote_stakes( bank );
390 0 : uchar __attribute__((aligned(FD_VOTE_STAKES_ITER_ALIGN))) iter_mem[ FD_VOTE_STAKES_ITER_FOOTPRINT ];
391 0 : for( fd_vote_stakes_iter_t * iter = fd_vote_stakes_iter_init( vote_stakes, fork_id, iter_kind, iter_mem );
392 0 : !fd_vote_stakes_iter_done( vote_stakes, fork_id, iter_kind, iter );
393 0 : fd_vote_stakes_iter_next( vote_stakes, fork_id, iter_kind, iter ) ) {
394 0 : fd_vote_stakes_ele_t ele[1];
395 0 : fd_vote_stakes_iter_ele( vote_stakes, fork_id, iter_kind, iter, ele );
396 0 : if( FD_UNLIKELY( fd_pubkey_eq( &ele->node_account, ctx->identity_pubkey ) ) ) return ele->rank;
397 0 : }
398 0 : return USHORT_MAX;
399 0 : }
400 :
401 : /* ag_rank_is_current returns 1 if the rank's last observed vote is
402 : recent enough. */
403 :
404 : static inline int
405 : ag_rank_is_current( ulong last_settled,
406 0 : ulong rank_last_voted ) {
407 0 : return fd_ulong_sat_sub( last_settled, rank_last_voted )<DELINQUENT_VALIDATOR_SLOT_DISTANCE;
408 0 : }
409 :
410 : /* ag_delinquent_epoch_roll clears the rank table when the epoch
411 : advances and raises the settled watermark. Ranks are remapped across
412 : the rollover, so the previous epoch's figure no longer describes the
413 : current stake set and reads unknown until a sample repopulates the
414 : table. The watermark cannot regress. */
415 :
416 : static void
417 : ag_delinquent_epoch_roll( fd_replay_tile_t * ctx,
418 : ulong reward_slot,
419 0 : ulong reward_epoch ) {
420 0 : if( FD_UNLIKELY( ctx->ag_last_voted_epoch==ULONG_MAX ||
421 0 : reward_epoch>ctx->ag_last_voted_epoch ) ) {
422 0 : fd_memset( ctx->ag_last_voted, 0, sizeof(ctx->ag_last_voted) );
423 0 : ctx->ag_last_voted_epoch = reward_epoch;
424 0 : ctx->delinquent_known = 0;
425 0 : ctx->ag_stamp_since = ULONG_MAX;
426 0 : }
427 0 : ctx->ag_last_settled = fd_ulong_max( ctx->ag_last_settled, reward_slot );
428 0 : }
429 :
430 : /* ag_update_delinquent stamps the reward cert's signers into
431 : ag_last_voted and recomputes delinquent stake. */
432 :
433 : static void
434 : ag_update_delinquent( fd_replay_tile_t * ctx,
435 : fd_bank_t const * bank,
436 : ulong reward_slot,
437 : ulong reward_epoch,
438 0 : fd_bls_set_t const * reward_set ) {
439 0 : if( FD_UNLIKELY( !ctx->wait_info ) ) return; /* not in this topology */
440 :
441 0 : ulong fork_id = bank->vote_stakes_fork_id;
442 0 : ulong fork_epoch = fd_vote_stakes_fork_epoch( fork_id );
443 :
444 : /* Roll on the fork's epoch, not the cert's: the cert trails the bank
445 : by FD_NUM_SLOTS_FOR_REWARD slots, so it still names the old epoch
446 : over the first slots of a new one. */
447 0 : ag_delinquent_epoch_roll( ctx, ctx->ag_last_settled, fork_epoch );
448 :
449 : /* Only T-2 exposes stake and is_valid. */
450 0 : if( FD_UNLIKELY( reward_epoch!=fork_epoch ) ) return;
451 :
452 : /* Ranks are epoch-scoped, and forks can complete an epoch-E block
453 : after an epoch-E+1 one. */
454 0 : if( FD_UNLIKELY( ctx->ag_last_voted_epoch!=ULONG_MAX &&
455 0 : reward_epoch<ctx->ag_last_voted_epoch ) ) return;
456 :
457 : /* Raise the watermark, so stamps age out as certs settle. */
458 0 : ag_delinquent_epoch_roll( ctx, reward_slot, reward_epoch );
459 :
460 : /* An empty cert is absence of evidence, so it stamps nothing. With no
461 : figure yet there is nothing to age either, but once one is known the
462 : walk below must still run: the roll above advanced ag_last_settled,
463 : so existing stamps age out even while certs are absent. */
464 0 : if( FD_UNLIKELY( fd_bls_set_is_null( reward_set ) && !ctx->delinquent_known ) ) return;
465 :
466 0 : if( FD_UNLIKELY( ctx->ag_stamp_since==ULONG_MAX ) ) ctx->ag_stamp_since = reward_slot;
467 :
468 : /* Stamp straight off the cert: O(signers), no vote stakes walk. No
469 : need to filter: entries the walk skips are never read. */
470 0 : for( ulong rank=fd_bls_set_const_iter_init( reward_set );
471 0 : !fd_bls_set_const_iter_done( rank );
472 0 : rank=fd_bls_set_const_iter_next( reward_set, rank ) ) {
473 0 : if( FD_UNLIKELY( rank>=AG_VAT_MAX ) ) break; /* FD_BLS_SET_MAX is sized separately */
474 0 : ctx->ag_last_voted[ rank ] = fd_ulong_max( ctx->ag_last_voted[ rank ], reward_slot );
475 0 : }
476 :
477 0 : if( FD_LIKELY( ctx->delinquent_known &&
478 0 : reward_slot<ctx->delinquent_sample_slot+DELINQUENT_SAMPLE_SLOTS ) ) return;
479 0 : if( FD_UNLIKELY( reward_slot<ctx->ag_stamp_since+DELINQUENT_SAMPLE_SLOTS ) ) return;
480 : /* Slots arrive out of order across forks, so hold the watermark. */
481 0 : ctx->delinquent_sample_slot = fd_ulong_max( ctx->delinquent_sample_slot, reward_slot );
482 :
483 0 : fd_vote_stakes_t * vote_stakes = fd_bank_vote_stakes( bank );
484 0 : ulong delinquent = 0UL;
485 0 : ulong total = 0UL;
486 :
487 0 : uchar __attribute__((aligned(FD_VOTE_STAKES_ITER_ALIGN))) iter_mem[ FD_VOTE_STAKES_ITER_FOOTPRINT ];
488 0 : for( fd_vote_stakes_iter_t * iter = fd_vote_stakes_iter_init( vote_stakes, fork_id, FD_VOTE_STAKES_ITER_T_2, iter_mem );
489 0 : !fd_vote_stakes_iter_done( vote_stakes, fork_id, FD_VOTE_STAKES_ITER_T_2, iter );
490 0 : fd_vote_stakes_iter_next( vote_stakes, fork_id, FD_VOTE_STAKES_ITER_T_2, iter ) ) {
491 0 : fd_vote_stakes_ele_t ele[1];
492 0 : fd_vote_stakes_iter_ele( vote_stakes, fork_id, FD_VOTE_STAKES_ITER_T_2, iter, ele );
493 0 : ulong stake = ele->stake;
494 0 : ushort rank = ele->rank;
495 0 : if( FD_UNLIKELY( !ele->is_valid || rank==FD_VOTE_STAKES_ALPENGLOW_RANK_NULL ) ) continue;
496 0 : FD_TEST( rank<AG_VAT_MAX );
497 :
498 0 : total += stake;
499 0 : if( FD_UNLIKELY( !ag_rank_is_current( ctx->ag_last_settled, ctx->ag_last_voted[ rank ] ) ) ) {
500 0 : delinquent += stake;
501 0 : }
502 0 : }
503 :
504 0 : ctx->delinquent_stake_lamports = delinquent;
505 0 : ctx->cluster_active_stake_lamports = total;
506 :
507 : /* Over-counts delinquency, so a partial union blocks rather than
508 : passes. */
509 0 : ctx->delinquent_known = 1;
510 0 : }
511 :
512 : static int
513 : replay_reward_cert_voted( fd_replay_tile_t * ctx,
514 : fd_bank_t * bank,
515 : ushort * rank_out,
516 0 : ushort * count_out ) {
517 0 : *rank_out = USHORT_MAX;
518 0 : *count_out = USHORT_MAX;
519 0 : if( FD_UNLIKELY( bank->f.slot<FD_NUM_SLOTS_FOR_REWARD ) ) return 0;
520 :
521 0 : ulong reward_slot = bank->f.slot-FD_NUM_SLOTS_FOR_REWARD;
522 0 : ulong reward_epoch = fd_slot_to_epoch( &bank->f.epoch_schedule, reward_slot, NULL );
523 0 : ushort rank = replay_voter_rank( ctx, bank, reward_epoch );
524 0 : *rank_out = rank;
525 :
526 0 : fd_block_footer_t const * footer = bank==ctx->leader_bank ? ctx->leader_footer : fd_sched_get_footer( ctx->sched, bank->idx );
527 0 : if( FD_UNLIKELY( !footer ) ) return 0;
528 :
529 : /* A validator included in both reward certificates is counted once. */
530 0 : fd_bls_set_t reward_set[ fd_bls_set_word_cnt ];
531 0 : fd_bls_set_null( reward_set );
532 0 : if( FD_UNLIKELY( footer->has_skip_reward_cert ) ) fd_bls_set_union( reward_set, reward_set, footer->skip_reward_cert.signer_set );
533 0 : if( FD_LIKELY( footer->has_notar_reward_cert ) ) fd_bls_set_union( reward_set, reward_set, footer->notar_reward_cert.signer_set );
534 0 : *count_out = (ushort)fd_bls_set_cnt( reward_set );
535 :
536 0 : ag_update_delinquent( ctx, bank, reward_slot, reward_epoch, reward_set );
537 :
538 0 : if( FD_UNLIKELY( rank==USHORT_MAX ) ) return 0;
539 :
540 : /* bits at or past the cert's nbits are left clear at decode, so the
541 : set test alone bounds the rank */
542 0 : int in_cert = fd_bls_set_test( reward_set, rank );
543 :
544 0 : if( FD_UNLIKELY( in_cert && ( ctx->metrics.voted_slot==ULONG_MAX || reward_slot>ctx->metrics.voted_slot ) ) ) ctx->metrics.voted_slot = reward_slot;
545 0 : return in_cert;
546 0 : }
547 :
548 : static void
549 : publish_epoch_info( fd_replay_tile_t * ctx,
550 : fd_stem_context_t * stem,
551 : fd_bank_t * bank,
552 0 : int next_epoch ) {
553 0 : fd_epoch_schedule_t const * schedule = &bank->f.epoch_schedule;
554 0 : ulong epoch = fd_slot_to_epoch( schedule, bank->f.slot, NULL ) + fd_ulong_if( next_epoch, 1UL, 0UL );
555 :
556 0 : fd_features_t const * features = &bank->f.features;
557 :
558 0 : fd_runtime_stack_t * runtime_stack = ctx->runtime_stack;
559 :
560 0 : fd_epoch_info_msg_t * epoch_info_msg = fd_chunk_to_laddr( ctx->epoch_out->mem, ctx->epoch_out->chunk );
561 :
562 0 : epoch_info_msg->staked_vote_cnt = next_epoch ? runtime_stack->epoch_weights.next_stake_weights_cnt : runtime_stack->epoch_weights.stake_weights_cnt;
563 0 : epoch_info_msg->staked_id_cnt = next_epoch ? runtime_stack->epoch_weights.next_id_weights_cnt : runtime_stack->epoch_weights.id_weights_cnt;
564 0 : epoch_info_msg->epoch_schedule = *schedule;
565 0 : epoch_info_msg->features = *features;
566 0 : epoch_info_msg->epoch = epoch;
567 0 : epoch_info_msg->start_slot = fd_epoch_slot0( schedule, epoch );
568 0 : epoch_info_msg->slot_cnt = fd_epoch_slot_cnt( schedule, epoch );
569 0 : epoch_info_msg->ns_per_slot = fd_slot_params_at_slot( bank, epoch_info_msg->start_slot ).ns_per_slot;
570 :
571 0 : fd_vote_stake_weight_t * stake_weights = fd_type_pun( epoch_info_msg + 1 );
572 0 : fd_vote_stake_weight_t * src_stake_weights = next_epoch ? runtime_stack->epoch_weights.next_stake_weights : runtime_stack->epoch_weights.stake_weights;
573 0 : memcpy( stake_weights, src_stake_weights, epoch_info_msg->staked_vote_cnt * sizeof(fd_vote_stake_weight_t) );
574 :
575 0 : fd_stake_weight_t * id_weights = fd_epoch_info_msg_id_weights( epoch_info_msg );
576 0 : fd_stake_weight_t * src_id_weights = next_epoch ? runtime_stack->epoch_weights.next_id_weights : runtime_stack->epoch_weights.id_weights;
577 0 : fd_memcpy( id_weights, src_id_weights, epoch_info_msg->staked_id_cnt * sizeof(fd_stake_weight_t) );
578 :
579 0 : ulong epoch_info_sz = fd_epoch_info_msg_sz( epoch_info_msg->staked_vote_cnt, epoch_info_msg->staked_id_cnt );
580 0 : ulong epoch_info_sig = 4UL;
581 0 : fd_stem_publish( stem, ctx->epoch_out->idx, epoch_info_sig, ctx->epoch_out->chunk, epoch_info_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
582 0 : ctx->epoch_out->chunk = fd_dcache_compact_next( ctx->epoch_out->chunk, epoch_info_sz, ctx->epoch_out->chunk0, ctx->epoch_out->wmark );
583 :
584 0 : fd_multi_epoch_leaders_epoch_msg_init( ctx->mleaders, epoch_info_msg );
585 0 : fd_multi_epoch_leaders_epoch_msg_fini( ctx->mleaders );
586 :
587 0 : ctx->next_leader_query_start = ULONG_MAX;
588 0 : }
589 :
590 : static void
591 : publish_replay_out( fd_replay_tile_t * ctx,
592 : fd_stem_context_t * stem,
593 : ulong sig,
594 0 : ulong sz ) {
595 0 : ulong tspub = fd_frag_meta_ts_comp( fd_tickcount() );
596 0 : if( FD_UNLIKELY( sig!=REPLAY_SIG_TXN_EXECUTED ) ) {
597 0 : fd_replay_out_link_t * out = sig==REPLAY_SIG_MISSING_FEC ? ctx->rotor_out : ctx->slot_out;
598 0 : if( FD_LIKELY( out->idx!=ULONG_MAX ) ) { /* backtest has no rotor */
599 0 : fd_memcpy( fd_chunk_to_laddr( out->mem, out->chunk ), fd_chunk_to_laddr_const( ctx->replay_out->mem, ctx->replay_out->chunk ), sz );
600 0 : fd_stem_publish( stem, out->idx, sig, out->chunk, sz, 0UL, 0UL, tspub );
601 0 : out->chunk = fd_dcache_compact_next( out->chunk, sz, out->chunk0, out->wmark );
602 0 : }
603 0 : }
604 0 : fd_stem_publish( stem, ctx->replay_out->idx, sig, ctx->replay_out->chunk, sz, 0UL, 0UL, tspub );
605 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sz, ctx->replay_out->chunk0, ctx->replay_out->wmark );
606 0 : }
607 :
608 : /**********************************************************************/
609 : /* Transaction execution state machine helpers */
610 : /**********************************************************************/
611 :
612 : static inline void
613 : timing_slot_release( fd_replay_tile_t * ctx,
614 0 : ulong bank_idx ) {
615 0 : ulong tslot = ctx->timing_slot_of_bank[ bank_idx ];
616 0 : if( FD_LIKELY( tslot!=fd_timing_slot_pool_idx_null( ctx->timing_slot_pool ) ) ) {
617 0 : fd_timing_slot_pool_idx_release( ctx->timing_slot_pool, tslot );
618 0 : ctx->timing_slot_of_bank[ bank_idx ] = fd_timing_slot_pool_idx_null( ctx->timing_slot_pool );
619 0 : }
620 0 : }
621 :
622 : static void
623 : replay_block_start( fd_replay_tile_t * ctx,
624 : ulong bank_idx,
625 : ulong parent_bank_idx,
626 0 : ulong slot ) {
627 0 : long before = fd_clock_tile_now( ctx->clock );
628 :
629 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
630 0 : FD_CHECK_CRIT( bank, "invariant violation: bank is NULL" );
631 0 : FD_CHECK_CRIT( bank->state==FD_BANK_STATE_INIT, "invariant violation: bank is not in correct state" );
632 :
633 0 : bank->preparation_begin_nanos = before;
634 :
635 0 : FD_TEST( ctx->timing_slot_of_bank[ bank_idx ]==fd_timing_slot_pool_idx_null( ctx->timing_slot_pool ) );
636 0 : if( FD_LIKELY( fd_timing_slot_pool_free( ctx->timing_slot_pool ) ) ) {
637 0 : ulong tslot = fd_timing_slot_pool_idx_acquire( ctx->timing_slot_pool );
638 0 : fd_timing_slot_pool_ele( ctx->timing_slot_pool, tslot )->cnt = 0UL;
639 0 : ctx->timing_slot_of_bank[ bank_idx ] = tslot;
640 0 : }
641 :
642 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
643 0 : FD_CHECK_CRIT( parent_bank, "invariant violation: parent bank is NULL" );
644 0 : FD_CHECK_CRIT( parent_bank->state==FD_BANK_STATE_FROZEN || parent_bank->state==FD_BANK_STATE_PRUNABLE, "invariant violation: parent bank is not in correct state" );
645 :
646 : /* Clone the bank from the parent. We must special case the first
647 : slot that is executed as the snapshot does not provide a parent
648 : block id. */
649 :
650 0 : bank = fd_banks_clone_from_parent( ctx->banks, bank_idx );
651 0 : if( FD_UNLIKELY( !bank ) ) {
652 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
653 0 : }
654 0 : fd_pubkey_t const * leader = fd_multi_epoch_leaders_get_leader_for_slot( ctx->mleaders, slot );
655 0 : FD_TEST( leader );
656 0 : bank->f.slot = slot;
657 0 : bank->leader = *leader;
658 0 : bank->txncache_fork_id = fd_txncache_attach_child ( ctx->txncache, parent_bank->txncache_fork_id );
659 0 : bank->progcache_fork_id = fd_progcache_attach_child( ctx->progcache, parent_bank->progcache_fork_id );
660 0 : bank->accdb_fork_id = fd_accdb_attach_child ( ctx->accdb, parent_bank->accdb_fork_id );
661 0 : bank->parent_accdb_fork_id = parent_bank->accdb_fork_id;
662 :
663 0 : ulong new_epoch = fd_slot_to_epoch( &parent_bank->f.epoch_schedule, slot, NULL );
664 0 : ulong root_epoch = fd_slot_to_epoch( &parent_bank->f.epoch_schedule, ctx->published_root_slot, NULL );
665 0 : if( FD_UNLIKELY( new_epoch>root_epoch+1UL ) ) {
666 0 : FD_LOG_CRIT(( "firedancer replay does not support replaying more than one epoch ahead of the current root" ));
667 0 : }
668 :
669 : /* Update required runtime state and handle potential boundary. */
670 :
671 0 : int is_epoch_boundary = 0;
672 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
673 0 : FD_CHECK_CRIT( !is_epoch_boundary || FD_FEATURE_ACTIVE_BANK( bank, alpenglow ), "alpenglow feature is not active" );
674 :
675 0 : ulong max_tick_height;
676 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( parent_bank->f.ticks_per_slot, slot, &max_tick_height ) ) ) {
677 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, parent_bank->f.ticks_per_slot ));
678 0 : }
679 0 : bank->f.max_tick_height = max_tick_height;
680 : /* in alpenglow, we expect only one tick per block. Instead of
681 : adjusting max tick height, we match agave behavior by setting
682 : tick height to max tick height - 1. These fields must stay in
683 : line with agave behavior. */
684 0 : bank->f.tick_height = bank->f.max_tick_height - 1UL;
685 0 : bank->f.slot_params.hashes_per_tick = 1UL;
686 0 : fd_sched_set_poh_params( ctx->sched, bank->idx, bank->f.tick_height, bank->f.max_tick_height, bank->f.slot_params.hashes_per_tick, &parent_bank->f.poh );
687 :
688 0 : FD_LOG_DEBUG(( "replay_block_start: bank_idx=%lu slot=%lu parent_bank_idx=%lu", bank_idx, slot, parent_bank_idx ));
689 0 : }
690 :
691 : static void
692 0 : cost_tracker_snap( fd_bank_t * bank, fd_replay_slot_completed_t * slot_info ) {
693 0 : if( FD_LIKELY( bank->cost_tracker_fork_id!=USHORT_MAX ) ) {
694 0 : fd_cost_tracker_t const * cost_tracker = fd_cost_tracker_store_pin( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id );
695 0 : if( FD_UNLIKELY( cost_tracker->block_cost_limit==0UL ) ) {
696 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
697 0 : } else {
698 0 : slot_info->cost_tracker.block_cost = cost_tracker->block_cost;
699 0 : slot_info->cost_tracker.allocated_accounts_data_size = cost_tracker->allocated_accounts_data_size;
700 0 : slot_info->cost_tracker.block_cost_limit = cost_tracker->block_cost_limit;
701 0 : slot_info->cost_tracker.account_cost_limit = cost_tracker->account_cost_limit;
702 0 : }
703 0 : fd_cost_tracker_store_unpin( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id );
704 0 : } else {
705 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
706 0 : }
707 0 : slot_info->cost_tracker.pool_idx = fd_ulong_if( bank->cost_tracker_fork_id==USHORT_MAX, ULONG_MAX, bank->cost_tracker_fork_id );
708 0 : }
709 :
710 : static int
711 0 : sched_dead_reason_to_event( int sched_reason ) {
712 0 : switch( sched_reason ) {
713 0 : case FD_SCHED_DEAD_REASON_UNPARSEABLE_CONTENT: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_UNPARSEABLE_CONTENT;
714 0 : case FD_SCHED_DEAD_REASON_SHORT_BLOCK: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_SHORT_BLOCK;
715 0 : case FD_SCHED_DEAD_REASON_TOO_MANY_TXNS: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TOO_MANY_TXNS;
716 0 : case FD_SCHED_DEAD_REASON_TOO_MANY_MICROBLOCKS: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TOO_MANY_MICROBLOCKS;
717 0 : case FD_SCHED_DEAD_REASON_DUPLICATE_ACCOUNT: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_DUPLICATE_ACCOUNT;
718 0 : case FD_SCHED_DEAD_REASON_TRAILING_ENTRY: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TRAILING_ENTRY;
719 0 : case FD_SCHED_DEAD_REASON_TOO_MANY_TICKS: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TOO_MANY_TICKS;
720 0 : case FD_SCHED_DEAD_REASON_TOO_FEW_TICKS: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TOO_FEW_TICKS;
721 0 : case FD_SCHED_DEAD_REASON_ZERO_MICROBLOCKS: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ZERO_MICROBLOCKS;
722 0 : case FD_SCHED_DEAD_REASON_WRONG_HASHES_PER_TICK: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_WRONG_HASHES_PER_TICK;
723 0 : case FD_SCHED_DEAD_REASON_INCONSISTENT_TICK_HASHES: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_INCONSISTENT_TICK_HASHES;
724 0 : case FD_SCHED_DEAD_REASON_TICK_HASHES_OVERFLOW: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TICK_HASHES_OVERFLOW;
725 0 : case FD_SCHED_DEAD_REASON_TICK_HASHES_OVERFLOW_INGEST: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TICK_HASHES_OVERFLOW_INGEST;
726 0 : case FD_SCHED_DEAD_REASON_ZERO_HASH_TICK: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ZERO_HASH_TICK;
727 0 : case FD_SCHED_DEAD_REASON_ZERO_HASH_TICK_INGEST: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ZERO_HASH_TICK_INGEST;
728 0 : case FD_SCHED_DEAD_REASON_TICK_HASH_MISMATCH: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TICK_HASH_MISMATCH;
729 0 : case FD_SCHED_DEAD_REASON_ENTRY_HASH_MISMATCH: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ENTRY_HASH_MISMATCH;
730 0 : case FD_SCHED_DEAD_REASON_ENTRY_HASH_MISMATCH_INGEST: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ENTRY_HASH_MISMATCH_INGEST;
731 0 : case FD_SCHED_DEAD_REASON_DEAD_ANCESTOR: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_PARENT_DEAD;
732 0 : case FD_SCHED_DEAD_REASON_BAD_BLOCK_MARKER: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_BAD_BLOCK_MARKER;
733 0 : case FD_SCHED_DEAD_REASON_ALPENGLOW_HASH_CNT: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ALPENGLOW_HASH_CNT;
734 0 : case FD_SCHED_DEAD_REASON_MISSING_PARENT_MARKER: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_MISSING_PARENT_MARKER;
735 0 : case FD_SCHED_DEAD_REASON_MULTIPLE_BLOCK_HEADERS: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_MULTIPLE_BLOCK_HEADERS;
736 0 : case FD_SCHED_DEAD_REASON_GENESIS_CERT_OUT_OF_ORDER: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_GENESIS_CERT_OUT_OF_ORDER;
737 0 : case FD_SCHED_DEAD_REASON_MULTIPLE_BLOCK_FOOTERS: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_MULTIPLE_BLOCK_FOOTERS;
738 0 : case FD_SCHED_DEAD_REASON_ENTRY_AFTER_BLOCK_FOOTER: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ENTRY_AFTER_BLOCK_FOOTER;
739 0 : case FD_SCHED_DEAD_REASON_INVALID_ALPENTICK_POSITION: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_INVALID_ALPENTICK_POSITION;
740 0 : case FD_SCHED_DEAD_REASON_MISSING_BLOCK_FOOTER: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_MISSING_BLOCK_FOOTER;
741 0 : case FD_SCHED_DEAD_REASON_SPURIOUS_UPDATE_PARENT: return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_SPURIOUS_UPDATE_PARENT;
742 0 : default: FD_LOG_CRIT(( "unmapped scheduler dead reason %d", sched_reason ));
743 0 : }
744 0 : }
745 :
746 : static int
747 0 : sched_block_dead_reason_to_event( fd_replay_tile_t * ctx, ulong bank_idx ) {
748 0 : if( FD_UNLIKELY( fd_sched_block_is_discarded( ctx->sched, bank_idx ) ) ) return FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD;
749 0 : return sched_dead_reason_to_event( fd_sched_get_dead_reason( ctx->sched, bank_idx ) );
750 0 : }
751 :
752 : /* The stats array is keyed by bank idx. */
753 :
754 : static void
755 : block_completed_event_fill_reception( fd_replay_tile_t * ctx,
756 : fd_event_block_completed_t * ev,
757 0 : ulong bank_idx ) {
758 0 : ev->lowest_verified_fec_index = 0U;
759 0 : ev->last_completed_fec_set_index = UINT_MAX;
760 0 : fd_rotor_fec_metrics_t const * rm = NULL;
761 :
762 0 : fd_bank_t const * bank = bank_idx==ULONG_MAX ? NULL : fd_banks_bank_query( ctx->banks, bank_idx );
763 0 : if( FD_UNLIKELY( !bank ) ) return;
764 :
765 0 : fd_block_id_ele_t const * ele = &ctx->block_id_arr[ bank_idx ];
766 0 : ev->fec_set_count = ele->fec_cnt;
767 :
768 0 : if( FD_LIKELY( bank->parent_idx!=ULONG_MAX ) ) {
769 0 : fd_block_id_ele_t const * parent = &ctx->block_id_arr[ bank->parent_idx ];
770 0 : fd_memcpy( ev->parent_block_id, parent->dmr.uc, sizeof(ev->parent_block_id) );
771 0 : if( !ev->parent_slot ) ev->parent_slot = parent->slot;
772 0 : }
773 :
774 : /* A mismatched bank_seq means the entry belongs to a previous
775 : occupant of this bank idx, i.e. we have no stats for this block. */
776 0 : fd_reception_stats_t const * s = &ctx->reception_stats[ bank_idx ];
777 0 : if( FD_UNLIKELY( s->bank_seq!=bank->bank_seq ) ) return;
778 0 : rm = &s->metrics.rotor;
779 0 : ev->votor_repaired = !!rm->votor_repaired;
780 :
781 0 : if( FD_UNLIKELY( !rm->stats_valid ) ) return;
782 :
783 0 : ev->turbine_shred_count = rm->blk_turbine_cnt;
784 0 : ev->repair_shred_count = rm->blk_repair_cnt;
785 0 : ev->recovered_shred_count = rm->blk_recovered_cnt;
786 0 : ev->data_shred_count = rm->blk_data_cnt;
787 0 : ev->parity_shred_count = rm->blk_parity_cnt;
788 0 : ev->slot_complete_flag = !!rm->blk_slot_complete;
789 :
790 0 : ev->last_completed_fec_set_index = rm->blk_last_completed_fec_idx;
791 0 : ev->repair_request_window_count = rm->blk_req_window_cnt;
792 0 : ev->repair_request_highest_window_count = rm->blk_req_highest_cnt;
793 0 : ev->repair_request_orphan_count = rm->blk_req_orphan_cnt;
794 0 : ev->repair_request_shred_block_id_count = rm->blk_req_shred_bid_cnt;
795 0 : ev->repair_request_parent_count = rm->blk_req_parent_cnt;
796 0 : ev->repair_request_fec_root_count = rm->blk_req_fec_root_cnt;
797 0 : ev->repair_requests_retransmitted = rm->blk_req_retransmit_cnt;
798 0 : ev->repair_responses_received = rm->blk_repair_responses;
799 0 : ev->first_shred_received_time = rm->blk_first_shred_ts_nanos;
800 0 : ev->last_shred_received_time = rm->blk_last_shred_ts_nanos;
801 0 : ev->first_repair_request_time = rm->blk_first_req_ts_nanos;
802 0 : ev->last_repair_received_time = rm->blk_last_repair_resp_ts_nanos;
803 :
804 : /* always zero for alpenglow*/
805 0 : ev->lowest_verified_fec_index = 0;
806 0 : ev->repair_failed_chain_verify = 0;
807 0 : ev->chain_confirmed = 0;
808 0 : }
809 :
810 : static void
811 : block_completed_event_fill_bank( fd_replay_tile_t * ctx,
812 : fd_event_block_completed_t * ev,
813 0 : fd_bank_t * bank ) {
814 0 : fd_bank_t * parent_bank = fd_banks_get_parent( ctx->banks, bank );
815 :
816 0 : int prepared = bank->preparation_begin_nanos!=0L;
817 0 : ulong slot = prepared ? bank->f.slot : ctx->block_id_arr[ bank->idx ].slot;
818 :
819 0 : ev->bank_seq = bank->bank_seq;
820 0 : ev->slot = slot;
821 0 : ev->parent_slot = prepared ? bank->f.parent_slot
822 0 : : bank->parent_idx!=ULONG_MAX ? ctx->block_id_arr[ bank->parent_idx ].slot
823 0 : : 0UL;
824 0 : if( FD_LIKELY( prepared ) ) ev->epoch = bank->f.epoch;
825 0 : else if( FD_LIKELY( ctx->notified_root_bank ) ) ev->epoch = fd_slot_to_epoch( &ctx->notified_root_bank->f.epoch_schedule, slot, NULL );
826 0 : fd_block_id_ele_t const * ele = &ctx->block_id_arr[ bank->idx ];
827 0 : fd_memcpy( ev->block_id, ele->block_id_seen ? ele->dmr.uc : ele->latest_mr.uc, sizeof(ev->block_id) );
828 0 : if( FD_LIKELY( bank->block_completed_nanos ) ) fd_memcpy( ev->bank_hash, bank->f.bank_hash.uc, sizeof(ev->bank_hash) );
829 :
830 0 : ev->first_fec_set_received_time = bank->is_leader ? 0UL : (ulong)bank->first_fec_set_received_nanos;
831 0 : ev->preparation_begin_time = (ulong)bank->preparation_begin_nanos;
832 0 : ev->first_transaction_scheduled_time = (ulong)bank->first_transaction_scheduled_nanos;
833 0 : ev->last_transaction_finished_time = (ulong)bank->last_transaction_finished_nanos;
834 0 : ev->block_completed_time = (ulong)bank->block_completed_nanos;
835 0 : ev->parent_block_completed_time = parent_bank ? (ulong)parent_bank->block_completed_nanos : 0UL;
836 :
837 0 : if( FD_UNLIKELY( bank->cost_tracker_fork_id!=USHORT_MAX ) ) {
838 0 : fd_cost_tracker_t const * ct = fd_cost_tracker_store_pin( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id );
839 0 : ev->cost_tracker_block_cost = ct->block_cost;
840 0 : ev->cost_tracker_allocated_accounts_data_size = ct->allocated_accounts_data_size;
841 0 : ev->cost_tracker_block_cost_limit = ct->block_cost_limit;
842 0 : ev->cost_tracker_account_cost_limit = ct->account_cost_limit;
843 0 : fd_cost_tracker_store_unpin( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id );
844 0 : }
845 :
846 0 : ev->bank_idx = bank->idx;
847 0 : if( FD_LIKELY( prepared ) ) {
848 0 : ev->txncache_fork_id = bank->txncache_fork_id.val;
849 0 : ev->progcache_fork_id = bank->progcache_fork_id;
850 0 : ev->accdb_fork_id = bank->accdb_fork_id.val;
851 0 : ev->vote_stakes_fork_id = bank->vote_stakes_fork_id;
852 0 : ev->stake_rewards_fork_id = bank->stake_rewards_fork_id;
853 0 : ev->epoch_credits_fork_id = bank->epoch_credits_fork_id;
854 0 : ev->stake_delegations_fork_id = bank->stake_delegations_fork_id;
855 0 : ev->cost_tracker_pool_idx = fd_ulong_if( bank->cost_tracker_fork_id==USHORT_MAX, ULONG_MAX, bank->cost_tracker_fork_id );
856 0 : }
857 :
858 0 : block_completed_event_fill_reception( ctx, ev, bank->idx );
859 0 : }
860 :
861 : static int
862 0 : pack_end_reason_to_event( int reason ) {
863 0 : switch( reason ) {
864 0 : case FD_PACK_END_SLOT_REASON_TIME: return FD_EVENT_BLOCK_COMPLETED_PACK_END_REASON_TIME;
865 0 : case FD_PACK_END_SLOT_REASON_MICROBLOCK: return FD_EVENT_BLOCK_COMPLETED_PACK_END_REASON_MICROBLOCK_LIMIT;
866 0 : case FD_PACK_END_SLOT_REASON_ABANDONED: return FD_EVENT_BLOCK_COMPLETED_PACK_END_REASON_ABANDONED;
867 0 : default: FD_LOG_CRIT(( "unmapped pack end reason %d", reason ));
868 0 : }
869 0 : }
870 :
871 : static void
872 : block_completed_event_fill_leader( fd_replay_tile_t * ctx,
873 : fd_event_block_completed_t * ev,
874 0 : fd_bank_t * bank ) {
875 0 : FD_TEST( ctx->leader_stats.slot==bank->f.slot );
876 :
877 0 : ev->became_leader_time = (ulong)ctx->leader_stats.became_leader_nanos;
878 0 : ev->leader_slot_start_time = (ulong)ctx->leader_stats.leader_slot_start_nanos;
879 :
880 0 : ev->first_fec_set_received_time = (ulong)ctx->leader_stats.first_fec_returned_nanos;
881 0 : ev->recovered_shred_count = 0UL;
882 :
883 0 : ev->microblock_count = ctx->leader_stats.microblock_count;
884 0 : ev->pack_block_cost = ctx->leader_stats.pack_block_cost;
885 0 : ev->pack_vote_cost = ctx->leader_stats.pack_vote_cost;
886 0 : ev->pack_data_bytes = ctx->leader_stats.pack_data_bytes;
887 0 : ev->bundle_txn_count = ctx->leader_stats.bundle_txn_count;
888 0 : ev->pack_start_time = (ulong)ctx->leader_stats.pack_start_nanos;
889 0 : ev->pack_end_time = (ulong)ctx->leader_stats.pack_end_nanos;
890 0 : ev->pack_end_reason = pack_end_reason_to_event( ctx->leader_stats.pack_end_reason );
891 0 : }
892 :
893 : static void
894 : block_completed_event_fill_leader_txn_timing( fd_replay_tile_t * ctx,
895 : fd_event_block_completed_t * ev,
896 0 : fd_bank_t * bank ) {
897 0 : ulong tidx = ctx->leader_stats.timing_table_idx;
898 0 : if( FD_UNLIKELY( !ctx->leader_txn_timing || tidx>=FD_LEADER_TXN_TIMING_TABLE_CNT ) ) return;
899 0 : fd_leader_txn_timing_table_t const * table = fd_leader_txn_timing_table_const( ctx->leader_txn_timing, tidx, ctx->max_txn_per_slot );
900 0 : if( FD_UNLIKELY( table->slot!=bank->f.slot ) ) return;
901 :
902 0 : ulong cnt = fd_ulong_min( table->cnt, FD_EVENT_BLOCK_COMPLETED_TXN_TIMING_MAX );
903 0 : long first_dispatch = LONG_MAX;
904 0 : for( ulong i=0UL; i<cnt; i++ ) {
905 0 : fd_leader_txn_timing_rec_t const * rec = &table->rec[ i ];
906 0 : first_dispatch = fd_long_min( first_dispatch, rec->dispatched_ticks );
907 0 : ev->txn_timing[ i ] = (fd_event_block_completed_txn_timing_t){
908 0 : .received_time = (ulong)rec->received_ns,
909 0 : .dispatched_time = rec->dispatched_ticks==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, rec->dispatched_ticks ),
910 0 : .replayed_time = rec->replayed_ticks ==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, rec->replayed_ticks ),
911 0 : .poh_mixed_time = rec->poh_mixed_ticks ==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, rec->poh_mixed_ticks ),
912 0 : };
913 0 : }
914 0 : ev->txn_timing_cnt = cnt;
915 :
916 0 : ev->first_transaction_scheduled_time = first_dispatch==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, first_dispatch );
917 0 : }
918 :
919 : static void
920 : report_block_completed( fd_replay_tile_t * ctx,
921 : fd_bank_t * bank,
922 : int is_leader,
923 0 : fd_replay_slot_completed_t const * slot_info ) {
924 0 : if( FD_LIKELY( !fd_event_tl ) ) return;
925 :
926 0 : fd_event_block_completed_t * ev = ctx->block_completed_event;
927 0 : memset( ev, 0, FD_EVENT_BLOCK_COMPLETED_PREFIX_SZ );
928 :
929 0 : block_completed_event_fill_bank( ctx, ev, bank );
930 :
931 0 : if( FD_LIKELY( slot_info->cost_tracker.block_cost_limit!=ULONG_MAX ) ) {
932 0 : ev->cost_tracker_block_cost = slot_info->cost_tracker.block_cost;
933 0 : ev->cost_tracker_allocated_accounts_data_size = slot_info->cost_tracker.allocated_accounts_data_size;
934 0 : ev->cost_tracker_block_cost_limit = slot_info->cost_tracker.block_cost_limit;
935 0 : ev->cost_tracker_account_cost_limit = slot_info->cost_tracker.account_cost_limit;
936 0 : }
937 0 : ev->cost_tracker_pool_idx = slot_info->cost_tracker.pool_idx;
938 :
939 0 : ev->root_slot = ctx->consensus_root_slot;
940 0 : ev->storage_slot = ctx->published_root_slot;
941 0 : ev->caught_up = !!ctx->caught_up;
942 0 : ev->turbine_slot = ctx->catch_up_max_fec_slot==ULONG_MAX ? 0UL : ctx->catch_up_max_fec_slot;
943 0 : ev->fork_width = ctx->banks->curr_fork_width;
944 0 : ev->snapshot_in_progress = !!ctx->snapmk.active;
945 0 : ev->live_bank_count = fd_banks_pool_used_cnt( ctx->banks );
946 0 : ev->is_leader = is_leader;
947 :
948 0 : ev->pack_end_reason = FD_EVENT_BLOCK_COMPLETED_PACK_END_REASON_NOT_LEADER;
949 0 : if( FD_UNLIKELY( is_leader ) ) block_completed_event_fill_leader( ctx, ev, bank );
950 :
951 0 : ev->dead = 0;
952 0 : ev->dead_reason = FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD;
953 0 : ev->dead_time = 0UL;
954 0 : ev->abandoned = 0;
955 0 : ev->abandoned_reason = FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED;
956 0 : ev->abandoned_time = 0UL;
957 :
958 0 : if( FD_UNLIKELY( is_leader ) ) {
959 0 : block_completed_event_fill_leader_txn_timing( ctx, ev, bank );
960 0 : } else {
961 0 : ulong tslot = ctx->timing_slot_of_bank[ bank->idx ];
962 0 : if( FD_LIKELY( tslot!=fd_timing_slot_pool_idx_null( ctx->timing_slot_pool ) ) ) {
963 0 : fd_replay_txn_timing_slot_t const * slot = fd_timing_slot_pool_ele( ctx->timing_slot_pool, tslot );
964 0 : fd_replay_txn_timing_t const * rec = ctx->timing_rec + tslot*ctx->max_txn_per_slot;
965 0 : ulong cnt = fd_ulong_min( slot->cnt, FD_EVENT_BLOCK_COMPLETED_TXN_TIMING_MAX );
966 0 : for( ulong i=0UL; i<cnt; i++ ) {
967 0 : fd_replay_txn_timing_t const * t = &rec[ i ];
968 0 : ev->txn_timing[ i ] = (fd_event_block_completed_txn_timing_t){
969 0 : .received_time = (ulong)t->received_ns, /* already wallclock */
970 0 : .parsed_time = t->parsed_ticks ==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, t->parsed_ticks ),
971 0 : .sigverify_dispatched_time = t->sigverify_disp_ticks==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, t->sigverify_disp_ticks ),
972 0 : .sigverify_done_time = t->sigverify_done_ticks==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, t->sigverify_done_ticks ),
973 0 : .dispatched_time = t->exec_disp_ticks ==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, t->exec_disp_ticks ),
974 0 : .replayed_time = t->exec_done_ticks ==LONG_MAX ? 0UL : (ulong)fd_clock_epoch_y( ctx->clock->epoch, t->exec_done_ticks ),
975 0 : };
976 0 : }
977 0 : ev->txn_timing_cnt = cnt;
978 0 : }
979 0 : }
980 :
981 0 : if( FD_UNLIKELY( !ev->first_transaction_scheduled_time ) ) ev->first_transaction_scheduled_time = ev->last_transaction_finished_time;
982 :
983 0 : fd_event_report_block_completed( ev );
984 0 : }
985 :
986 : static void
987 : update_metric_identity_balance( fd_replay_tile_t * ctx,
988 : fd_accdb_fork_id_t fork_id,
989 0 : fd_pubkey_t const * identity ) {
990 0 : ulong identity_balance = fd_accdb_lamports( ctx->accdb, fork_id, identity->uc );
991 0 : FD_MGAUGE_SET( REPLAY, IDENTITY_BALANCE_LAMPORTS, identity_balance );
992 0 : }
993 :
994 : static void
995 : update_metric_vote_account( fd_replay_tile_t * ctx,
996 : fd_bank_t const * bank,
997 : fd_accdb_fork_id_t fork_id,
998 0 : fd_pubkey_t const * vote_key ) {
999 0 : ulong epoch_credits = 0UL;
1000 0 : fd_acc_t ro = fd_accdb_read_one( ctx->accdb, fork_id, vote_key->uc );
1001 : /* Unstaked voters are never admitted; don't flag them. Once set,
1002 : sticky until an epoch boundary actually admits the account. */
1003 0 : int admitted = fd_vote_stakes_query_t_1( fd_bank_vote_stakes( bank ), bank->vote_stakes_fork_id, vote_key, NULL );
1004 : /* SIMD-0123: delegator rewards do not count towards the balance
1005 : https://github.com/anza-xyz/agave/blob/v4.4/vote/src/vote_account.rs#L225-L234 */
1006 0 : ulong available_balance = ro.lamports;
1007 0 : if( FD_FEATURE_ACTIVE_BANK( bank, block_revenue_sharing ) ) {
1008 0 : ulong pending_delegator_rewards = 0UL;
1009 0 : if( FD_LIKELY( !fd_vote_account_pending_delegator_rewards( ro.data, ro.data_len, &pending_delegator_rewards ) ) ) {
1010 0 : available_balance = fd_ulong_sat_sub( ro.lamports, pending_delegator_rewards );
1011 0 : }
1012 0 : }
1013 0 : if( FD_UNLIKELY( !fd_stakes_vote_account_is_admissible( bank, available_balance, ro.owner, ro.data, ro.data_len ) ) ) {
1014 0 : ctx->vote_account_inadmissible = ctx->vote_account_staked || admitted;
1015 0 : } else if( FD_UNLIKELY( ctx->vote_account_inadmissible ) ) {
1016 0 : ctx->vote_account_inadmissible = ctx->vote_account_staked && !admitted;
1017 0 : }
1018 :
1019 0 : if( FD_LIKELY( ro.lamports ) ) {
1020 0 : fd_vote_state_versioned_t vsv[1];
1021 0 : if( FD_LIKELY( fd_vote_state_versioned_deserialize( vsv, ro.data, ro.data_len ) ) ) {
1022 0 : fd_vote_epoch_credits_t const * ec = fd_vsv_get_epoch_credits( vsv );
1023 0 : if( !deq_fd_vote_epoch_credits_t_empty( ec ) ) {
1024 0 : fd_vote_epoch_credits_t const * last_ec = deq_fd_vote_epoch_credits_t_peek_tail_const( ec );
1025 0 : if( last_ec->epoch==bank->f.epoch ) {
1026 0 : epoch_credits = fd_ulong_sat_sub( last_ec->credits, last_ec->prev_credits ); /* credits is the lifetime total */
1027 0 : }
1028 0 : }
1029 0 : }
1030 0 : }
1031 0 : fd_accdb_unread_one( ctx->accdb, &ro );
1032 :
1033 0 : FD_MGAUGE_SET( REPLAY, EPOCH_CREDITS, epoch_credits );
1034 0 : FD_MGAUGE_SET( REPLAY, IS_VOTE_ACCOUNT_INADMISSIBLE, (ulong)ctx->vote_account_inadmissible );
1035 0 : }
1036 :
1037 : static void
1038 : update_metric_active_stake( fd_bank_t const * bank,
1039 0 : fd_pubkey_t const * vote_key ) {
1040 0 : ulong my_active_stake = 0UL;
1041 :
1042 0 : fd_vote_stakes_ele_t ele[1];
1043 0 : if( fd_vote_stakes_query_t_1( fd_bank_vote_stakes( bank ), bank->vote_stakes_fork_id, vote_key, ele ) ) my_active_stake = ele->stake;
1044 :
1045 0 : FD_MGAUGE_SET( REPLAY, ACTIVE_STAKE_LAMPORTS, my_active_stake );
1046 0 : }
1047 :
1048 : static void
1049 : update_metric_balances( fd_replay_tile_t * ctx,
1050 0 : fd_bank_t * bank ) {
1051 0 : fd_accdb_fork_id_t fork_id = bank->accdb_fork_id;
1052 0 : fd_node_info_t node_info[1]; fd_node_info_read( node_info, ctx->node_info );
1053 0 : if( !fd_pubkey_check_zero( &node_info->identity ) ) {
1054 0 : update_metric_identity_balance( ctx, fork_id, &node_info->identity );
1055 0 : }
1056 :
1057 0 : if( !fd_pubkey_check_zero( &node_info->vote_account ) ) {
1058 0 : update_metric_vote_account( ctx, bank, fork_id, &node_info->vote_account );
1059 0 : update_metric_active_stake( bank, &node_info->vote_account );
1060 0 : }
1061 0 : FD_MGAUGE_SET( REPLAY, CLUSTER_ACTIVE_STAKE_LAMPORTS, bank->f.total_epoch_stake );
1062 0 : }
1063 :
1064 : /* try_advance_root_ag attempts to advance root to finalized_block_id. For
1065 : something to be rooted, it must be BOTH finalized AND replayed. This
1066 : walks up the bank lineage in the block id map to find and set root.
1067 :
1068 : ON FINALIZED
1069 :
1070 : If finalized is ahead of replayed => root does not advance
1071 : If replayed is ahead of finalized => root advances
1072 :
1073 : ON REPLAYED
1074 :
1075 : If finalized is ahead of replayed => root advances
1076 : If replayed is ahead of finalized => root does not advance
1077 :
1078 : consensus_root moves straight to the finalized block, and
1079 : try_notify_consensus_root hands it off as REPLAY_SIG_ROOT_ADVANCED. */
1080 :
1081 : static void
1082 : try_advance_root_ag( fd_replay_tile_t * ctx,
1083 0 : ag_block_id_t finalized_block_id ) {
1084 :
1085 0 : if( FD_UNLIKELY( finalized_block_id.slot<=ctx->consensus_root_slot ) ) return;
1086 :
1087 0 : ag_block_id_t ancestor_block_id = finalized_block_id;
1088 0 : fd_bank_t * finalized_bank = NULL;
1089 :
1090 0 : while( FD_LIKELY( ancestor_block_id.slot>ctx->consensus_root_slot ) ) {
1091 0 : fd_block_id_ele_t * ele = fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &ancestor_block_id, NULL, ctx->block_id_arr );
1092 0 : fd_bank_t * bank = ele ? fd_banks_bank_query( ctx->banks, fd_block_id_ele_get_idx( ctx->block_id_arr, ele ) ) : NULL;
1093 0 : if( FD_UNLIKELY( !bank || bank->bank_seq!=ele->bank_seq || bank->state!=FD_BANK_STATE_FROZEN || bank->parent_idx==ULONG_MAX ) ) break; /* not replayed */
1094 0 : if( !finalized_bank ) finalized_bank = bank;
1095 0 : ancestor_block_id = ctx->block_id_arr[ bank->parent_idx ].block_info;
1096 0 : }
1097 :
1098 0 : if( FD_LIKELY( ancestor_block_id.slot==ctx->consensus_root_slot && !memcmp( ancestor_block_id.hash, ctx->consensus_root.uc, sizeof(fd_hash_t) ) ) ) {
1099 0 : ctx->consensus_root_slot = finalized_block_id.slot;
1100 0 : memcpy( ctx->consensus_root.uc, finalized_block_id.hash, sizeof(fd_hash_t) );
1101 0 : if( FD_UNLIKELY( ctx->next_leader_slot!=ULONG_MAX && finalized_block_id.slot>ctx->votor_leader->parent_slot ) ) ctx->next_leader_slot = ULONG_MAX;
1102 0 : update_metric_balances( ctx, finalized_bank ); /* Alpenglow has no optimistic confirmation */
1103 0 : return;
1104 0 : }
1105 :
1106 : /* When a block id is finalized ahead of replay, we need to cache it
1107 : and process it when replay catches up. Certificates can arrive
1108 : out of order, so lo keeps the oldest and hi the newest. */
1109 :
1110 0 : if( FD_UNLIKELY( finalized_block_id.slot<ctx->finalized_block_id_lo.slot || ctx->finalized_block_id_lo.slot<=ctx->consensus_root_slot ) ) {
1111 0 : if( FD_UNLIKELY( ctx->finalized_block_id_lo.slot>ctx->finalized_block_id_hi.slot ) ) fd_swap( ctx->finalized_block_id_lo, ctx->finalized_block_id_hi );
1112 0 : ctx->finalized_block_id_lo = finalized_block_id;
1113 0 : } else if( FD_LIKELY( finalized_block_id.slot>ctx->finalized_block_id_hi.slot ) ) {
1114 0 : ctx->finalized_block_id_hi = finalized_block_id;
1115 0 : }
1116 0 : }
1117 :
1118 : static void
1119 : publish_slot_completed( fd_replay_tile_t * ctx,
1120 : fd_stem_context_t * stem,
1121 : fd_bank_t * bank,
1122 : int is_initial,
1123 : int is_leader,
1124 : ulong execution_fees_pre_settle,
1125 : ulong priority_fees_pre_settle,
1126 0 : fd_block_footer_t const * footer ) {
1127 :
1128 0 : ulong slot = bank->f.slot;
1129 :
1130 0 : if( FD_UNLIKELY( slot>=ctx->reset_slot ) ) {
1131 0 : ctx->reset_slot = slot;
1132 : /* Do not touch next_leader_slot: try_fini_leader uses ULONG_MAX
1133 : to continue the leader window. */
1134 0 : refresh_epoch_info( ctx, bank, slot );
1135 0 : }
1136 :
1137 0 : if( FD_UNLIKELY( is_initial ) ) bank->block_completed_nanos = fd_clock_tile_now( ctx->clock );
1138 :
1139 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->idx ];
1140 :
1141 : /* HACKY: hacky way of checking if we should send a null parent block
1142 : id */
1143 0 : fd_hash_t block_id = block_id_ele->dmr;
1144 0 : fd_hash_t parent_block_id = {0};
1145 0 : if( FD_LIKELY( !is_initial ) ) {
1146 0 : fd_block_id_ele_t const * parent_ele = &ctx->block_id_arr[ bank->parent_idx ];
1147 0 : parent_block_id = parent_ele->dmr;
1148 0 : }
1149 :
1150 0 : fd_hash_t const * bank_hash = &bank->f.bank_hash;
1151 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->f.block_hash_queue );
1152 0 : FD_TEST( block_hash );
1153 :
1154 0 : if( FD_LIKELY( !is_initial ) ) fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
1155 :
1156 0 : fd_epoch_schedule_t const * epoch_schedule = &bank->f.epoch_schedule;
1157 0 : ulong slot_idx;
1158 0 : ulong epoch = fd_slot_to_epoch( epoch_schedule, slot, &slot_idx );
1159 :
1160 0 : ctx->metrics.slots_total++;
1161 0 : ctx->metrics.transactions_total = bank->f.parent_txn_count + bank->f.txn_count;
1162 :
1163 : /* Caught up once replay completes a slot within a few slots of the
1164 : cluster tip. Require the tip to have advanced a few times first so
1165 : a brief view of the tip right after boot does not count. */
1166 0 : if( FD_UNLIKELY( !ctx->caught_up && !is_initial &&
1167 0 : ctx->catch_up_tip_advance_cnt>=12UL &&
1168 0 : ctx->catch_up_max_fec_slot<slot+3UL ) ) {
1169 0 : ctx->caught_up = 1;
1170 0 : FD_MGAUGE_SET( REPLAY, IS_CAUGHT_UP, 1UL );
1171 0 : metrics_trigger( ctx, FD_METRICS_SAMPLE_REASON_CAUGHT_UP, slot );
1172 0 : double boot_secs = (double)(fd_log_wallclock()-ctx->boot_timestamp_nanos)/1e9;
1173 0 : FD_LOG_NOTICE(( "caught up to cluster at slot %s%lu%s %s(%.1f seconds since boot)%s",
1174 0 : fd_log_style_bold(), slot, fd_log_style_normal(),
1175 0 : fd_log_style_dim(), boot_secs, fd_log_style_normal() ));
1176 0 : }
1177 :
1178 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1179 0 : slot_info->slot = slot;
1180 0 : slot_info->root_slot = ctx->consensus_root_slot;
1181 0 : slot_info->storage_slot = ctx->published_root_slot;
1182 0 : slot_info->epoch = epoch;
1183 0 : slot_info->slot_in_epoch = slot_idx;
1184 0 : slot_info->slots_per_epoch = fd_epoch_slot_cnt( epoch_schedule, epoch );
1185 0 : slot_info->block_height = bank->f.block_height;
1186 0 : slot_info->parent_slot = bank->f.parent_slot;
1187 0 : slot_info->block_id = block_id;
1188 0 : slot_info->parent_block_id = parent_block_id;
1189 0 : slot_info->bank_hash = *bank_hash;
1190 0 : slot_info->block_hash = *block_hash;
1191 0 : slot_info->transaction_count = bank->f.parent_txn_count + bank->f.txn_count;
1192 :
1193 0 : fd_inflation_t inflation = bank->f.inflation;
1194 0 : slot_info->inflation.foundation = inflation.foundation;
1195 0 : slot_info->inflation.foundation_term = inflation.foundation_term;
1196 0 : slot_info->inflation.terminal = inflation.terminal;
1197 0 : slot_info->inflation.initial = inflation.initial;
1198 0 : slot_info->inflation.taper = inflation.taper;
1199 :
1200 0 : fd_rent_t rent = bank->f.rent;
1201 0 : slot_info->rent.burn_percent = rent.burn_percent;
1202 0 : slot_info->rent.lamports_per_uint8_year = rent.lamports_per_uint8_year;
1203 0 : slot_info->rent.exemption_threshold = rent.exemption_threshold;
1204 :
1205 0 : slot_info->first_fec_set_received_nanos = bank->first_fec_set_received_nanos;
1206 0 : slot_info->preparation_begin_nanos = bank->preparation_begin_nanos;
1207 0 : slot_info->first_transaction_scheduled_nanos = bank->first_transaction_scheduled_nanos;
1208 0 : slot_info->last_transaction_finished_nanos = bank->last_transaction_finished_nanos;
1209 0 : slot_info->completion_time_nanos = fd_clock_tile_now( ctx->clock );
1210 0 : slot_info->footer = footer ? *footer : (fd_block_footer_t){0};
1211 0 : if( !slot_info->first_transaction_scheduled_nanos ) { /* edge case: empty slot */
1212 0 : slot_info->first_transaction_scheduled_nanos = slot_info->last_transaction_finished_nanos;
1213 0 : }
1214 :
1215 : /* refcnt should be incremented by 1 for each consumer that uses
1216 : `bank_idx`. Each consumer should decrement the bank's refcnt once
1217 : they are done using the bank. */
1218 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->refcnt++; /* rpc tile */
1219 0 : slot_info->bank_idx = bank->idx;
1220 0 : slot_info->bank_seq = bank->bank_seq;
1221 0 : slot_info->accdb_fork_id = bank->accdb_fork_id;
1222 :
1223 0 : fd_bank_t * parent_bank = fd_banks_get_parent( ctx->banks, bank );
1224 0 : slot_info->parent_bank_idx = parent_bank ? parent_bank->idx : ULONG_MAX;
1225 0 : slot_info->parent_bank_seq = parent_bank ? parent_bank->bank_seq : ULONG_MAX;
1226 0 : if( FD_LIKELY( parent_bank ) ) {
1227 0 : ulong total_txn_cnt = bank->f.txn_count;
1228 0 : ulong nonvote_txn_cnt = bank->f.nonvote_txn_count;
1229 0 : ulong failed_txn_cnt = bank->f.failed_txn_count;
1230 0 : ulong nonvote_failed_txn_cnt = bank->f.nonvote_failed_txn_count;
1231 :
1232 0 : slot_info->nonvote_success = nonvote_txn_cnt - nonvote_failed_txn_cnt;
1233 0 : slot_info->nonvote_failed = nonvote_failed_txn_cnt;
1234 0 : slot_info->vote_failed = failed_txn_cnt - nonvote_failed_txn_cnt;
1235 0 : slot_info->vote_success = total_txn_cnt - nonvote_txn_cnt - slot_info->vote_failed;
1236 0 : } else {
1237 0 : slot_info->vote_failed = ULONG_MAX;
1238 0 : slot_info->vote_success = ULONG_MAX;
1239 0 : slot_info->nonvote_success = ULONG_MAX;
1240 0 : slot_info->nonvote_failed = ULONG_MAX;
1241 0 : }
1242 :
1243 0 : slot_info->is_leader = is_leader;
1244 0 : slot_info->transaction_fee = execution_fees_pre_settle;
1245 0 : slot_info->transaction_fee -= (slot_info->transaction_fee>>1); /* burn */
1246 0 : slot_info->priority_fee = priority_fees_pre_settle;
1247 0 : slot_info->tips = bank->f.tips;
1248 0 : slot_info->shred_cnt = bank->f.shred_cnt;
1249 :
1250 0 : slot_info->voted = replay_reward_cert_voted( ctx, bank, &slot_info->voted_rank, &slot_info->vote_count );
1251 :
1252 0 : slot_info->vote_balance = ULONG_MAX;
1253 0 : slot_info->vote_commission = USHORT_MAX;
1254 0 : if( FD_UNLIKELY( ctx->has_vote_account && slot%512UL==0UL ) ) {
1255 0 : fd_acc_t acc = fd_accdb_read_one( ctx->accdb, bank->accdb_fork_id, ctx->vote_account->uc );
1256 0 : if( FD_LIKELY( acc.lamports ) ) {
1257 0 : slot_info->vote_balance = acc.lamports;
1258 :
1259 0 : ushort bps;
1260 0 : if( FD_LIKELY( !fd_vote_account_commission_bps( acc.data, acc.data_len, FD_FEATURE_ACTIVE_BANK( bank, commission_rate_in_basis_points ), &bps ) ) ) slot_info->vote_commission = bps;
1261 0 : }
1262 0 : fd_accdb_unread_one( ctx->accdb, &acc );
1263 0 : }
1264 :
1265 0 : publish_replay_out( ctx, stem, REPLAY_SIG_SLOT_COMPLETED, sizeof(fd_replay_slot_completed_t) );
1266 :
1267 0 : fd_replay_next_leader_t * next_leader = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1268 0 : next_leader->slot = ctx->reset_slot;
1269 0 : next_leader->next_leader_slot = ctx->leader_bank ? ctx->leader_bank->f.slot :
1270 0 : query_next_leader_slot( ctx, fd_ulong_max( ctx->reset_slot+1UL, fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot+1UL ) ) );
1271 0 : publish_replay_out( ctx, stem, REPLAY_SIG_NEXT_LEADER, sizeof(fd_replay_next_leader_t) );
1272 :
1273 : /* Skip the telemetry event for the initial boot block (snapshot /
1274 : genesis): it was not replayed. */
1275 0 : if( FD_LIKELY( !is_initial ) ) report_block_completed( ctx, bank, is_leader, slot_info );
1276 0 : timing_slot_release( ctx, bank->idx );
1277 :
1278 0 : if( FD_UNLIKELY( slot==ctx->finalized_block_id_lo.slot && !memcmp( block_id.uc, ctx->finalized_block_id_lo.hash, sizeof(fd_hash_t) ) ) ) { /* finalized ahead of replay, replay caught up */
1279 0 : try_advance_root_ag( ctx, ctx->finalized_block_id_lo );
1280 0 : ctx->finalized_block_id_lo = ctx->finalized_block_id_hi; /* hi becomes lo once lo is reached */
1281 0 : }
1282 0 : }
1283 :
1284 : static void
1285 : report_block_incomplete( fd_replay_tile_t * ctx,
1286 : ulong slot,
1287 : fd_hash_t const * block_id,
1288 : fd_bank_t * bank,
1289 : int dead_reason,
1290 0 : int abandoned_reason ) {
1291 0 : if( FD_LIKELY( !fd_event_tl ) ) return;
1292 :
1293 0 : fd_event_block_completed_t * ev = ctx->block_completed_event;
1294 0 : memset( ev, 0, FD_EVENT_BLOCK_COMPLETED_PREFIX_SZ );
1295 0 : if( FD_LIKELY( bank ) ) {
1296 0 : block_completed_event_fill_bank( ctx, ev, bank );
1297 0 : ev->is_leader = !!bank->is_leader;
1298 0 : } else {
1299 0 : ev->slot = slot;
1300 0 : if( FD_LIKELY( ctx->notified_root_bank ) ) ev->epoch = fd_slot_to_epoch( &ctx->notified_root_bank->f.epoch_schedule, slot, NULL );
1301 0 : fd_memcpy( ev->block_id, block_id->uc, sizeof(ev->block_id) );
1302 0 : block_completed_event_fill_reception( ctx, ev, ULONG_MAX );
1303 0 : }
1304 0 : ev->root_slot = ctx->consensus_root_slot;
1305 0 : ev->storage_slot = ctx->published_root_slot;
1306 0 : ev->caught_up = !!ctx->caught_up;
1307 0 : ev->turbine_slot = ctx->catch_up_max_fec_slot==ULONG_MAX ? 0UL : ctx->catch_up_max_fec_slot;
1308 0 : ev->fork_width = ctx->banks->curr_fork_width;
1309 0 : ev->snapshot_in_progress = !!ctx->snapmk.active;
1310 0 : ev->live_bank_count = fd_banks_pool_used_cnt( ctx->banks );
1311 :
1312 0 : int dead = dead_reason !=FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD;
1313 0 : int abandoned = abandoned_reason!=FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED;
1314 0 : long now = fd_clock_tile_now( ctx->clock );
1315 :
1316 0 : ev->dead = dead;
1317 0 : ev->dead_reason = dead_reason;
1318 0 : ev->dead_time = dead ? (ulong)now : 0UL;
1319 0 : ev->abandoned = abandoned;
1320 0 : ev->abandoned_reason = abandoned_reason;
1321 0 : ev->abandoned_time = abandoned ? (ulong)now : 0UL;
1322 :
1323 0 : ev->pack_end_reason = FD_EVENT_BLOCK_COMPLETED_PACK_END_REASON_NOT_LEADER;
1324 0 : if( FD_UNLIKELY( bank && bank->is_leader ) ) {
1325 0 : block_completed_event_fill_leader( ctx, ev, bank );
1326 0 : block_completed_event_fill_leader_txn_timing( ctx, ev, bank );
1327 0 : }
1328 :
1329 0 : fd_event_report_block_completed( ev );
1330 0 : }
1331 :
1332 : static void
1333 : publish_slot_dead( fd_replay_tile_t * ctx,
1334 : fd_stem_context_t * stem,
1335 : ulong slot,
1336 : fd_hash_t const * block_id,
1337 0 : fd_block_footer_t const * footer ) {
1338 0 : fd_replay_slot_dead_t * slot_dead = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1339 0 : slot_dead->slot = slot;
1340 0 : slot_dead->block_id = *block_id;
1341 0 : slot_dead->footer = footer ? *footer : (fd_block_footer_t){0};
1342 0 : publish_replay_out( ctx, stem, REPLAY_SIG_SLOT_DEAD, sizeof(fd_replay_slot_dead_t) );
1343 0 : }
1344 :
1345 : static void
1346 : publish_txn_executed( fd_replay_tile_t * ctx,
1347 : fd_stem_context_t * stem,
1348 : ulong bank_idx,
1349 0 : ulong txn_idx ) {
1350 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
1351 :
1352 0 : FD_TEST( txn_info->index_in_slot<ctx->max_txn_per_slot );
1353 0 : ulong tslot = ctx->timing_slot_of_bank[ bank_idx ];
1354 0 : if( FD_LIKELY( tslot!=fd_timing_slot_pool_idx_null( ctx->timing_slot_pool ) ) ) {
1355 0 : fd_replay_txn_timing_slot_t * slot = fd_timing_slot_pool_ele( ctx->timing_slot_pool, tslot );
1356 0 : fd_replay_txn_timing_t * t = ctx->timing_rec + tslot*ctx->max_txn_per_slot + txn_info->index_in_slot;
1357 0 : t->received_ns = txn_info->received_ns;
1358 0 : t->parsed_ticks = txn_info->tick_parsed;
1359 0 : t->sigverify_disp_ticks = txn_info->tick_sigverify_disp;
1360 0 : t->sigverify_done_ticks = txn_info->tick_sigverify_done;
1361 0 : t->exec_disp_ticks = txn_info->tick_exec_disp;
1362 0 : t->exec_done_ticks = txn_info->tick_exec_done;
1363 0 : slot->cnt = fd_ulong_max( slot->cnt, txn_info->index_in_slot+1UL );
1364 0 : }
1365 :
1366 0 : fd_replay_txn_executed_t * txn_executed = fd_type_pun( fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk ) );
1367 0 : *txn_executed->txn = *fd_sched_get_txn( ctx->sched, txn_idx );
1368 0 : txn_executed->txn_err = txn_info->txn_err;
1369 0 : txn_executed->is_committable = !!(txn_info->flags&FD_SCHED_TXN_IS_COMMITTABLE);
1370 0 : txn_executed->is_fees_only = !!(txn_info->flags&FD_SCHED_TXN_IS_FEES_ONLY);
1371 0 : txn_executed->is_noop = !!(txn_info->flags&FD_SCHED_TXN_IS_NOOP);
1372 0 : txn_executed->is_simple_vote = txn_info->is_simple_vote;
1373 0 : txn_executed->tick_parsed = txn_info->tick_parsed;
1374 0 : txn_executed->tick_sigverify_disp = txn_info->tick_sigverify_disp;
1375 0 : txn_executed->tick_sigverify_done = txn_info->tick_sigverify_done;
1376 0 : txn_executed->tick_exec_disp = txn_info->tick_exec_disp;
1377 0 : txn_executed->tick_exec_done = txn_info->tick_exec_done;
1378 :
1379 0 : txn_executed->tick_load_start = txn_info->tick_load_start;
1380 0 : txn_executed->tick_check_start = txn_info->tick_check_start;
1381 0 : txn_executed->tick_exec_start = txn_info->tick_exec_start;
1382 0 : txn_executed->tick_commit_start = txn_info->tick_commit_start;
1383 0 : txn_executed->tick_commit_end = txn_info->tick_commit_end;
1384 :
1385 0 : txn_executed->slot = txn_info->slot;
1386 0 : txn_executed->bank_seq = txn_info->bank_seq;
1387 0 : txn_executed->index_in_slot = txn_info->index_in_slot;
1388 0 : txn_executed->exec_tile_idx = txn_info->exec_tile_idx;
1389 0 : txn_executed->sigverify_exec_tile_idx = txn_info->sigverify_exec_tile_idx;
1390 0 : txn_executed->compute_units_consumed = txn_info->compute_units_consumed;
1391 0 : txn_executed->max_compute_units = txn_info->max_compute_units;
1392 0 : txn_executed->transaction_fee = txn_info->transaction_fee;
1393 0 : txn_executed->priority_fee = txn_info->priority_fee;
1394 0 : txn_executed->tips = txn_info->tips;
1395 0 : publish_replay_out( ctx, stem, REPLAY_SIG_TXN_EXECUTED, sizeof(*txn_executed) );
1396 0 : }
1397 :
1398 : /* Emit the runtime_block event for a finalized bank (replayed or
1399 : leader-produced). The fee arguments are the bank's values snapshotted
1400 : before fd_runtime_block_execute_finalize, which settles (zeroes) the
1401 : fee fields and applies the tip commission. */
1402 :
1403 : static void
1404 : replay_runtime_block_emit( fd_replay_tile_t * ctx,
1405 : fd_bank_t * bank,
1406 : ulong execution_fees,
1407 : ulong priority_fees,
1408 0 : ulong tips ) {
1409 0 : fd_block_id_ele_t const * ele = &ctx->block_id_arr[ bank->idx ];
1410 0 : fd_hash_t const * block_id = &ele->dmr;
1411 0 : fd_hash_t parent_block_id = {0};
1412 0 : if( FD_LIKELY( bank->parent_idx!=ULONG_MAX ) ) {
1413 0 : fd_block_id_ele_t const * parent_ele = &ctx->block_id_arr[ bank->parent_idx ];
1414 0 : parent_block_id = parent_ele->dmr;
1415 0 : }
1416 0 : fd_pubkey_t leader = {0};
1417 0 : fd_pubkey_t const * _leader = fd_multi_epoch_leaders_get_leader_for_slot( ctx->mleaders, bank->f.slot );
1418 0 : if( FD_LIKELY( _leader ) ) leader = *_leader;
1419 0 : fd_sol_sysvar_clock_t clock = {0};
1420 0 : if( FD_UNLIKELY( !fd_sysvar_clock_read( ctx->accdb, bank->accdb_fork_id, &clock ) ) ) FD_LOG_ERR(( "failed to read clock sysvar for slot %lu", bank->f.slot ));
1421 0 : fd_event_runtime_block_emit( bank, block_id->uc, parent_block_id.uc, leader.uc,
1422 0 : execution_fees, priority_fees, tips, bank->f.shred_cnt, &clock,
1423 0 : ctx->fec_chain + bank->idx*FD_FEC_BLK_MAX,
1424 0 : ctx->block_id_arr[ bank->idx ].fec_cnt );
1425 0 : }
1426 :
1427 : static void
1428 : mark_bank_dead( fd_replay_tile_t * ctx,
1429 : fd_stem_context_t * stem,
1430 : ulong bank_idx,
1431 : int dead_reason,
1432 : int abandoned_reason,
1433 : fd_block_footer_t const * footer );
1434 :
1435 :
1436 : /**********************************************************************/
1437 : /* Leader bank management */
1438 : /**********************************************************************/
1439 :
1440 : static fd_bank_t *
1441 : prepare_leader_bank( fd_replay_tile_t * ctx,
1442 : fd_bank_t * parent_bank,
1443 : ulong slot,
1444 0 : long now ) {
1445 :
1446 : /* Make sure that we are not already leader. */
1447 0 : FD_TEST( ctx->leader_bank==NULL );
1448 :
1449 0 : ctx->leader_bank = fd_banks_new_bank( ctx->banks, parent_bank->idx, now, 1 );
1450 0 : if( FD_UNLIKELY( !ctx->leader_bank ) ) {
1451 0 : FD_LOG_CRIT(( "invariant violation: leader bank is NULL for slot %lu", slot ));
1452 0 : }
1453 :
1454 0 : ctx->leader_bank = fd_banks_clone_from_parent( ctx->banks, ctx->leader_bank->idx );
1455 0 : if( FD_UNLIKELY( !ctx->leader_bank ) ) {
1456 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for slot %lu", slot ));
1457 0 : }
1458 :
1459 : /* pin the cost tracker for the duration of the leader slot */
1460 0 : fd_cost_tracker_store_pin( fd_bank_cost_tracker( ctx->leader_bank ), ctx->leader_bank->cost_tracker_fork_id );
1461 :
1462 0 : ctx->leader_bank->preparation_begin_nanos = now;
1463 :
1464 0 : ctx->leader_bank->f.slot = slot;
1465 0 : ctx->leader_bank->leader = ctx->identity_pubkey[ 0 ];
1466 :
1467 : /* Clear the previous occupant of this reused bank idx from whichever
1468 : map it is currently keyed in. */
1469 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ ctx->leader_bank->idx ];
1470 0 : if( FD_LIKELY( fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &block_id_ele->block_info, NULL, ctx->block_id_arr )==block_id_ele ) ) {
1471 0 : FD_TEST( fd_ag_block_id_map_ele_remove( ctx->ag_block_id_map, &block_id_ele->block_info, NULL, ctx->block_id_arr ) );
1472 0 : }
1473 :
1474 0 : block_id_ele->block_id_seen = 0;
1475 0 : block_id_ele->slot = slot;
1476 0 : block_id_ele->bank_seq = ctx->leader_bank->bank_seq;
1477 0 : block_id_ele->latest_fec_idx = 0U;
1478 0 : block_id_ele->fec_cnt = 0U;
1479 0 : memset( &block_id_ele->latest_mr, 0, sizeof(fd_hash_t) );
1480 0 : memset( &block_id_ele->dmr, 0, sizeof(fd_hash_t) );
1481 :
1482 0 : ctx->leader_bank->txncache_fork_id = fd_txncache_attach_child ( ctx->txncache, parent_bank->txncache_fork_id );
1483 0 : ctx->leader_bank->progcache_fork_id = fd_progcache_attach_child( ctx->progcache, parent_bank->progcache_fork_id );
1484 0 : ctx->leader_bank->accdb_fork_id = fd_accdb_attach_child ( ctx->accdb, parent_bank->accdb_fork_id );
1485 0 : ctx->leader_bank->parent_accdb_fork_id = parent_bank->accdb_fork_id;
1486 :
1487 0 : int is_epoch_boundary = 0;
1488 0 : fd_runtime_block_execute_prepare( ctx->banks, ctx->leader_bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
1489 0 : FD_CHECK_CRIT( !is_epoch_boundary || FD_FEATURE_ACTIVE_BANK( ctx->leader_bank, alpenglow ), "alpenglow feature is not active" );
1490 :
1491 0 : ulong max_tick_height;
1492 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( parent_bank->f.ticks_per_slot, slot, &max_tick_height ) ) ) {
1493 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, parent_bank->f.ticks_per_slot ));
1494 0 : }
1495 0 : ctx->leader_bank->f.max_tick_height = max_tick_height;
1496 :
1497 : /* Now that a bank has been created for the leader slot, increment the
1498 : reference count until we are done with the leader slot. */
1499 0 : ctx->leader_bank->refcnt++;
1500 :
1501 0 : return ctx->leader_bank;
1502 0 : }
1503 :
1504 : static inline void
1505 0 : maybe_switch_identity( fd_replay_tile_t * ctx ) {
1506 :
1507 0 : if( FD_LIKELY( fd_keyswitch_state_query( ctx->keyswitch )!=FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) return;
1508 :
1509 : /* Switch identity */
1510 :
1511 0 : FD_LOG_DEBUG(( "keyswitch: switching identity" ));
1512 :
1513 0 : memcpy( ctx->identity_pubkey, ctx->keyswitch->bytes, 32UL );
1514 0 : ctx->identity_dirty = 1;
1515 :
1516 0 : ctx->metrics.voted_slot = ULONG_MAX;
1517 :
1518 0 : fd_node_info_write_begin( ctx->node_info );
1519 0 : ctx->node_info->info.identity = *ctx->identity_pubkey;
1520 0 : fd_node_info_write_end ( ctx->node_info );
1521 :
1522 0 : ctx->next_leader_query_start = ULONG_MAX;
1523 :
1524 0 : ctx->identity_vote_rooted = 0;
1525 0 : ctx->identity_idx++;
1526 0 : fd_vote_tracker_reset( ctx->vote_tracker );
1527 :
1528 : /* identity_vote_rooted was just reset, so the gate rests on the flag
1529 : alone and a surviving auto would read as armed (see fd_wfs.h). */
1530 0 : if( FD_UNLIKELY( ctx->wait_for_vote_to_start_leader==FD_BOOLAU_AUTO ) ) {
1531 0 : FD_LOG_CRIT(( "wait_for_vote_to_start_leader unresolved (auto) at identity rotation" ));
1532 0 : }
1533 :
1534 : /* Save the current sequence so the voter knows what sequence to
1535 : consume up to. */
1536 0 : ctx->keyswitch->result = fd_mcache_seq_query( ctx->slot_out_seq );
1537 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
1538 0 : }
1539 :
1540 : static void
1541 : construct_footer_certs( fd_replay_tile_t const * ctx,
1542 : ulong leader_slot,
1543 : ulong migration_slot,
1544 0 : fd_block_footer_t * footer ) {
1545 0 : fd_votor_certed_t const * fin = ctx->votor_final;
1546 0 : int use_fast = fin->slot!=ULONG_MAX && fin->kind==AG_CERT_KIND_FAST_FINAL;
1547 0 : int use_slow = fin->slot!=ULONG_MAX && fin->kind==AG_CERT_KIND_FINAL;
1548 :
1549 0 : footer->has_fast_final_cert = use_fast;
1550 0 : footer->has_final_cert = use_slow;
1551 0 : if( footer->has_fast_final_cert ) {
1552 0 : footer->has_fast_final_cert = fd_block_footer_cert_from_agg( &footer->fast_final_cert, fin->slot, fin->block_id.uc, &fin->agg );
1553 0 : }
1554 0 : if( footer->has_final_cert ) {
1555 0 : footer->has_final_cert = fd_block_footer_cert_from_agg( &footer->final_cert, fin->slot, NULL, &fin->agg ) &&
1556 0 : fd_block_footer_cert_from_agg( &footer->notar_cert, fin->slot, fin->block_id.uc, &fin->agg2 );
1557 0 : }
1558 :
1559 0 : int reward_ok = migration_slot!=ULONG_MAX &&
1560 0 : leader_slot>=migration_slot+FD_NUM_SLOTS_FOR_REWARD+1UL;
1561 0 : if( FD_LIKELY( reward_ok ) ) {
1562 0 : ulong reward_slot = leader_slot-FD_NUM_SLOTS_FOR_REWARD;
1563 0 : fd_votor_reward_t const * reward = &ctx->votor_reward[ reward_slot%(FD_NUM_SLOTS_FOR_REWARD+AG_SLOTS_PER_WINDOW+1UL) ];
1564 0 : footer->has_notar_reward_cert = reward->slot==reward_slot && !fd_bls_set_is_null( reward->agg_notar.set ) && fd_block_footer_cert_from_agg( &footer->notar_reward_cert, reward_slot, reward->block_id.uc, &reward->agg_notar );
1565 0 : footer->has_skip_reward_cert = reward->slot==reward_slot && !fd_bls_set_is_null( reward->agg_skip.set ) && fd_block_footer_cert_from_agg( &footer->skip_reward_cert, reward_slot, NULL, &reward->agg_skip );
1566 0 : }
1567 0 : }
1568 :
1569 : static void
1570 : publish_leader_footer( fd_replay_tile_t * ctx,
1571 : fd_stem_context_t * stem,
1572 : ulong slot,
1573 0 : fd_block_footer_t const * footer ) {
1574 0 : fd_replay_leader_footer_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1575 0 : msg->slot = slot;
1576 0 : msg->footer = *footer;
1577 :
1578 0 : ulong sz = sizeof(fd_replay_leader_footer_t);
1579 0 : publish_replay_out( ctx, stem, REPLAY_SIG_LEADER_FOOTER, sz );
1580 0 : }
1581 :
1582 : /* For the most part, try_become_leader_ag mirrors try_become_leader
1583 : with modifications for Alpenglow-specific changes, in particular
1584 : 1. no more PoH-hashing and 2. picking the reset bank is based on
1585 : ParentReady rather than fork choice. */
1586 :
1587 : static inline int
1588 : try_become_leader_ag( fd_replay_tile_t * ctx,
1589 : fd_stem_context_t * stem,
1590 : uint reason,
1591 0 : ulong end_slot ) {
1592 :
1593 : /* Mirrors check from try_become_leader. */
1594 :
1595 0 : if( FD_LIKELY( ctx->next_leader_slot==ULONG_MAX ||
1596 0 : ctx->is_leader ||
1597 : /* (!ctx->identity_vote_rooted && ctx->wait_for_vote_to_start_leader) || // deprecated */
1598 0 : ctx->replay_out->idx==ULONG_MAX ||
1599 0 : !ctx->wfs_complete ) ) {
1600 0 : return 0;
1601 0 : }
1602 :
1603 0 : ulong parent_slot = ctx->votor_leader->parent_slot;
1604 0 : fd_hash_t const * parent_block_id = &ctx->votor_leader->parent_block_id;
1605 :
1606 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_ele_query( ctx, parent_block_id, parent_slot );
1607 0 : if( FD_UNLIKELY( !block_id_ele ) ) return 0;
1608 0 : fd_bank_t * reset_bank = fd_banks_bank_query( ctx->banks, fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele ) );
1609 0 : if( FD_UNLIKELY( !reset_bank || reset_bank->bank_seq!=block_id_ele->bank_seq || reset_bank->state!=FD_BANK_STATE_FROZEN ) ) return 0;
1610 :
1611 : /* A finalization above the parent, or a root that excludes it, means
1612 : we should give up our leader slot. */
1613 0 : if( FD_UNLIKELY( ( ctx->votor_final->slot!=ULONG_MAX && ctx->votor_final->slot>parent_slot ) ||
1614 0 : ( ctx->consensus_root_slot!=ULONG_MAX && ctx->consensus_root_slot>parent_slot ) ||
1615 0 : ( ctx->consensus_root_slot==parent_slot && !fd_hash_eq( &ctx->consensus_root, parent_block_id ) ) ) ) {
1616 0 : ctx->next_leader_slot = ULONG_MAX;
1617 0 : return 0;
1618 0 : }
1619 :
1620 : /* Certs made a later window ParentReady after votor sent this one:
1621 : the cluster skipped it. */
1622 0 : if( FD_UNLIKELY( ctx->next_leader_slot<ctx->votor_parent_ready_slot ) ) {
1623 0 : ctx->metrics.leader_window_superseded++;
1624 0 : ctx->next_leader_slot = ULONG_MAX;
1625 0 : return 0;
1626 0 : }
1627 :
1628 0 : if( FD_UNLIKELY( !fd_banks_can_start_bank( ctx->banks ) ) ) return 0;
1629 0 : if( FD_UNLIKELY( ctx->halt_replay ) ) return 0;
1630 0 : if( !ctx->supports_leader ) return 0;
1631 0 : if( FD_UNLIKELY( fd_accdb_cmd_pending( ctx->accdb ) ) ) return 0; /* leader bank attaches an accdb fork */
1632 :
1633 : /* Don't become leader if the slot is not scheduled for the identity.
1634 : This can only happen in cases where the identity just switched. */
1635 0 : fd_pubkey_t const * leader = fd_multi_epoch_leaders_get_leader_for_slot( ctx->mleaders, ctx->next_leader_slot );
1636 0 : if( FD_UNLIKELY( !leader || !fd_pubkey_eq( leader, ctx->identity_pubkey ) ) ) {
1637 0 : ctx->next_leader_slot = ULONG_MAX;
1638 0 : return 0;
1639 0 : }
1640 :
1641 : /* The cluster already produced a slot of this window, eg. our own
1642 : past slots replayed during catch-up after a restart. */
1643 0 : ulong window_end = ctx->next_leader_slot - ctx->next_leader_slot%AG_SLOTS_PER_WINDOW + AG_SLOTS_PER_WINDOW;
1644 0 : for( ulong idx=0UL; idx<ctx->block_id_len; idx++ ) {
1645 0 : fd_block_id_ele_t const * ele = &ctx->block_id_arr[ idx ];
1646 0 : if( FD_LIKELY( ele->slot<ctx->next_leader_slot || ele->slot>=window_end ) ) continue;
1647 0 : fd_bank_t const * bank = fd_banks_bank_query( ctx->banks, idx );
1648 0 : if( FD_UNLIKELY( bank && bank->bank_seq==ele->bank_seq && !bank->is_leader ) ) {
1649 0 : ctx->metrics.leader_slot_already_produced++;
1650 0 : ctx->next_leader_slot = ULONG_MAX;
1651 0 : return 0;
1652 0 : }
1653 0 : }
1654 :
1655 : /* In Alpenglow, the "reset" block is signaled by ParentReady (a state
1656 : transition in the Votor consensus logic). ParentReady can occur
1657 : ahead of replay, so we publish reset here, once we know we are
1658 : certain we can build off the ParentReady bank. */
1659 :
1660 0 : ctx->reset_dmr = *parent_block_id;
1661 0 : ctx->reset_slot = parent_slot;
1662 0 : ctx->reset_timestamp_nanos = fd_clock_tile_now( ctx->clock );
1663 :
1664 0 : FD_TEST( ctx->next_leader_slot>ctx->reset_slot );
1665 0 : long now = fd_tickcount();
1666 0 : ctx->next_leader_tickcount = now;
1667 :
1668 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1669 :
1670 0 : reset->bank_idx = reset_bank->idx;
1671 0 : reset->timestamp = ctx->reset_timestamp_nanos;
1672 0 : reset->completed_slot = ctx->reset_slot;
1673 0 : reset->hashcnt_per_tick = reset_bank->f.slot_params.hashes_per_tick;
1674 0 : reset->ticks_per_slot = reset_bank->f.ticks_per_slot;
1675 0 : reset->tick_duration_ns = reset_bank->f.slot_params.ns_per_slot_adjusted/reset->ticks_per_slot;
1676 :
1677 0 : fd_memcpy( reset->completed_cmr, &block_id_ele->latest_mr, sizeof(fd_hash_t) );
1678 0 : fd_memcpy( reset->completed_dmr, parent_block_id, sizeof(fd_hash_t) );
1679 :
1680 0 : fd_blockhashes_t const * block_hash_queue = &reset_bank->f.block_hash_queue;
1681 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
1682 0 : FD_TEST( last_hash );
1683 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
1684 :
1685 0 : reset->max_microblocks_in_slot = FD_POH_ALPENGLOW_MAX_MICROBLOCKS_PER_SLOT;
1686 0 : reset->next_leader_slot = ctx->next_leader_slot;
1687 0 : reset->wfs_paused = !ctx->wfs_complete;
1688 :
1689 0 : publish_replay_out( ctx, stem, REPLAY_SIG_RESET, sizeof(fd_poh_reset_t) );
1690 :
1691 : /* The remainder of this function mirrors try_become_leader. */
1692 :
1693 0 : long now_nanos = fd_clock_epoch_y( ctx->clock->epoch, now );
1694 :
1695 0 : ctx->is_leader = 1;
1696 0 : ctx->recv_poh = 0;
1697 :
1698 0 : memset( &ctx->leader_stats, 0, sizeof(ctx->leader_stats) );
1699 0 : ctx->leader_stats.slot = ctx->next_leader_slot;
1700 0 : ctx->leader_stats.timing_table_idx = ULONG_MAX;
1701 0 : ctx->leader_stats.became_leader_nanos = now_nanos;
1702 0 : ctx->leader_stats.leader_slot_start_nanos = fd_clock_epoch_y( ctx->clock->epoch, ctx->next_leader_tickcount );
1703 :
1704 0 : FD_TEST( ctx->highwater_leader_slot==ULONG_MAX || ctx->highwater_leader_slot<ctx->next_leader_slot );
1705 0 : ctx->highwater_leader_slot = ctx->next_leader_slot;
1706 :
1707 0 : FD_LOG_INFO(( "becoming leader for slot %lu, parent slot is %lu", ctx->next_leader_slot, ctx->reset_slot ));
1708 :
1709 0 : fd_bank_t * bank = prepare_leader_bank( ctx, reset_bank, ctx->next_leader_slot, now_nanos );
1710 :
1711 0 : fd_bundle_crank_tip_payment_config_t config[1] = { 0 };
1712 0 : fd_pubkey_t tip_receiver_owner = {0};
1713 :
1714 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
1715 0 : fd_acct_addr_t tip_payment_config[1];
1716 0 : fd_acct_addr_t tip_receiver[1];
1717 0 : fd_bundle_crank_get_addresses( ctx->bundle.gen, bank->f.epoch, tip_payment_config, tip_receiver );
1718 :
1719 0 : fd_acc_t tip_config_acc = fd_accdb_read_one( ctx->accdb, bank->accdb_fork_id, tip_payment_config->b );
1720 0 : if( FD_UNLIKELY( !tip_config_acc.lamports ) ) {
1721 0 : FD_BASE58_ENCODE_32_BYTES( tip_payment_config->b, tip_config_acc_b58 );
1722 0 : FD_LOG_WARNING(( "tip payment config account %s does not exist", tip_config_acc_b58 ));
1723 0 : fd_accdb_unread_one( ctx->accdb, &tip_config_acc );
1724 0 : } else if( FD_UNLIKELY( tip_config_acc.data_len<sizeof(fd_bundle_crank_tip_payment_config_t) ) ) {
1725 0 : FD_LOG_HEXDUMP_WARNING(( "invalid tip payment config account data", tip_config_acc.data, tip_config_acc.data_len ));
1726 0 : fd_accdb_unread_one( ctx->accdb, &tip_config_acc );
1727 0 : } else {
1728 0 : memcpy( config, tip_config_acc.data, sizeof(fd_bundle_crank_tip_payment_config_t) );
1729 0 : fd_accdb_unread_one( ctx->accdb, &tip_config_acc );
1730 0 : }
1731 :
1732 : /* It is possible that the tip receiver account does not exist yet
1733 : if it is the first time in an epoch. */
1734 0 : fd_acc_t tip_receiver_acc = fd_accdb_read_one( ctx->accdb, bank->accdb_fork_id, tip_receiver->b );
1735 0 : if( FD_LIKELY( tip_receiver_acc.lamports ) ) {
1736 0 : fd_memcpy( tip_receiver_owner.uc, tip_receiver_acc.owner, 32UL );
1737 0 : }
1738 0 : fd_accdb_unread_one( ctx->accdb, &tip_receiver_acc );
1739 0 : }
1740 :
1741 :
1742 : /* Like Agave's block_timeout, the block ends relative to the window's
1743 : ParentReady. If that is already past, end at once rather than run
1744 : into votor's skip timeout. */
1745 0 : long slot_end_ns = ctx->votor_leader->parent_ready_ns
1746 0 : + (long)( ( ctx->next_leader_slot%AG_SLOTS_PER_WINDOW+1UL )*bank->f.slot_params.ns_per_slot )
1747 0 : - (long)FD_TARGET_SLOT_ADJUSTMENT_NS - AG_TIME_TO_COMPLETE_BROADCAST_NS;
1748 :
1749 0 : fd_became_leader_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1750 0 : msg->slot = ctx->next_leader_slot;
1751 0 : msg->block_height = bank->f.block_height;
1752 0 : msg->slot_start_ns = now_nanos;
1753 0 : msg->slot_end_ns = fd_long_max( slot_end_ns, now_nanos );
1754 0 : msg->bank_idx = bank->idx;
1755 0 : msg->bank_seq = bank->bank_seq;
1756 0 : msg->ticks_per_slot = bank->f.ticks_per_slot;
1757 0 : msg->hashcnt_per_tick = 1UL; /* one tick per block, no hash budget (see replay_block_start) */
1758 0 : msg->tick_duration_ns = bank->f.slot_params.ns_per_slot_adjusted/msg->ticks_per_slot;
1759 0 : msg->bundle->config[0] = config[0];
1760 0 : memcpy( msg->bundle->last_blockhash, bank->f.poh.hash, sizeof(fd_hash_t) );
1761 0 : memcpy( msg->bundle->tip_receiver_owner, tip_receiver_owner.uc, sizeof(fd_pubkey_t) );
1762 :
1763 0 : msg->max_microblocks_in_slot = FD_POH_ALPENGLOW_MAX_MICROBLOCKS_PER_SLOT;
1764 :
1765 0 : msg->total_skipped_ticks = 0UL; /* even when slots are skipped, ticks increment by exactly one for every block */
1766 0 : msg->epoch = fd_slot_to_epoch( &bank->f.epoch_schedule, ctx->next_leader_slot, NULL );
1767 :
1768 0 : fd_cost_tracker_t const * cost_tracker = fd_cost_tracker_store_peek( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id );
1769 :
1770 0 : msg->limits.slot_max_cost = cost_tracker->block_cost_limit;
1771 0 : msg->limits.slot_max_vote_cost = FD_PACK_MAX_VOTE_COST_PER_BLOCK_UPPER_BOUND;
1772 0 : msg->limits.slot_max_write_cost_per_acct = cost_tracker->account_cost_limit;
1773 0 : msg->limits.slot_max_allocated_data_per_block = cost_tracker->data_size_limit;
1774 0 : msg->limits.slot_max_data_shreds = bank->f.slot_params.max_shred_idx;
1775 :
1776 0 : publish_replay_out( ctx, stem, REPLAY_SIG_BECAME_LEADER, sizeof(fd_became_leader_t) );
1777 0 : metrics_trigger( ctx, reason, reason==FD_METRICS_SAMPLE_REASON_LEADER_START ? msg->slot : end_slot );
1778 :
1779 0 : ctx->leader_window_start_ns = ctx->votor_leader->parent_ready_ns;
1780 0 : ctx->next_leader_slot = ULONG_MAX;
1781 0 : ctx->next_leader_tickcount = LONG_MAX;
1782 :
1783 0 : return 1;
1784 0 : }
1785 :
1786 : /* replay_block_complete checks the footer's bank hash, then freezes and
1787 : publishes a finalized bank, whose footer certs all verified. Returns
1788 : 1 if it marked the bank dead. */
1789 :
1790 : static int
1791 : replay_block_complete( fd_replay_tile_t * ctx,
1792 : fd_stem_context_t * stem,
1793 : fd_bank_t * bank,
1794 : fd_block_footer_t const * footer,
1795 : ulong execution_fees_pre_settle,
1796 0 : ulong priority_fees_pre_settle ) {
1797 0 : fd_hash_t const * footer_bank_hash = &footer->bank_hash;
1798 0 : if( FD_UNLIKELY( memcmp( footer_bank_hash->uc, bank->f.bank_hash.uc, sizeof(fd_hash_t) ) ) ) {
1799 0 : FD_BASE58_ENCODE_32_BYTES( footer_bank_hash->uc, footer_bank_hash_b58 );
1800 0 : FD_BASE58_ENCODE_32_BYTES( bank->f.bank_hash.uc, executed_bank_hash_b58 );
1801 0 : FD_LOG_WARNING(( "slot %lu: bank hash mismatch, footer declares %s but executed %s. ", bank->f.slot, footer_bank_hash_b58, executed_bank_hash_b58 ));
1802 0 : mark_bank_dead( ctx, stem, bank->idx, FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_BAD_FOOTER, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED, footer );
1803 0 : return 1;
1804 0 : } else {
1805 0 : FD_BASE58_ENCODE_32_BYTES( footer_bank_hash->uc, footer_bank_hash_b58 );
1806 0 : FD_BASE58_ENCODE_32_BYTES( bank->f.bank_hash.uc, executed_bank_hash_b58 );
1807 0 : FD_LOG_INFO(( "slot %lu: bank hash matches, footer declares %s, executed %s", bank->f.slot, footer_bank_hash_b58, executed_bank_hash_b58 ));
1808 0 : }
1809 :
1810 : /* Copy out cost tracker fields before freezing */
1811 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1812 0 : cost_tracker_snap( bank, slot_info );
1813 :
1814 : /* fetch identity balance infrequently and after set-identity. */
1815 0 : slot_info->identity_balance = ULONG_MAX;
1816 0 : if( FD_UNLIKELY( ctx->identity_dirty || bank->f.slot%4096UL==0UL ) ) {
1817 0 : ctx->identity_dirty = 0;
1818 0 : slot_info->identity_balance = fd_accdb_lamports( ctx->accdb, bank->accdb_fork_id, ctx->identity_pubkey->uc );
1819 0 : }
1820 :
1821 : /* Mark the bank as frozen. */
1822 0 : fd_block_id_ele_t const * block_id_ele = &ctx->block_id_arr[ bank->idx ];
1823 0 : bank->f.block_id = block_id_ele->dmr;
1824 0 : fd_banks_mark_bank_frozen( bank );
1825 0 : bank->block_completed_nanos = fd_clock_tile_now( ctx->clock );
1826 :
1827 : /**********************************************************************/
1828 : /* Bank hash comparison, and halt if there's a mismatch after replay */
1829 : /**********************************************************************/
1830 :
1831 : /* Must be last so we can measure completion time correctly, even
1832 : though we could technically do this before the hash cmp. */
1833 0 : publish_slot_completed( ctx, stem, bank, 0, 0 /* is_leader */, execution_fees_pre_settle, priority_fees_pre_settle, footer );
1834 :
1835 : /* The footer's finalization cert can also finalize. */
1836 0 : if( footer->has_fast_final_cert ) { try_advance_root_ag( ctx, ag_block_id( footer->fast_final_cert.slot, footer->fast_final_cert.block_id.uc ) ); }
1837 0 : else if( footer->has_final_cert ) { try_advance_root_ag( ctx, ag_block_id( footer->final_cert.slot, footer->notar_cert.block_id.uc ) ); }
1838 :
1839 : /* If enabled, dump the block to a file and reset the dumping
1840 : context state */
1841 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
1842 0 : fd_dump_block_to_protobuf( ctx->block_dump_ctx, ctx->banks, bank, ctx->accdb, ctx->dump_proto_ctx, ctx->runtime_stack );
1843 0 : fd_block_dump_context_reset( ctx->block_dump_ctx );
1844 0 : }
1845 0 : return 0;
1846 0 : }
1847 :
1848 : static int
1849 : replay_block_finalize( fd_replay_tile_t * ctx,
1850 : fd_stem_context_t * stem,
1851 0 : fd_bank_t * bank ) {
1852 0 : bank->last_transaction_finished_nanos = fd_clock_tile_now( ctx->clock );
1853 :
1854 : /* Set poh hash in bank. */
1855 0 : fd_hash_t * poh = fd_sched_get_poh( ctx->sched, bank->idx );
1856 0 : bank->f.poh = *poh;
1857 :
1858 : /* Set shred count in bank. */
1859 0 : bank->f.shred_cnt = fd_sched_get_shred_cnt( ctx->sched, bank->idx );
1860 :
1861 0 : ulong execution_fees_pre_settle = bank->f.execution_fees;
1862 0 : ulong priority_fees_pre_settle = bank->f.priority_fees;
1863 0 : ulong tips_pre_settle = bank->f.tips;
1864 :
1865 0 : fd_block_footer_t const * footer = NULL;
1866 0 : footer = fd_sched_get_footer( ctx->sched, bank->idx );
1867 :
1868 : /* Footer certs go to idle exec tiles as cert verify tasks, the rest
1869 : verify here, then finalize applies the footer while the tasks run.
1870 : The bank completes on its last cert task done. */
1871 0 : fd_block_footer_cert_t const * cert [ FD_ALPENGLOW_FOOTER_CERT_MAX ];
1872 0 : uint tag [ FD_ALPENGLOW_FOOTER_CERT_MAX ];
1873 0 : ulong quorum[ FD_ALPENGLOW_FOOTER_CERT_MAX ];
1874 :
1875 0 : ulong cnt = 0UL;
1876 0 : ulong tiles = 0UL;
1877 0 : if( footer && !ctx->capture_ctx && !ctx->report_runtime_diffs ) {
1878 0 : cnt = fd_alpenglow_footer_certs( footer, cert, tag, quorum );
1879 0 : tiles = fd_sched_cert_dispatch( ctx->sched, bank->idx, cnt );
1880 0 : }
1881 :
1882 0 : if( FD_UNLIKELY( !tiles ) ) {
1883 : /* Nowhere to dispatch to ... just run cert verification here, now. */
1884 0 : if( FD_UNLIKELY( fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx, footer, ctx->shred_version, ctx->footer_cache ) ) ) goto bad_footer;
1885 0 : } else {
1886 0 : FD_TEST( ctx->shred_version );
1887 0 : ulong k = 0UL;
1888 0 : for( ; tiles; k++, tiles=fd_ulong_pop_lsb( tiles ) ) {
1889 0 : ulong t = (ulong)fd_ulong_find_lsb( tiles );
1890 :
1891 0 : ctx->cert_payload_sz[ t ] = fd_alpenglow_cert_payload( cert[ k ], tag[ k ], ctx->shred_version, ctx->cert_payload[ t ] );
1892 :
1893 0 : blst_p2_affine const * hm = fd_footer_cache_hash_query( ctx->footer_cache, ctx->cert_payload[ t ], ctx->cert_payload_sz[ t ] );
1894 :
1895 0 : fd_replay_out_link_t * exec_out = &ctx->exec_out[ t ];
1896 0 : fd_execrp_cert_verify_msg_t * msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1897 0 : msg->bank_idx = bank->idx;
1898 0 : msg->quorum = quorum[ k ];
1899 0 : msg->tag = tag[ k ];
1900 0 : msg->shred_version = ctx->shred_version;
1901 0 : msg->has_hm = !!hm;
1902 0 : if( hm ) msg->hm = *hm;
1903 0 : *msg->cert = *cert[ k ];
1904 :
1905 0 : bank->refcnt++;
1906 :
1907 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_CERT_VERIFY<<32) | t, exec_out->chunk, sizeof(fd_execrp_cert_verify_msg_t), 0UL, 0UL, 0UL );
1908 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*msg), exec_out->chunk0, exec_out->wmark );
1909 0 : }
1910 0 : bank->cert_task_cnt = k;
1911 0 : bank->execution_fees_pre_settle = execution_fees_pre_settle;
1912 0 : bank->priority_fees_pre_settle = priority_fees_pre_settle;
1913 0 : for( ulong i=k; i<cnt; i++ ) {
1914 0 : if( FD_UNLIKELY( fd_alpenglow_cert_verify( bank, cert[ i ], tag[ i ], quorum[ i ], ctx->shred_version, ctx->footer_cache, NULL, NULL ) ) ) goto bad_footer;
1915 0 : }
1916 0 : if( FD_UNLIKELY( fd_runtime_block_execute_finalize_unverified( bank, ctx->accdb, footer, ctx->footer_cache ) ) ) goto bad_footer;
1917 0 : }
1918 :
1919 0 : if( FD_UNLIKELY( ctx->report_runtime_diffs ) ) replay_runtime_block_emit( ctx, bank, execution_fees_pre_settle, priority_fees_pre_settle, tips_pre_settle );
1920 :
1921 0 : if( FD_UNLIKELY( bank->cert_task_cnt ) ) return 0; /* completes on its last cert task done */
1922 0 : return replay_block_complete( ctx, stem, bank, footer, execution_fees_pre_settle, priority_fees_pre_settle );
1923 :
1924 0 : bad_footer:
1925 0 : mark_bank_dead( ctx, stem, bank->idx, FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_BAD_FOOTER, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED, NULL );
1926 0 : return 1;
1927 0 : }
1928 :
1929 : static int
1930 : try_fini_leader( fd_replay_tile_t * ctx,
1931 0 : fd_stem_context_t * stem ) {
1932 :
1933 : /* If we are leader, we can only unbecome the leader iff we have
1934 : received the poh hash from the poh tile and block id from reasm.
1935 : The block id entry is claimed for the leader slot in
1936 : prepare_leader_bank, so a slot mismatch here means the claim
1937 : discipline broke. */
1938 :
1939 0 : if( FD_LIKELY( !ctx->is_leader ) ) return 0;
1940 0 : if( !ctx->recv_poh ) return 0;
1941 :
1942 : /* The consensus root only advances through FROZEN banks and the
1943 : leader bank is frozen only below, so a root at or past our slot
1944 : can never include the leader bank.
1945 :
1946 : We can't include votor-only messages, because it's possible our
1947 : block was retransmitted but replay hasn't processed the FEC yet
1948 : (due to bespoke repair <=> replay concurrency). */
1949 :
1950 0 : if( FD_UNLIKELY( !ctx->block_id_arr[ ctx->leader_bank->idx ].block_id_seen &&
1951 0 : ctx->consensus_root_slot!=ULONG_MAX &&
1952 0 : ctx->consensus_root_slot>=ctx->leader_bank->f.slot ) ) {
1953 0 : ulong bank_idx = ctx->leader_bank->idx;
1954 0 : metrics_trigger( ctx, FD_METRICS_SAMPLE_REASON_LEADER_END, ctx->leader_bank->f.slot );
1955 0 : ctx->leader_bank->refcnt--;
1956 0 : ctx->leader_bank = NULL;
1957 0 : ctx->recv_poh = 0;
1958 0 : ctx->is_leader = 0;
1959 0 : mark_bank_dead( ctx, stem, bank_idx, FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_PRUNED, NULL );
1960 0 : return 1;
1961 0 : }
1962 :
1963 0 : if( !ctx->block_id_arr[ ctx->leader_bank->idx ].block_id_seen ) return 0;
1964 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->idx ].slot==ctx->leader_bank->f.slot );
1965 :
1966 0 : ulong curr_slot = ctx->leader_bank->f.slot;
1967 :
1968 0 : ulong execution_fees_pre_settle;
1969 0 : ulong priority_fees_pre_settle;
1970 0 : ulong tips_pre_settle;
1971 :
1972 : /* Already finalized above, when the footer was published. */
1973 :
1974 0 : execution_fees_pre_settle = ctx->leader_execution_fees;
1975 0 : priority_fees_pre_settle = ctx->leader_priority_fees;
1976 0 : tips_pre_settle = ctx->leader_tips;
1977 :
1978 0 : ctx->leader_bank->f.shred_cnt = (ulong)ctx->block_id_arr[ ctx->leader_bank->idx ].fec_cnt*FD_FEC_SHRED_CNT;
1979 :
1980 0 : if( FD_UNLIKELY( ctx->report_runtime_diffs ) ) replay_runtime_block_emit( ctx, ctx->leader_bank, execution_fees_pre_settle, priority_fees_pre_settle, tips_pre_settle );
1981 :
1982 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1983 0 : cost_tracker_snap( ctx->leader_bank, slot_info );
1984 0 : slot_info->identity_balance = ULONG_MAX;
1985 0 : if( FD_UNLIKELY( ctx->identity_dirty || curr_slot%4096UL==0UL ) ) {
1986 0 : ctx->identity_dirty = 0;
1987 0 : slot_info->identity_balance = fd_accdb_lamports( ctx->accdb, ctx->leader_bank->accdb_fork_id, ctx->identity_pubkey->uc );
1988 0 : }
1989 :
1990 0 : ctx->leader_bank->f.block_id = ctx->block_id_arr[ ctx->leader_bank->idx ].dmr;
1991 0 : fd_banks_mark_bank_frozen( ctx->leader_bank );
1992 0 : ctx->leader_bank->block_completed_nanos = fd_clock_tile_now( ctx->clock );
1993 :
1994 0 : publish_slot_completed( ctx, stem, ctx->leader_bank, 0, 1 /* is_leader */, execution_fees_pre_settle, priority_fees_pre_settle, ctx->leader_footer );
1995 :
1996 : /* The reference on the bank is finally no longer needed. */
1997 0 : ctx->leader_bank->refcnt--;
1998 :
1999 0 : fd_bank_t * completed = ctx->leader_bank;
2000 :
2001 : /* We are no longer leader so we can clear the bank index we use for
2002 : being the leader. */
2003 0 : ctx->leader_bank = NULL;
2004 0 : ctx->recv_poh = 0;
2005 0 : ctx->is_leader = 0;
2006 :
2007 0 : if( FD_UNLIKELY( ctx->next_leader_slot==ULONG_MAX && (curr_slot+1UL)%AG_SLOTS_PER_WINDOW ) ) {
2008 0 : *ctx->votor_leader = (fd_votor_leader_t){
2009 0 : .slot = curr_slot+1UL,
2010 0 : .parent_slot = curr_slot,
2011 0 : .parent_block_id = ctx->block_id_arr[ completed->idx ].dmr,
2012 0 : .parent_ready_ns = ctx->leader_window_start_ns
2013 0 : };
2014 0 : ctx->next_leader_slot = curr_slot+1UL;
2015 0 : if( try_become_leader_ag( ctx, stem, FD_METRICS_SAMPLE_REASON_LEADER_END_START, curr_slot ) ) return 1;
2016 0 : }
2017 :
2018 0 : metrics_trigger( ctx, FD_METRICS_SAMPLE_REASON_LEADER_END, curr_slot );
2019 0 : return 1;
2020 0 : }
2021 :
2022 : static void
2023 : publish_root_advanced( fd_replay_tile_t * ctx,
2024 : fd_stem_context_t * stem,
2025 0 : fd_bank_t * bank ) {
2026 :
2027 : /* If the new consensus root is in the next epoch from the one the
2028 : replay tile currently holds, send the next epoch's leader schedule.
2029 : We can't use the new root's parent slot safely here. */
2030 0 : if( FD_UNLIKELY( bank->f.epoch>fd_slot_to_epoch( &bank->f.epoch_schedule, ctx->notified_root_slot, NULL ) ) ) {
2031 0 : fd_runtime_update_next_leaders( bank, ctx->runtime_stack );
2032 0 : publish_epoch_info( ctx, stem, bank, 1 );
2033 0 : }
2034 :
2035 0 : if( ctx->rpc_enabled ) {
2036 0 : bank->refcnt++;
2037 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc", bank->idx, bank->f.slot, bank->refcnt ));
2038 :
2039 : /* Under Alpenglow Agave treats a rooted bank as both confirmed
2040 : and finalized. The rpc tile holds an independent reference for
2041 : each alias. */
2042 0 : bank->refcnt++;
2043 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc (confirmed alias)", bank->idx, bank->f.slot, bank->refcnt ));
2044 0 : }
2045 :
2046 : /* Increment the reference count on the consensus root bank to account
2047 : for the number of resolv tiles that are waiting on it. */
2048 0 : bank->refcnt += ctx->resolv_tile_cnt;
2049 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for resolv", bank->idx, bank->f.slot, bank->refcnt ));
2050 :
2051 0 : fd_replay_root_advanced_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2052 : /* f.* first: gcc 11 SLP otherwise materialises the 27 KB idx..f.slot gap element-wise */
2053 0 : msg->slot = bank->f.slot;
2054 0 : msg->bank_hash = bank->f.bank_hash;
2055 0 : msg->block_id = bank->f.block_id;
2056 0 : msg->bank_idx = bank->idx;
2057 0 : msg->bank_seq = bank->bank_seq;
2058 :
2059 0 : publish_replay_out( ctx, stem, REPLAY_SIG_ROOT_ADVANCED, sizeof(fd_replay_root_advanced_t) );
2060 :
2061 : /* Drop every FEC set below the new root from the store. */
2062 :
2063 0 : fd_store_publish( ctx->store, ctx->map_join, bank->f.slot );
2064 0 : }
2065 :
2066 : /* publish_boot_slot announces the snapshot slot, which the validator
2067 : did not replay, as completed and rooted. */
2068 :
2069 : static void
2070 : publish_boot_slot( fd_replay_tile_t * ctx,
2071 : fd_stem_context_t * stem,
2072 0 : fd_bank_t * bank ) {
2073 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2074 0 : cost_tracker_snap( bank, slot_info );
2075 0 : slot_info->identity_balance = fd_accdb_lamports( ctx->accdb, bank->accdb_fork_id, ctx->identity_pubkey->uc );
2076 :
2077 0 : fd_block_footer_t footer = { .bank_hash = bank->f.bank_hash }; /* dummy footer for the snapshot slot */
2078 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */, 0, 0, &footer );
2079 0 : publish_root_advanced( ctx, stem, bank );
2080 0 : }
2081 :
2082 : /* Determine the default slot params to use for slots where no
2083 : reduce_slot_time feature gate is in effect. This is important for
2084 : the inflation calculations, which use the slot times for
2085 : historical slots as input. Therefore we need the same semantics
2086 : as Agave, even after the reduce_slot_time feature gates are
2087 : active. */
2088 : static fd_slot_params_t
2089 0 : restore_default_slot_params( fd_bank_t const * bank ) {
2090 :
2091 : /* A reduction is effective if the effective ns_per_slot is less than
2092 : the 400ms value.
2093 : https://github.com/anza-xyz/agave/blob/v4.2/runtime/src/slot_params.rs#L332-L350 */
2094 0 : int reduction_effective = fd_slot_params_lookup( &FD_SLOT_PARAMS_400MS,
2095 0 : &bank->f.features,
2096 0 : &bank->f.epoch_schedule,
2097 0 : bank->f.slot ).ns_per_slot < FD_SLOT_PARAMS_400MS.ns_per_slot;
2098 :
2099 : /* In order to behave correctly in real networks, if a reduction is
2100 : effective then we use the 400ms slot params as the default. */
2101 0 : if( reduction_effective ) {
2102 0 : return FD_SLOT_PARAMS_400MS;
2103 0 : }
2104 :
2105 : /* If a reduction is not effective, then we can rely on the slot
2106 : times having remained constant throughout the lifetime of the
2107 : cluster, and can use the slot params from the manifest. Note that
2108 : in test clusters these may differ from the 400ms values. */
2109 0 : return bank->f.slot_params;
2110 0 : }
2111 :
2112 : /* refresh_vote_account_staked recomputes vote_account_staked from the
2113 : root stake delegations and seeds vote_account_inadmissible if stake
2114 : effective in the root's epoch was not admitted at its boundary.
2115 : Admitted implies staked, so only a non-admitted vote account pays
2116 : for the scan. Called at boot and when the root crosses an epoch. */
2117 :
2118 : static void
2119 : refresh_vote_account_staked( fd_replay_tile_t * ctx,
2120 0 : fd_bank_t * bank ) {
2121 0 : fd_node_info_t node_info[1]; fd_node_info_read( node_info, ctx->node_info );
2122 0 : fd_pubkey_t const * vote_key = &node_info->vote_account;
2123 0 : int has_vote_key = !fd_pubkey_check_zero( vote_key );
2124 0 : int admitted = has_vote_key && fd_vote_stakes_query_t_1( fd_bank_vote_stakes( bank ), bank->vote_stakes_fork_id, vote_key, NULL );
2125 0 : ctx->vote_account_staked = admitted;
2126 0 : if( FD_LIKELY( admitted || !has_vote_key ) ) return;
2127 :
2128 0 : fd_stake_history_t stake_history_[1];
2129 0 : fd_stake_history_t const * stake_history = fd_sysvar_cache_stake_history_view( &bank->f.sysvar_cache, stake_history_ );
2130 0 : fd_stake_history_entry_t status = fd_stake_delegations_vote_account_status(
2131 0 : fd_banks_stake_delegations_root_query( ctx->banks ), vote_key, bank->f.epoch, stake_history,
2132 0 : &bank->f.warmup_cooldown_rate_epoch, FD_FEATURE_ACTIVE_BANK( bank, upgrade_bpf_stake_program_to_v5_1 ) );
2133 0 : ctx->vote_account_staked = status.effective || status.activating;
2134 0 : ctx->vote_account_inadmissible = !!status.effective;
2135 0 : }
2136 :
2137 : static void
2138 : init_after_snapshot( fd_replay_tile_t * ctx,
2139 0 : fd_stem_context_t * stem ) {
2140 : /* snapin built the root stake delegations while writing the accounts
2141 : db index. Refresh finalizes them in memory. */
2142 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
2143 0 : if( FD_UNLIKELY( !bank ) ) {
2144 0 : FD_LOG_CRIT(( "invariant violation: replay bank is NULL at bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
2145 0 : }
2146 :
2147 0 : char const * one_offs[ 16UL ];
2148 0 : for( ulong i=0UL; i<ctx->enable_features_cnt; i++ ) one_offs[ i ] = ctx->enable_features[ i ];
2149 0 : fd_features_enable_one_offs( &bank->f.features, one_offs, (uint)ctx->enable_features_cnt, 0UL );
2150 0 : FD_CHECK_CRIT( FD_FEATURE_ACTIVE_BANK( bank, alpenglow ), "alpenglow feature is not active" );
2151 0 : bank->f.alpenglow_migration_slot = fd_alpenglow_migration_slot( bank, ctx->accdb );
2152 :
2153 : /* Set slot params based on the feature gates in the snapshot,
2154 : and assert that these are consistent with the values from the
2155 : manifest. These assertions match Agave:
2156 : https://github.com/anza-xyz/agave/blob/v4.2/runtime/src/bank.rs#L4839-L4869 */
2157 0 : fd_slot_params_t manifest_params = bank->f.slot_params;
2158 0 : bank->f.slot_params_default = restore_default_slot_params( bank );
2159 0 : bank->f.slot_params = fd_slot_params_at_slot( bank, bank->f.slot );
2160 0 : FD_TEST( bank->f.slot_params.ns_per_slot == manifest_params.ns_per_slot );
2161 0 : FD_TEST( bank->f.slot_params.slots_per_year == manifest_params.slots_per_year );
2162 :
2163 0 : fd_runtime_update_next_leaders( bank, ctx->runtime_stack );
2164 0 : fd_runtime_update_leaders( bank, ctx->runtime_stack );
2165 :
2166 : /* Typically, when we cross an epoch boundary during normal
2167 : operation, we publish the stake weights for the new epoch. But
2168 : since we are starting from a snapshot, we need to publish two
2169 : epochs worth of stake weights: the previous epoch (which is
2170 : needed for voting on the current epoch), and the current epoch
2171 : (which is needed for voting on the next epoch). */
2172 0 : publish_epoch_info( ctx, stem, bank, 0 );
2173 0 : publish_epoch_info( ctx, stem, bank, 1 );
2174 0 : fd_pubkey_t const * leader = fd_multi_epoch_leaders_get_leader_for_slot( ctx->mleaders, bank->f.slot );
2175 0 : FD_TEST( leader );
2176 0 : bank->leader = *leader;
2177 :
2178 0 : fd_progcache_reset( ctx->progcache );
2179 0 : bank->progcache_fork_id = fd_progcache_fork_id_initial();
2180 :
2181 0 : bank->f.warmup_cooldown_rate_epoch = fd_slot_to_epoch( &bank->f.epoch_schedule, bank->f.features.reduce_stake_warmup_cooldown, NULL );
2182 0 : fd_stake_delegations_t * root_delegations = fd_banks_stake_delegations_root_query( ctx->banks );
2183 0 : fd_stake_history_t stake_history_[1];
2184 0 : fd_stake_history_t const * stake_history = fd_sysvar_cache_stake_history_view( &bank->f.sysvar_cache, stake_history_ );
2185 : /* Despite claims like https://github.com/solana-program/stake/pull/81
2186 : that the stake history sysvar is contiguous, testnet has in fact
2187 : had a gap at epoch 386. */
2188 0 : if( FD_UNLIKELY( !fd_sysvar_stake_history_is_contiguous( stake_history ) ) ) {
2189 0 : FD_LOG_INFO(( "stake history sysvar (covering epoch %lu to %lu over %lu entries) is not contiguous; some fast paths will be disabled", stake_history->entries[ 0 ].epoch, stake_history->entries[ stake_history->len-1UL ].epoch, stake_history->len ));
2190 0 : }
2191 0 : fd_stake_delegations_refresh(
2192 0 : root_delegations,
2193 0 : bank->f.epoch,
2194 0 : stake_history, /* may be NULL */
2195 0 : &bank->f.warmup_cooldown_rate_epoch,
2196 0 : FD_FEATURE_ACTIVE_BANK( bank, upgrade_bpf_stake_program_to_v5_1 ),
2197 0 : FD_FEATURE_ACTIVE_BANK( bank, remove_inactive_stakes ) );
2198 0 : fd_stake_history_entry_t totals = fd_stake_delegations_totals( root_delegations );
2199 0 : bank->f.total_effective_stake = totals.effective;
2200 0 : bank->f.total_activating_stake = totals.activating;
2201 0 : bank->f.total_deactivating_stake = totals.deactivating;
2202 :
2203 : /* Emit the stake-delegations boot baseline from the finalized root
2204 : cache (snapshot accepted and refreshed, or genesis loaded), so the
2205 : rows match the post-reconciliation cache rather than the raw
2206 : account stream. */
2207 0 : if( FD_UNLIKELY( ctx->report_runtime_diffs ) ) {
2208 0 : fd_stake_delegations_iter_t iter_[1];
2209 0 : for( fd_stake_delegations_iter_t * iter = fd_stake_delegations_iter_init( iter_, root_delegations );
2210 0 : !fd_stake_delegations_iter_done( iter );
2211 0 : fd_stake_delegations_iter_next( iter ) ) {
2212 0 : fd_stake_delegation_t const * d = fd_stake_delegations_iter_ele( iter );
2213 0 : ulong ae = d->activation_epoch ==(ushort)USHORT_MAX ? ULONG_MAX : (ulong)d->activation_epoch;
2214 0 : ulong de = d->deactivation_epoch==(ushort)USHORT_MAX ? ULONG_MAX : (ulong)d->deactivation_epoch;
2215 0 : fd_event_runtime_stake_delegation_bootup_emit( bank->f.slot, bank->f.epoch, d->stake_account.uc,
2216 0 : d->vote_account.uc, d->stake, ae, de, d->credits_observed );
2217 0 : }
2218 0 : }
2219 :
2220 0 : fd_vote_stakes_refresh( fd_bank_vote_stakes( bank ), bank->vote_stakes_fork_id, ctx->accdb, bank->accdb_fork_id );
2221 :
2222 0 : refresh_vote_account_staked( ctx, bank );
2223 :
2224 : /* After both snapshots have been loaded in, we can determine if we should
2225 : start distributing rewards. */
2226 :
2227 0 : fd_rewards_recalculate_partitioned_rewards( bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx );
2228 :
2229 : /* Signals fd_startup_gate */
2230 0 : FD_MGAUGE_SET( REPLAY, IS_RUNTIME_LOADED, 1UL );
2231 0 : metrics_trigger( ctx, FD_METRICS_SAMPLE_REASON_REPLAY_READY, bank->f.slot );
2232 0 : }
2233 :
2234 : static ulong
2235 : enforce_nanosecond_clock_bounds( fd_replay_tile_t * ctx,
2236 : fd_bank_t * bank,
2237 0 : ulong now_nanos ) {
2238 0 : long lo, hi;
2239 0 : fd_alpenglow_producer_time_bounds( bank, ctx->accdb, &lo, &hi );
2240 0 : return (ulong)fd_long_max( lo, fd_long_min( hi, (long)fd_ulong_min( now_nanos, (ulong)LONG_MAX ) ) );
2241 0 : }
2242 :
2243 : static void
2244 : process_poh_message( fd_replay_tile_t * ctx,
2245 : fd_stem_context_t * stem,
2246 0 : fd_poh_leader_slot_ended_t const * slot_ended ) {
2247 :
2248 0 : FD_TEST( ctx->is_booted );
2249 0 : FD_TEST( ctx->is_leader );
2250 0 : FD_TEST( ctx->leader_bank!=NULL );
2251 :
2252 0 : FD_TEST( ctx->highwater_leader_slot>=slot_ended->slot );
2253 0 : FD_TEST( ctx->next_leader_slot>ctx->highwater_leader_slot );
2254 :
2255 0 : if( FD_LIKELY( ctx->leader_stats.slot==slot_ended->slot ) ) {
2256 0 : ctx->leader_stats.microblock_count = slot_ended->microblock_count;
2257 0 : ctx->leader_stats.pack_block_cost = slot_ended->pack_block_cost;
2258 0 : ctx->leader_stats.pack_vote_cost = slot_ended->pack_vote_cost;
2259 0 : ctx->leader_stats.pack_data_bytes = slot_ended->pack_data_bytes;
2260 0 : ctx->leader_stats.bundle_txn_count = slot_ended->bundle_txn_count;
2261 0 : ctx->leader_stats.pack_end_reason = slot_ended->pack_end_reason;
2262 0 : ctx->leader_stats.pack_start_nanos = slot_ended->pack_start_ns;
2263 0 : ctx->leader_stats.pack_end_nanos = slot_ended->pack_end_ns;
2264 0 : ctx->leader_stats.timing_table_idx = slot_ended->timing_table_idx;
2265 0 : }
2266 :
2267 : /* unpin when we know leader bank done executing txns */
2268 0 : fd_cost_tracker_store_unpin( fd_bank_cost_tracker( ctx->leader_bank ), ctx->leader_bank->cost_tracker_fork_id );
2269 :
2270 0 : if( FD_UNLIKELY( !slot_ended->completed ) ) {
2271 : /* The leader slot was aborted by a reset mid-production. The
2272 : block-complete entry was never emitted, so no slot-complete FEC
2273 : (and thus no block id) will ever arrive. */
2274 0 : ulong bank_idx = ctx->leader_bank->idx;
2275 0 : metrics_trigger( ctx, FD_METRICS_SAMPLE_REASON_LEADER_END, slot_ended->slot );
2276 0 : ctx->leader_bank->refcnt--;
2277 0 : ctx->leader_bank = NULL;
2278 0 : ctx->recv_poh = 0;
2279 0 : ctx->is_leader = 0;
2280 0 : mark_bank_dead( ctx, stem, bank_idx, FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_RESET, NULL );
2281 0 : return;
2282 0 : }
2283 :
2284 : /* Update the poh hash in the bank. We will want to maintain a refcnt
2285 : on the bank until we have received the block id for the block after
2286 : it has been shredded. */
2287 :
2288 0 : memcpy( &ctx->leader_bank->f.poh, slot_ended->blockhash, sizeof(fd_hash_t) );
2289 :
2290 0 : ctx->recv_poh = 1;
2291 :
2292 0 : ctx->leader_bank->last_transaction_finished_nanos = fd_clock_tile_now( ctx->clock );
2293 0 : fd_sched_block_add_done( ctx->sched, ctx->leader_bank->idx, ctx->leader_bank->parent_idx, ctx->leader_bank->f.slot );
2294 :
2295 : /* cache fees before runtime zeros */
2296 :
2297 0 : ctx->leader_execution_fees = ctx->leader_bank->f.execution_fees;
2298 0 : ctx->leader_priority_fees = ctx->leader_bank->f.priority_fees;
2299 0 : ctx->leader_tips = ctx->leader_bank->f.tips;
2300 :
2301 0 : fd_block_footer_t * footer = ctx->leader_footer;
2302 0 : fd_memset( footer, 0, sizeof(fd_block_footer_t) );
2303 0 : footer->block_producer_time_nanos = enforce_nanosecond_clock_bounds( ctx, ctx->leader_bank, (ulong)fd_clock_tile_now( ctx->clock ) );
2304 :
2305 0 : ulong migration_slot = fd_alpenglow_migration_slot( ctx->leader_bank, ctx->accdb );
2306 0 : construct_footer_certs( ctx, ctx->leader_bank->f.slot, migration_slot, footer );
2307 :
2308 : /* The block goes out regardless: the certs are already committed to
2309 : the bank hash, so there is nothing left to fall back to. */
2310 0 : if( FD_UNLIKELY( fd_runtime_block_execute_finalize( ctx->leader_bank, ctx->accdb, ctx->capture_ctx, footer, ctx->shred_version, ctx->footer_cache ) ) ) {
2311 0 : FD_LOG_WARNING(( "slot %lu: our own block footer certs did not apply; the block we produce will be dead to the cluster", ctx->leader_bank->f.slot ));
2312 0 : }
2313 0 : footer->bank_hash = ctx->leader_bank->f.bank_hash;
2314 :
2315 0 : publish_leader_footer( ctx, stem, ctx->leader_bank->f.slot, footer );
2316 0 : }
2317 :
2318 : static void
2319 : boot_genesis( fd_replay_tile_t * ctx,
2320 : fd_stem_context_t * stem,
2321 0 : fd_genesis_meta_t const * meta ) {
2322 :
2323 0 : if( FD_UNLIKELY( ctx->wfs_enabled ) ) {
2324 0 : FD_LOG_ERR(( "wait_for_supermajority is enabled but the validator is booting from genesis. "
2325 0 : "Remove [consensus.wait_for_supermajority_at_slot] and "
2326 0 : "[consensus.wait_for_supermajority_with_bank_hash] configuration and restart." ));
2327 0 : }
2328 :
2329 : /* Firedancer only boots Alpenglow clusters from genesis. The
2330 : genesis certificate certifies slot 0 with a zero block id. */
2331 0 : fd_memset( &ctx->initial_block_id, 0, sizeof(fd_hash_t) );
2332 :
2333 : /* If we are bootstrapping, we can't wait to wait for our identity
2334 : vote to be rooted as this creates a circular dependency. */
2335 0 : ctx->identity_vote_rooted = 1;
2336 :
2337 0 : ctx->caught_up = 1;
2338 0 : FD_MGAUGE_SET( REPLAY, IS_CAUGHT_UP, 1UL );
2339 0 : metrics_trigger( ctx, FD_METRICS_SAMPLE_REASON_CAUGHT_UP, 0UL );
2340 :
2341 0 : ctx->hard_fork_cnt = 0UL;
2342 :
2343 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
2344 0 : FD_TEST( meta->bootstrap && meta->has_lthash );
2345 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
2346 :
2347 0 : fd_bank_t * bank = fd_banks_init_bank( ctx->banks );
2348 0 : FD_TEST( bank );
2349 0 : bank->f.slot = 0UL;
2350 0 : FD_TEST( bank->idx==FD_REPLAY_BOOT_BANK_IDX );
2351 :
2352 0 : static const fd_accdb_fork_id_t accdb_root = { .val = USHORT_MAX };
2353 0 : bank->accdb_fork_id = fd_accdb_attach_child( ctx->accdb, accdb_root );
2354 0 : bank->parent_accdb_fork_id = bank->accdb_fork_id;
2355 :
2356 0 : fd_runtime_read_genesis( ctx->banks, bank, ctx->accdb, NULL, &meta->genesis_hash, &meta->lthash, ctx->genesis, genesis_blob, ctx->runtime_stack );
2357 0 : FD_TEST( fd_sysvar_cache_restore( bank, ctx->accdb ) );
2358 :
2359 0 : bank->txncache_fork_id = fd_txncache_attach_child ( ctx->txncache, (fd_txncache_fork_id_t){USHORT_MAX} );
2360 0 : bank->progcache_fork_id = fd_progcache_attach_child( ctx->progcache, fd_progcache_fork_id_initial() );
2361 :
2362 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->f.block_hash_queue );
2363 0 : fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
2364 :
2365 : /* We call this after fd_runtime_read_genesis, which sets up the
2366 : slot_bank needed in blockstore_init. */
2367 0 : init_after_snapshot( ctx, stem );
2368 :
2369 0 : ctx->published_root_slot = 0UL;
2370 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, 0UL );
2371 :
2372 0 : bank->f.block_height = 1UL;
2373 :
2374 0 : ctx->consensus_root = ctx->initial_block_id;
2375 0 : ctx->consensus_root_slot = 0UL;
2376 0 : ctx->finalized_block_id_lo = ag_block_id( 0UL, ctx->initial_block_id.uc );
2377 0 : ctx->finalized_block_id_hi = ctx->finalized_block_id_lo;
2378 0 : ctx->notified_root = ctx->initial_block_id;
2379 0 : ctx->notified_root_slot = 0UL;
2380 0 : ctx->notified_root_bank = bank;
2381 0 : ctx->published_root_slot = 0UL;
2382 0 : ctx->published_root_bank_idx = 0UL;
2383 0 : if( FD_UNLIKELY( ctx->snapmk.full_interval_blocks ) ) {
2384 0 : ulong interval = ctx->snapmk.full_interval_blocks;
2385 0 : ctx->snapmk.next_full_block_height = ((bank->f.block_height/interval)+1UL)*interval;
2386 0 : }
2387 0 : if( FD_UNLIKELY( ctx->snapmk.incremental_interval_blocks ) ) {
2388 0 : ulong interval = ctx->snapmk.incremental_interval_blocks;
2389 0 : ctx->snapmk.next_incremental_block_height = ((bank->f.block_height/interval)+1UL)*interval;
2390 0 : }
2391 :
2392 0 : ctx->reset_slot = 0UL;
2393 0 : ctx->reset_dmr = ctx->initial_block_id;
2394 0 : ctx->epoch_end_slot = 0UL;
2395 0 : ctx->slots_per_epoch = 0UL;
2396 0 : ctx->ns_per_slot = 0UL;
2397 0 : ctx->reset_timestamp_nanos = fd_clock_tile_now( ctx->clock );
2398 :
2399 0 : ctx->has_cluster_type = 1;
2400 :
2401 0 : ctx->is_booted = 1;
2402 :
2403 0 : fd_hash_t initial_block_id = ctx->initial_block_id;
2404 :
2405 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
2406 0 : block_id_ele->latest_mr = initial_block_id;
2407 0 : block_id_ele->dmr = initial_block_id;
2408 0 : block_id_ele->slot = 0UL;
2409 0 : block_id_ele->bank_seq = bank->bank_seq;
2410 0 : block_id_ele->block_id_seen = 1;
2411 0 : block_id_ele->latest_fec_idx = 0U;
2412 0 : block_id_ele->block_info = ag_block_id( 0UL, initial_block_id.uc );
2413 0 : bank->f.block_id = initial_block_id;
2414 0 : FD_TEST( fd_ag_block_id_map_ele_insert( ctx->ag_block_id_map, block_id_ele, ctx->block_id_arr ) );
2415 :
2416 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2417 0 : cost_tracker_snap( bank, slot_info );
2418 :
2419 0 : slot_info->identity_balance = fd_accdb_lamports( ctx->accdb, bank->accdb_fork_id, ctx->identity_pubkey->uc );
2420 :
2421 0 : FD_LOG_INFO(( "replay ready at slot %lu (%.3f s after snapshot done, %.3f s since boot)",
2422 0 : 0UL, 0.0, (double)(fd_log_wallclock()-ctx->boot_timestamp_nanos)/1e9 ));
2423 :
2424 0 : fd_block_footer_t footer = { .bank_hash = bank->f.bank_hash }; /* dummy footer for genesis slot */
2425 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */, 0, 0, &footer );
2426 0 : publish_root_advanced( ctx, stem, bank );
2427 :
2428 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
2429 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2430 :
2431 0 : reset->bank_idx = bank->idx;
2432 0 : reset->timestamp = ctx->reset_timestamp_nanos;
2433 0 : reset->completed_slot = ctx->reset_slot;
2434 0 : reset->hashcnt_per_tick = bank->f.slot_params.hashes_per_tick;
2435 0 : reset->ticks_per_slot = bank->f.ticks_per_slot;
2436 0 : reset->tick_duration_ns = bank->f.slot_params.ns_per_slot_adjusted/reset->ticks_per_slot;
2437 :
2438 0 : fd_memcpy( reset->completed_cmr, &block_id_ele->latest_mr, sizeof(fd_hash_t) );
2439 0 : fd_memcpy( reset->completed_dmr, &block_id_ele->dmr, sizeof(fd_hash_t) );
2440 :
2441 0 : fd_blockhashes_t const * block_hash_queue = &bank->f.block_hash_queue;
2442 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
2443 0 : FD_TEST( last_hash );
2444 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
2445 :
2446 0 : reset->max_microblocks_in_slot = fd_poh_max_microblocks_per_slot( bank->f.ticks_per_slot, reset->hashcnt_per_tick );
2447 0 : reset->next_leader_slot = ctx->next_leader_slot;
2448 0 : reset->wfs_paused = !ctx->wfs_complete;
2449 :
2450 0 : publish_replay_out( ctx, stem, REPLAY_SIG_RESET, sizeof(fd_poh_reset_t) );
2451 0 : }
2452 0 : }
2453 :
2454 : static inline void
2455 0 : maybe_verify_cluster_type( fd_replay_tile_t * ctx ) {
2456 0 : if( FD_UNLIKELY( !ctx->has_cluster_type || !ctx->has_genesis_hash ) ) {
2457 0 : return;
2458 0 : }
2459 :
2460 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash->uc, hash_cstr );
2461 0 : ulong cluster = fd_genesis_cluster_identify( hash_cstr );
2462 : /* Map pyth-related clusters to unknown. */
2463 0 : switch( cluster ) {
2464 0 : case FD_CLUSTER_PYTHNET:
2465 0 : case FD_CLUSTER_PYTHTEST:
2466 0 : cluster = FD_CLUSTER_UNKNOWN;
2467 0 : }
2468 :
2469 0 : if( FD_UNLIKELY( cluster!=ctx->cluster_type ) ) {
2470 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis hash of `%s` which means the cluster is %s "
2471 0 : "but the snapshot you loaded is for a different cluster %s. If you are trying to join the "
2472 0 : "%s cluster, you can delete the genesis.bin file and restart the node to download the correct "
2473 0 : "genesis file automatically.",
2474 0 : ctx->genesis_path,
2475 0 : hash_cstr,
2476 0 : fd_genesis_cluster_name( cluster ),
2477 0 : fd_genesis_cluster_name( ctx->cluster_type ),
2478 0 : fd_genesis_cluster_name( cluster ) ));
2479 0 : }
2480 0 : }
2481 :
2482 : static void
2483 : on_snapshot_message( fd_replay_tile_t * ctx,
2484 : fd_stem_context_t * stem,
2485 : ulong in_idx,
2486 : ulong chunk,
2487 0 : ulong sig ) {
2488 0 : ulong msg = fd_ssmsg_sig_message( sig );
2489 0 : if( FD_LIKELY( msg==FD_SSMSG_DONE ) ) {
2490 0 : long snapshot_done_nanos = fd_log_wallclock();
2491 :
2492 : /* An end of message notification indicates the snapshot is loaded.
2493 : Replay is able to start executing from this point onwards. */
2494 : /* TODO: replay should finish booting. Could make replay a
2495 : state machine and set the state here accordingly. */
2496 0 : ctx->is_booted = 1;
2497 :
2498 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
2499 0 : if( FD_UNLIKELY( !bank ) ) {
2500 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
2501 0 : }
2502 :
2503 0 : ulong snapshot_slot = bank->f.slot;
2504 :
2505 0 : fd_hash_t bank_hash = bank->f.bank_hash;
2506 0 : if( FD_UNLIKELY( ctx->wfs_enabled ) ) {
2507 : /* Classify boot against WFS (see fd_wfs.h). */
2508 0 : int wfs_mode = fd_wfs_mode( ctx->wait_for_supermajority_at_slot,
2509 0 : ctx->wfs_hash_is_zero,
2510 0 : (ulong)ctx->expected_shred_version,
2511 0 : snapshot_slot );
2512 0 : switch( wfs_mode ) {
2513 0 : case FD_WFS_MODE_MATCH: {
2514 0 : if( FD_UNLIKELY( memcmp( ctx->expected_bank_hash.uc, bank_hash.uc, sizeof(fd_hash_t) ) ) ) {
2515 0 : FD_BASE58_ENCODE_32_BYTES( ctx->expected_bank_hash.uc, expected_bank_hash_cstr );
2516 0 : FD_BASE58_ENCODE_32_BYTES( bank_hash.uc, actual_bank_hash_cstr );
2517 0 : FD_LOG_ERR(( "[consensus.wait_for_supermajority_with_bank_hash] expected_bank_hash=%s does not match snapshot slot"
2518 0 : "=%lu bank_hash=%s. If you are loading a snapshot from the network, check that the slot matches the "
2519 0 : "cluster restart slot.", expected_bank_hash_cstr, snapshot_slot, actual_bank_hash_cstr ));
2520 0 : }
2521 : /* Not gated on wfs_complete: gossip can signal the cluster
2522 : back before we boot, which says nothing about our vote. */
2523 0 : if( FD_UNLIKELY( ctx->wait_for_vote_to_start_leader==FD_BOOLAU_AUTO ) ) {
2524 0 : FD_LOG_NOTICE(( "auto-overriding wait_for_vote_to_start_leader to false "
2525 0 : "(WFS target slot %lu, boot slot %lu)",
2526 0 : ctx->wait_for_supermajority_at_slot, snapshot_slot ));
2527 0 : ctx->wait_for_vote_to_start_leader = FD_BOOLAU_FALSE;
2528 0 : }
2529 0 : if( FD_LIKELY( !ctx->wfs_complete ) ) {
2530 0 : FD_LOG_NOTICE(( "waiting for supermajority at snapshot slot %lu", snapshot_slot ));
2531 0 : }
2532 0 : break;
2533 0 : }
2534 0 : case FD_WFS_MODE_ERROR: {
2535 0 : FD_LOG_ERR(( "snapshot slot %lu is behind [consensus.wait_for_supermajority_at_slot] %lu "
2536 0 : "and no incremental snapshot bridged the gap. "
2537 0 : "Provide a snapshot at that slot, or clear it and "
2538 0 : "[consensus.wait_for_supermajority_with_bank_hash] to disable WFS.",
2539 0 : snapshot_slot, ctx->wait_for_supermajority_at_slot ));
2540 0 : break;
2541 0 : }
2542 0 : case FD_WFS_MODE_NOOP: {
2543 0 : FD_LOG_WARNING(( "snapshot slot %lu is ahead of [consensus.wait_for_supermajority_at_slot] %lu, "
2544 0 : "skipping wait for supermajority and bank hash verification",
2545 0 : snapshot_slot, ctx->wait_for_supermajority_at_slot ));
2546 0 : ctx->wfs_complete = 1;
2547 0 : ctx->wait_for_vote_to_start_leader = wfs_resolve_leader_gate_auto( ctx->wait_for_vote_to_start_leader );
2548 0 : break;
2549 0 : }
2550 0 : default: {
2551 : /* DISABLED and UNRESOLVED are unreachable under wfs_enabled. */
2552 0 : FD_LOG_CRIT(( "unexpected WFS mode %d (%s) at boot slot %lu",
2553 0 : wfs_mode, fd_wfs_mode_str( wfs_mode ), snapshot_slot ));
2554 0 : }
2555 0 : }
2556 :
2557 0 : if( FD_UNLIKELY( ctx->wait_for_vote_to_start_leader==FD_BOOLAU_AUTO ) ) {
2558 0 : FD_LOG_CRIT(( "wait_for_vote_to_start_leader unresolved (auto) after WFS boot classification" ));
2559 0 : }
2560 0 : }
2561 :
2562 : /* Manifest message must arrive before DONE */
2563 0 : if( FD_UNLIKELY( !ctx->has_expected_genesis_timestamp ) ) {
2564 0 : FD_LOG_CRIT(( "snapshot DONE received before manifest" ));
2565 0 : }
2566 :
2567 : /* FIXME: This is a hack when the block id of the snapshot slot
2568 : is not provided in the snapshot (Agave versions <4.1). A
2569 : possible solution is to get the block id of the snapshot slot
2570 : from repair. */
2571 0 : fd_hash_t manifest_block_id = ctx->has_manifest_block_id ? ctx->manifest_block_id : ctx->initial_block_id;
2572 :
2573 0 : FD_TEST( fd_sysvar_cache_restore( bank, ctx->accdb ) );
2574 : /* Agave zeroes manifest rent_params; reload from sysvar account */
2575 0 : FD_TEST( fd_sysvar_rent_read( ctx->accdb, bank->accdb_fork_id, &bank->f.rent ) );
2576 :
2577 0 : ctx->consensus_root = manifest_block_id;
2578 0 : ctx->consensus_root_slot = snapshot_slot;
2579 0 : ctx->finalized_block_id_lo = ag_block_id( snapshot_slot, manifest_block_id.uc );
2580 0 : ctx->finalized_block_id_hi = ctx->finalized_block_id_lo;
2581 0 : ctx->notified_root = manifest_block_id;
2582 0 : ctx->notified_root_slot = snapshot_slot;
2583 0 : ctx->notified_root_bank = bank;
2584 0 : ctx->published_root_slot = ctx->consensus_root_slot;
2585 0 : ctx->published_root_bank_idx = 0UL;
2586 0 : if( FD_UNLIKELY( ctx->snapmk.full_interval_blocks ) ) {
2587 0 : ulong interval = ctx->snapmk.full_interval_blocks;
2588 0 : ctx->snapmk.next_full_block_height = ((bank->f.block_height/interval)+1UL)*interval;
2589 0 : }
2590 0 : if( FD_UNLIKELY( ctx->snapmk.incremental_interval_blocks ) ) {
2591 0 : ulong interval = ctx->snapmk.incremental_interval_blocks;
2592 0 : ctx->snapmk.next_incremental_block_height = ((bank->f.block_height/interval)+1UL)*interval;
2593 0 : }
2594 :
2595 0 : ctx->reset_slot = snapshot_slot;
2596 0 : ctx->reset_dmr = manifest_block_id;
2597 0 : refresh_epoch_info( ctx, bank, snapshot_slot );
2598 0 : ctx->reset_timestamp_nanos = fd_clock_tile_now( ctx->clock );
2599 :
2600 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, snapshot_slot );
2601 0 : FD_TEST( bank->idx==0UL );
2602 :
2603 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
2604 0 : block_id_ele->latest_mr = manifest_block_id;
2605 0 : block_id_ele->dmr = manifest_block_id;
2606 0 : block_id_ele->slot = snapshot_slot;
2607 0 : block_id_ele->bank_seq = bank->bank_seq;
2608 0 : block_id_ele->block_id_seen = 1;
2609 0 : block_id_ele->latest_fec_idx = 0U;
2610 0 : bank->f.block_id = manifest_block_id;
2611 :
2612 0 : block_id_ele->block_info = ag_block_id( snapshot_slot, manifest_block_id.uc );
2613 0 : FD_TEST( fd_ag_block_id_map_ele_insert( ctx->ag_block_id_map, block_id_ele, ctx->block_id_arr ) );
2614 :
2615 : /* We call this after fd_runtime_read_genesis, which sets up the
2616 : slot_bank needed in blockstore_init. */
2617 0 : init_after_snapshot( ctx, stem );
2618 :
2619 0 : long now = fd_log_wallclock();
2620 0 : FD_LOG_INFO(( "replay ready at slot %lu (%.3f s after snapshot done, %.3f s since boot)",
2621 0 : snapshot_slot, (double)(now-snapshot_done_nanos)/1e9, (double)(now-ctx->boot_timestamp_nanos)/1e9 ));
2622 :
2623 0 : publish_boot_slot( ctx, stem, bank );
2624 :
2625 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
2626 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2627 :
2628 0 : reset->bank_idx = bank->idx;
2629 0 : reset->timestamp = ctx->reset_timestamp_nanos;
2630 0 : reset->completed_slot = ctx->reset_slot;
2631 0 : reset->hashcnt_per_tick = bank->f.slot_params.hashes_per_tick;
2632 0 : reset->ticks_per_slot = bank->f.ticks_per_slot;
2633 0 : reset->tick_duration_ns = bank->f.slot_params.ns_per_slot_adjusted/reset->ticks_per_slot;
2634 :
2635 0 : fd_memcpy( reset->completed_cmr, &block_id_ele->latest_mr, sizeof(fd_hash_t) );
2636 0 : fd_memcpy( reset->completed_dmr, &block_id_ele->dmr, sizeof(fd_hash_t) );
2637 :
2638 0 : fd_blockhashes_t const * block_hash_queue = &bank->f.block_hash_queue;
2639 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
2640 0 : FD_TEST( last_hash );
2641 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
2642 :
2643 0 : reset->max_microblocks_in_slot = fd_poh_max_microblocks_per_slot( bank->f.ticks_per_slot, reset->hashcnt_per_tick );
2644 0 : reset->next_leader_slot = ctx->next_leader_slot;
2645 0 : reset->wfs_paused = !ctx->wfs_complete;
2646 :
2647 0 : publish_replay_out( ctx, stem, REPLAY_SIG_RESET, sizeof(fd_poh_reset_t) );
2648 0 : }
2649 :
2650 0 : return;
2651 0 : }
2652 :
2653 0 : switch( msg ) {
2654 0 : case FD_SSMSG_MANIFEST_FULL:
2655 0 : case FD_SSMSG_MANIFEST_INCREMENTAL: {
2656 : /* We may either receive a full snapshot manifest or an
2657 : incremental snapshot manifest. Note that this external message
2658 : id is only used temporarily because replay cannot yet receive
2659 : the firedancer-internal snapshot manifest message. */
2660 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark ) )
2661 0 : FD_LOG_ERR(( "chunk %lu from in %d corrupt, not in range [%lu,%lu]", chunk, ctx->in_kind[ in_idx ], ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
2662 :
2663 : /* Malformed manifests are rejected recoverably by snapin via
2664 : fd_ssload_manifest_validate. If recover fails here, then the
2665 : bank is partially mutated, and we must abort. */
2666 0 : if( FD_UNLIKELY( fd_ssload_recover( fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ),
2667 0 : ctx->banks,
2668 0 : fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX ),
2669 0 : ctx->blockhash_seed ) ) ) {
2670 0 : FD_LOG_ERR(( "Snapshot manifest recovery failed, aborting." ));
2671 0 : }
2672 :
2673 0 : ctx->has_cluster_type = 1;
2674 0 : ctx->cluster_type = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX )->f.cluster_type;
2675 :
2676 0 : fd_snapshot_manifest_t const * manifest = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2677 : /* hard_fork_cnt already validated by fd_ssload_recover. */
2678 0 : ctx->hard_fork_cnt = manifest->hard_fork_cnt;
2679 0 : for( ulong i=0UL; i<manifest->hard_fork_cnt; i++ ) {
2680 0 : ctx->hard_forks[ i ] = manifest->hard_forks[ i ];
2681 0 : }
2682 0 : ctx->has_expected_genesis_timestamp = 1;
2683 0 : ctx->expected_genesis_timestamp = manifest->creation_time_seconds;
2684 0 : ctx->has_manifest_block_id = manifest->has_block_id;
2685 0 : if( manifest->has_block_id ) memcpy( ctx->manifest_block_id.uc, manifest->block_id, 32UL );
2686 0 : if( FD_UNLIKELY( msg==FD_SSMSG_MANIFEST_FULL ) ) {
2687 0 : ctx->snapmk.base_slot = manifest->slot;
2688 0 : }
2689 0 : break;
2690 0 : }
2691 0 : default: {
2692 0 : FD_LOG_ERR(( "Received unknown snapshot message with msg %lu", msg ));
2693 0 : return;
2694 0 : }
2695 0 : }
2696 :
2697 0 : return;
2698 0 : }
2699 :
2700 : static void
2701 : dispatch_task( fd_replay_tile_t * ctx,
2702 : fd_stem_context_t * stem,
2703 0 : fd_sched_task_t * task ) {
2704 :
2705 0 : switch( task->task_type ) {
2706 0 : case FD_SCHED_TT_TXN_EXEC: {
2707 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_exec->txn_idx );
2708 :
2709 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_exec->bank_idx );
2710 0 : FD_TEST( bank );
2711 :
2712 : /* Add the transaction to the block dumper if necessary. This
2713 : logic doesn't need to be fork-aware since it's only meant to
2714 : be used in backtest. */
2715 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
2716 0 : fd_dump_block_to_protobuf_collect_tx( ctx->block_dump_ctx, txn_p );
2717 0 : }
2718 :
2719 0 : bank->refcnt++;
2720 :
2721 0 : fd_cost_tracker_store_pin( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id );
2722 :
2723 0 : if( FD_UNLIKELY( !bank->first_transaction_scheduled_nanos ) ) bank->first_transaction_scheduled_nanos = fd_clock_tile_now( ctx->clock );
2724 :
2725 0 : FD_TEST( task->txn_exec->exec_idx<ctx->exec_cnt );
2726 0 : fd_replay_out_link_t * exec_out = &ctx->exec_out[ task->txn_exec->exec_idx ];
2727 0 : fd_execrp_txn_exec_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
2728 0 : fd_txn_p_copy( exec_msg->txn, txn_p );
2729 0 : exec_msg->bank_idx = task->txn_exec->bank_idx;
2730 0 : exec_msg->txn_idx = task->txn_exec->txn_idx;
2731 0 : memcpy( exec_msg->fec_merkle_root, ctx->block_id_arr[ task->txn_exec->bank_idx ].latest_mr.uc, 32UL );
2732 0 : exec_msg->index_in_slot = fd_sched_get_txn_info( ctx->sched, task->txn_exec->txn_idx )->index_in_slot;
2733 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) {
2734 0 : exec_msg->capture_txn_idx = ctx->capture_ctx->current_txn_idx++;
2735 0 : }
2736 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_TXN_EXEC<<32) | task->txn_exec->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
2737 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
2738 0 : break;
2739 0 : }
2740 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
2741 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_sigverify->txn_idx );
2742 :
2743 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_sigverify->bank_idx );
2744 0 : FD_TEST( bank );
2745 0 : bank->refcnt++;
2746 :
2747 0 : FD_TEST( task->txn_sigverify->exec_idx<ctx->exec_cnt );
2748 0 : fd_replay_out_link_t * exec_out = &ctx->exec_out[ task->txn_sigverify->exec_idx ];
2749 0 : fd_execrp_txn_sigverify_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
2750 0 : fd_txn_p_copy( exec_msg->txn, txn_p );
2751 0 : exec_msg->bank_idx = task->txn_sigverify->bank_idx;
2752 0 : exec_msg->txn_idx = task->txn_sigverify->txn_idx;
2753 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_TXN_SIGVERIFY<<32) | task->txn_sigverify->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
2754 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
2755 0 : break;
2756 0 : };
2757 0 : case FD_SCHED_TT_POH_HASH: {
2758 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->poh_hash->bank_idx );
2759 0 : FD_TEST( bank );
2760 0 : bank->refcnt++;
2761 :
2762 0 : FD_TEST( task->poh_hash->exec_idx<ctx->exec_cnt );
2763 0 : fd_replay_out_link_t * exec_out = &ctx->exec_out[ task->poh_hash->exec_idx ];
2764 0 : fd_execrp_poh_hash_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
2765 0 : exec_msg->bank_idx = task->poh_hash->bank_idx;
2766 0 : exec_msg->cnt = task->poh_hash->cnt;
2767 0 : exec_msg->hashcnt = task->poh_hash->hashcnt;
2768 0 : FD_STATIC_ASSERT( FD_SCHED_POH_PARA==FD_EXECRP_POH_PARA, poh batch width mismatch );
2769 0 : memcpy( exec_msg->hash, task->poh_hash->hash, task->poh_hash->cnt*sizeof(fd_hash_t) );
2770 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_POH_HASH<<32) | task->poh_hash->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
2771 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
2772 0 : break;
2773 0 : };
2774 0 : default: {
2775 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
2776 0 : }
2777 0 : }
2778 0 : }
2779 :
2780 : static void
2781 : mark_bank_dead( fd_replay_tile_t * ctx,
2782 : fd_stem_context_t * stem,
2783 : ulong bank_idx,
2784 : int dead_reason,
2785 : int abandoned_reason,
2786 0 : fd_block_footer_t const * footer ) {
2787 0 : ulong * dead_idxs = ctx->dead_idxs;
2788 0 : ulong dead_idxs_cnt = 0UL;
2789 0 : fd_banks_mark_bank_dead( ctx->banks, bank_idx, dead_idxs, &dead_idxs_cnt );
2790 :
2791 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank_idx ];
2792 0 : if( block_id_ele->block_id_seen ) publish_slot_dead( ctx, stem, block_id_ele->slot, &block_id_ele->dmr, footer );
2793 0 : else if( dead_reason!=FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD && !memcmp( block_id_ele->block_info.hash, ((fd_hash_t){0}).uc, sizeof(fd_hash_t) ) ) publish_slot_dead( ctx, stem, block_id_ele->slot, &(fd_hash_t){ 0 }, footer ); /* an incomplete turbine copy, keyed {slot,0}: rotor frees its eager blk */
2794 :
2795 : /* Report each newly dead bank now (dead_idxs excludes already-dead,
2796 : already-reported subtrees): the failing bank with its real reason and
2797 : descendants as parent_dead, or the whole lineage with the caller's
2798 : abandoned flavor. Exactly one of the two reasons is set.
2799 : A previously PRUNABLE (evicted) bank converted by this walk reports
2800 : here like any other: eviction emits no row, and the sched-drain
2801 : emission is gated on the state still being PRUNABLE. Blocks still
2802 : receiving FECs dedup the slot-completion report via dead_reported. */
2803 0 : int abandoned = abandoned_reason!=FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED;
2804 0 : for( ulong i=0UL; i<dead_idxs_cnt; i++ ) {
2805 0 : fd_block_id_ele_t * ele = &ctx->block_id_arr[ dead_idxs[ i ] ];
2806 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, dead_idxs[ i ] );
2807 0 : timing_slot_release( ctx, dead_idxs[ i ] );
2808 0 : int dr = abandoned ? FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD
2809 0 : : dead_idxs[ i ]==bank_idx ? dead_reason
2810 0 : : FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_PARENT_DEAD;
2811 0 : fd_hash_t * block_id = ele->block_id_seen ? &ele->dmr : &ele->latest_mr;
2812 0 : report_block_incomplete( ctx, ele->slot, block_id, bank, dr, abandoned_reason );
2813 0 : }
2814 0 : }
2815 :
2816 : static int
2817 : try_replay( fd_replay_tile_t * ctx,
2818 0 : fd_stem_context_t * stem ) {
2819 :
2820 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return 0;
2821 :
2822 : /* Hold off executing until the computed shred version is known, so
2823 : footer certs verify under it. */
2824 0 : if( FD_UNLIKELY( !ctx->shred_version ) ) return 0;
2825 :
2826 : /* Starting a block attaches an accdb fork, which would spin on a
2827 : pending root. Nothing of the block dispatches before its start. */
2828 0 : if( FD_UNLIKELY( fd_sched_block_start_pending( ctx->sched ) && fd_accdb_cmd_pending( ctx->accdb ) ) ) return 0;
2829 :
2830 0 : int charge_busy = 0;
2831 0 : fd_sched_task_t task[ 1 ];
2832 0 : if( FD_UNLIKELY( !fd_sched_task_next_ready( ctx->sched, task ) ) ) {
2833 0 : return charge_busy; /* Nothing to execute or do. */
2834 0 : }
2835 :
2836 0 : charge_busy = 1;
2837 :
2838 0 : switch( task->task_type ) {
2839 0 : case FD_SCHED_TT_BLOCK_START: {
2840 0 : replay_block_start( ctx, task->block_start->bank_idx, task->block_start->parent_bank_idx, task->block_start->slot );
2841 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX, NULL );
2842 0 : break;
2843 0 : }
2844 0 : case FD_SCHED_TT_BLOCK_END: {
2845 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->block_end->bank_idx );
2846 0 : int dead = 0;
2847 0 : if( FD_LIKELY( bank->state==FD_BANK_STATE_REPLAYABLE ) ) dead = replay_block_finalize( ctx, stem, bank );
2848 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX, NULL );
2849 0 : if( FD_UNLIKELY( dead ) ) fd_sched_block_abandon( ctx->sched, bank->idx, FD_SCHED_ABANDON_INVALID );
2850 0 : break;
2851 0 : }
2852 0 : case FD_SCHED_TT_TXN_EXEC:
2853 0 : case FD_SCHED_TT_TXN_SIGVERIFY:
2854 0 : case FD_SCHED_TT_POH_HASH: {
2855 : /* Common case: we have a transaction we need to execute. */
2856 0 : dispatch_task( ctx, stem, task );
2857 0 : break;
2858 0 : }
2859 0 : case FD_SCHED_TT_MARK_DEAD: {
2860 0 : int dr = sched_block_dead_reason_to_event( ctx, task->mark_dead->bank_idx );
2861 0 : int ar = dr==FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD ? FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_PRUNED
2862 0 : : FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED;
2863 0 : mark_bank_dead( ctx, stem, task->mark_dead->bank_idx, dr, ar, NULL );
2864 0 : break;
2865 0 : }
2866 0 : default: {
2867 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
2868 0 : }
2869 0 : }
2870 :
2871 0 : return charge_busy;
2872 0 : }
2873 :
2874 0 : #define PROCESS_FEC_WAIT 0 /* Put back on dcache for later retry */
2875 0 : #define PROCESS_FEC_OK 1 /* Process FEC set */
2876 0 : #define PROCESS_FEC_DROP 2 /* Drop FEC set and notify rotor to start redelivery */
2877 0 : #define PROCESS_FEC_SKIP 3 /* Good to skip this FEC set, drop from dcache */
2878 :
2879 : /* live_block_id_ele returns ele if it is backed by a live bank (same
2880 : bank_seq, not PRUNABLE), else NULL. Map entries outlive their banks
2881 : through eviction. TODO if we remove block_id_eles with banks, we
2882 : can get rid of this */
2883 : static fd_block_id_ele_t *
2884 : live_block_id_ele( fd_replay_tile_t * ctx,
2885 0 : fd_block_id_ele_t * ele ) {
2886 0 : if( FD_UNLIKELY( !ele ) ) return NULL;
2887 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, fd_block_id_ele_get_idx( ctx->block_id_arr, ele ) );
2888 0 : if( FD_UNLIKELY( !bank || bank->bank_seq!=ele->bank_seq || bank->state==FD_BANK_STATE_PRUNABLE ) ) return NULL;
2889 0 : return ele;
2890 0 : }
2891 :
2892 : static int
2893 : can_process_rotor_fec( fd_replay_tile_t * ctx,
2894 : fd_rotor_replay_fec_t * fec,
2895 0 : int * evict_banks_out ) {
2896 : /* We can process a FEC set if a few conditions are met:
2897 : - sched has capacity
2898 : - banks has capacity. Evict if we don't (see below)
2899 : Our leader FECs bind to the existing leader bank and never enter
2900 : sched, so neither capacity gate applies to them. One that is not
2901 : for the current leader slot has no bank to bind to. */
2902 :
2903 0 : if( FD_UNLIKELY( fec->is_leader && ( !ctx->leader_bank || ctx->leader_bank->f.slot!=fec->slot ) ) ) return PROCESS_FEC_SKIP;
2904 :
2905 0 : if( FD_UNLIKELY( !fec->is_leader && fd_sched_can_ingest_cnt( ctx->sched )==0UL ) ) {
2906 0 : FD_TEST( !fd_sched_is_drained( ctx->sched ) );
2907 0 : ctx->metrics.sched_full++;
2908 0 : return PROCESS_FEC_WAIT;
2909 0 : }
2910 :
2911 : /* Recovery redelivery re-publishes the entire ancestry path from the
2912 : rotor root, but most of those blocks are already replayed. A
2913 : fully replayed block is always keyed by {slot, block_id_real}; a
2914 : turbine block still in flight is keyed by {slot, 0}. Rotor fills
2915 : in block_id on redelivered FECs whenever it knows it, so try the id
2916 : key first and fall back to the slot key for a turbine version.
2917 :
2918 : A turbine FEC only counts a {slot, block_id} hit if that block is
2919 : complete. An incomplete one is the repaired copy of the same
2920 : block, a different stream that may be ahead of or behind the
2921 : turbine copy: rotor merges the two copies into the turbine one
2922 : without changing its known_id, so the turbine copy keeps building
2923 : its own {slot, 0} bank, and the repaired one is abandoned once the
2924 : turbine bank completes and takes its {slot, block_id} key. */
2925 0 : {
2926 0 : fd_block_id_ele_t * self = NULL;
2927 0 : if( FD_LIKELY( !fd_hash_check_zero( &fec->block_id ) ) ) {
2928 0 : ag_block_id_t self_key = ag_block_id( fec->slot, fec->block_id.uc );
2929 0 : self = live_block_id_ele( ctx, fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &self_key, NULL, ctx->block_id_arr ) );
2930 0 : if( FD_UNLIKELY( self && !self->block_id_seen && !fec->known_id ) ) self = NULL;
2931 0 : }
2932 0 : if( FD_LIKELY( !self && !fec->known_id ) ) {
2933 0 : ag_block_id_t self_key = { .slot = fec->slot };
2934 0 : self = live_block_id_ele( ctx, fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &self_key, NULL, ctx->block_id_arr ) );
2935 0 : }
2936 0 : if( FD_UNLIKELY( self && ( self->block_id_seen || self->latest_fec_idx>=fec->fec_set_idx ) ) ) return PROCESS_FEC_SKIP;
2937 0 : }
2938 :
2939 : /* Rooted past: the block is rooted or on a dead fork, so no
2940 : redelivery can help. A DROP would ask for another, and during
2941 : catch-up that one also lands behind the backlog, rooted past. */
2942 :
2943 0 : if( FD_UNLIKELY( ctx->published_root_slot!=ULONG_MAX &&
2944 0 : ( fec->slot<=ctx->published_root_slot ||
2945 0 : fec->parent_slot<ctx->published_root_slot ) ) ) {
2946 0 : ctx->metrics.fec_rooted_past++;
2947 0 : return PROCESS_FEC_SKIP;
2948 0 : }
2949 :
2950 0 : ulong parent_bank_idx = UINT_MAX;
2951 0 : fd_block_id_ele_t * parent = NULL;
2952 0 : if( FD_UNLIKELY( fec->fec_set_idx==0 ) ) {
2953 0 : ag_block_id_t parent_key = ag_block_id( fec->parent_slot, fec->parent_block_id.uc );
2954 0 : parent = fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &parent_key, NULL, ctx->block_id_arr );
2955 0 : if( FD_UNLIKELY( !parent ) ) {
2956 0 : FD_BASE58_ENCODE_32_BYTES( fec->parent_block_id.uc, parent_key_b58 );
2957 0 : FD_LOG_INFO(( "parent bank not found for slot %lu fec set idx %u, parent slot %lu parent block_id %s", fec->slot, fec->fec_set_idx, fec->parent_slot, parent_key_b58 ));
2958 :
2959 : /* We cannot tell here why the parent is missing -- dropped on a
2960 : dead lineage, evicted with its tracking slot reused, or rooted
2961 : past -- and the latter two are routine and recoverable, so do
2962 : not emit events; they would be emitted in bulk during the
2963 : redelivery that recovers them. */
2964 0 : ctx->metrics.parent_unavailable++;
2965 0 : return PROCESS_FEC_DROP;
2966 0 : }
2967 0 : parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent );
2968 0 : } else {
2969 0 : ag_block_id_t key = { .slot = fec->slot };
2970 0 : if( FD_LIKELY( fec->known_id ) ) key = ag_block_id( fec->slot, fec->block_id.uc );
2971 :
2972 0 : parent = fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &key, NULL, ctx->block_id_arr );
2973 0 : if( FD_UNLIKELY( !parent ) ) {
2974 0 : FD_BASE58_ENCODE_32_BYTES( fec->block_id.uc, block_id_b58 );
2975 0 : FD_LOG_INFO(( "parent bank not found for slot %lu fec set idx %u slot_bid %s. parent slot %lu", fec->slot, fec->fec_set_idx, block_id_b58, fec->parent_slot ));
2976 0 : ctx->metrics.parent_unavailable++;
2977 0 : return PROCESS_FEC_DROP; // either pruned or bank evicted
2978 0 : }
2979 0 : parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent );
2980 0 : }
2981 :
2982 0 : ctx->metrics.reasm_latest_slot = fec->slot;
2983 0 : ctx->metrics.reasm_latest_fec_idx = fec->fec_set_idx;
2984 :
2985 : /* If the FEC we are building off of is for a prunable bank, we must
2986 : wait to process the FEC until the bank has been evicted. */
2987 0 : fd_bank_t * parent_fec_bank = parent_bank_idx==UINT_MAX ? NULL : fd_banks_bank_query( ctx->banks, parent_bank_idx );
2988 0 : if( FD_UNLIKELY( parent_fec_bank && parent_fec_bank->bank_seq==parent->bank_seq && parent_fec_bank->state==FD_BANK_STATE_PRUNABLE ) ) {
2989 0 : FD_LOG_DEBUG(( "waiting to process FEC set (slot=%lu, fec_set_idx=%u) because parent bank is being pruned", fec->slot, fec->fec_set_idx ));
2990 0 : return PROCESS_FEC_WAIT;
2991 0 : }
2992 :
2993 0 : if( FD_UNLIKELY( ctx->is_leader &&
2994 0 : fec->fec_set_idx==0U &&
2995 0 : parent_bank_idx==ctx->leader_bank->idx &&
2996 0 : parent->bank_seq==ctx->leader_bank->bank_seq ) ) {
2997 : /* This guards against a rare race where we receive the FEC set for
2998 : the slot right after our leader rotation before we freeze the
2999 : bank for the last slot in our leader rotation. Leader slot
3000 : freezing happens only after if we've received the final PoH hash
3001 : from the poh tile as well as the final FEC set for the leader
3002 : slot. So the race happens when FEC sets are delivered and
3003 : processed sooner than the PoH hash, aka when the
3004 : poh=>shred=>replay path for the block id beats the poh=>replay
3005 : path for the poh hash. To mitigate this race, we must block on
3006 : ingesting the FEC set for the ensuing slot before the leader
3007 : bank freezes, because that would violate ordering invariants in
3008 : banks and sched. Both the bank index and sequence must match
3009 : because bank indices are recycled after eviction. */
3010 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->idx ].block_id_seen );
3011 0 : FD_TEST( !ctx->recv_poh );
3012 0 : ctx->metrics.leader_bid_wait++;
3013 0 : return PROCESS_FEC_WAIT;
3014 0 : }
3015 :
3016 : /* Should we evict banks if there are no more free banks? The answer
3017 : is it depends. Eviction should only happen if we can make no
3018 : forward replay progress. This can only happen if:
3019 : 1. banks are full
3020 : 2. sched is drained: pending txns could complete a block and
3021 : eventually advance the root.
3022 : AND
3023 : 3. next reasm FEC needs a new bank. A fec that chains off of a
3024 : bank that is already allocated can be processed. A FEC can
3025 : require a new bank in three ways:
3026 : - fec_set_idx==0: we don't have any free banks to provision a
3027 : new bank for this FEC.
3028 : - equivocation: a FEC may be in the middle of a block, but if
3029 : it's the first equivocating FEC detected, we need to allocate
3030 : a new bank for the version of the block.
3031 : - backfill: the parent FEC's bank was never created or has been
3032 : evicted and must be reconstructed.
3033 :
3034 :
3035 : - Without reasm there is no backfill: if the parent bank was
3036 : evicted or its bank idx was reused, this FEC can't be
3037 : processed, so drop it and notify rotor to redeliver.
3038 : - Rotor equivocating slots are guaranteed to be delivered starting
3039 : from FEC 0, so even equivocations mid-slot are delivered from
3040 : fec 0 */
3041 :
3042 0 : int invalid_parent = !parent_fec_bank || parent_fec_bank->bank_seq!=parent->bank_seq;
3043 0 : if( FD_UNLIKELY( invalid_parent ) ) {
3044 0 : FD_LOG_INFO(( "parent bank evicted for slot %lu fec set idx %u, parent slot %lu", fec->slot, fec->fec_set_idx, fec->parent_slot ));
3045 0 : ctx->metrics.parent_unavailable++;
3046 0 : return PROCESS_FEC_DROP;
3047 0 : } else if( FD_UNLIKELY( fec->fec_set_idx!=0U && parent->latest_fec_idx!=fec->fec_set_idx - FD_FEC_SHRED_CNT ) ) {
3048 : /* Similar to the very first condition in can_process_rotor_fec,
3049 : but this would hit if we were halfway through replaying a slot
3050 : and then requested redelivery from rotor. Then we can skip
3051 : replaying the first half of the slot. */
3052 0 : FD_LOG_INFO(( "fec redelivered for slot %lu fec set idx %u, parent slot %lu. bank_idx %lu, latest_fec_idx %u", fec->slot, fec->fec_set_idx, fec->parent_slot, fd_block_id_ele_get_idx( ctx->block_id_arr, parent ), parent->latest_fec_idx ));
3053 0 : return PROCESS_FEC_SKIP; // context for slot exists, but this is an earlier or non-contiguous FEC. Safe to skip.
3054 0 : }
3055 :
3056 0 : if( FD_UNLIKELY( !fd_banks_can_start_bank( ctx->banks ) ) ) {
3057 0 : int is_new_block = fec->fec_set_idx==0U && !fec->is_leader;
3058 0 : if( FD_UNLIKELY( is_new_block ) ) {
3059 0 : ctx->metrics.banks_full++;
3060 0 : if( FD_UNLIKELY( fd_sched_is_drained( ctx->sched ) && !ctx->snapmk.active ) ) *evict_banks_out = 1; /* see can_process_fec */
3061 0 : return PROCESS_FEC_WAIT;
3062 0 : }
3063 0 : }
3064 :
3065 : /* Otherwise, banks may not be full, so we can always create a new
3066 : bank if needed. Or, if banks are full, the current fec set's
3067 : ancestor (idx 0) already created a bank for this slot. */
3068 0 : return PROCESS_FEC_OK;
3069 0 : }
3070 :
3071 : static int
3072 : try_notify_consensus_root( fd_replay_tile_t * ctx,
3073 0 : fd_stem_context_t * stem ) {
3074 :
3075 : /* Other tiles cannot prune while we are leader - replay may still be
3076 : waiting for the remaining FEC sets of our own slots to be
3077 : delivered, and if the upstream tiles prune too early, they may fail
3078 : to forward our own leader FECs. Then we would be permanently stuck
3079 : in the leader state. */
3080 0 : if( FD_UNLIKELY( ctx->is_leader ) ) return 0;
3081 :
3082 0 : if( FD_LIKELY( ctx->notified_root_slot==ctx->consensus_root_slot &&
3083 0 : fd_hash_eq( &ctx->notified_root, &ctx->consensus_root ) ) ) return 0;
3084 :
3085 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_ele_query( ctx, &ctx->consensus_root, ctx->consensus_root_slot );
3086 0 : if( FD_UNLIKELY( !block_id_ele ) ) return 0;
3087 :
3088 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele ) );
3089 0 : if( FD_UNLIKELY( !bank ||
3090 0 : bank->bank_seq!=block_id_ele->bank_seq ||
3091 0 : !fd_hash_eq( &bank->f.block_id, &ctx->consensus_root ) ||
3092 0 : bank->state==FD_BANK_STATE_PRUNABLE ) ) return 0;
3093 :
3094 0 : fd_sched_root_notify( ctx->sched, bank->idx );
3095 0 : publish_root_advanced( ctx, stem, bank );
3096 :
3097 0 : ctx->notified_root = ctx->consensus_root;
3098 0 : ctx->notified_root_slot = ctx->consensus_root_slot;
3099 0 : ctx->notified_root_bank = bank;
3100 0 : return 1;
3101 0 : }
3102 :
3103 : /* Periodic snapshots use block height, matching Agave. Manually
3104 : scheduled snapshots remain slot based. */
3105 :
3106 : static int
3107 : snapshot_due_for_root( fd_replay_tile_t * ctx,
3108 : ulong published_root_block_height,
3109 : ulong advanceable_root_slot,
3110 : ulong advanceable_root_block_height,
3111 : ulong consensus_root_block_height,
3112 0 : int * out_incremental ) {
3113 0 : *out_incremental = 0;
3114 0 : if( FD_LIKELY( !ctx->snapmk.supported || ctx->snapmk.active ) ) return 0;
3115 :
3116 0 : int caught_up = ctx->caught_up; /* suspend periodic snaps until caught up */
3117 0 : ulong full_target = ULONG_MAX;
3118 0 : ulong interval = ctx->snapmk.full_interval_blocks;
3119 0 : if( FD_UNLIKELY( interval && caught_up ) ) {
3120 0 : if( FD_UNLIKELY( ctx->snapmk.next_full_block_height==ULONG_MAX ) ) {
3121 0 : ctx->snapmk.next_full_block_height = ((published_root_block_height/interval)+1UL)*interval;
3122 0 : }
3123 :
3124 : /* If snapshot production fell behind by more than one interval,
3125 : skip ahead to the latest due interval. */
3126 0 : full_target = fd_ulong_max( ctx->snapmk.next_full_block_height,
3127 0 : (consensus_root_block_height/interval)*interval );
3128 0 : }
3129 :
3130 : /* Manual slot scheduling always requests a full snapshot. */
3131 0 : if( FD_UNLIKELY( advanceable_root_slot>=ctx->snapmk.scheduled_at_slot ) ) return 1;
3132 :
3133 : /* An incremental snapshot is only possible once a full snapshot
3134 : exists to serve as its base. */
3135 0 : ulong incremental_target = ULONG_MAX;
3136 0 : ulong incr_interval = ctx->snapmk.incremental_interval_blocks;
3137 0 : if( FD_UNLIKELY( incr_interval && caught_up && ctx->snapmk.base_slot!=ULONG_MAX ) ) {
3138 0 : if( FD_UNLIKELY( ctx->snapmk.next_incremental_block_height==ULONG_MAX ) ) {
3139 0 : ctx->snapmk.next_incremental_block_height = ((published_root_block_height/incr_interval)+1UL)*incr_interval;
3140 0 : }
3141 0 : incremental_target = fd_ulong_max( ctx->snapmk.next_incremental_block_height,
3142 0 : (consensus_root_block_height/incr_interval)*incr_interval );
3143 0 : }
3144 :
3145 : /* A full snapshot due at the same block height supersedes the
3146 : incremental snapshot. */
3147 0 : ulong target = full_target;
3148 0 : int incremental = incremental_target<full_target;
3149 0 : if( FD_UNLIKELY( incremental ) ) target = incremental_target;
3150 0 : if( FD_LIKELY( advanceable_root_block_height<target ) ) return 0;
3151 0 : *out_incremental = incremental;
3152 0 : return 1;
3153 0 : }
3154 :
3155 : static void
3156 : snapmk_start( fd_replay_tile_t * ctx,
3157 : fd_stem_context_t * stem,
3158 : int incremental );
3159 :
3160 : static int
3161 : try_advance_published_root( fd_replay_tile_t * ctx,
3162 0 : fd_stem_context_t * stem ) {
3163 :
3164 0 : if( FD_LIKELY( ctx->published_root_slot==ctx->consensus_root_slot ) ) return 0;
3165 :
3166 : /* accdb pauses advance_root while producing a snapshot, so submitting
3167 : one would stall the next wait_cmd until the snapshot completes. */
3168 0 : if( FD_UNLIKELY( ctx->snapmk.active ) ) return 0;
3169 :
3170 : /* Don't spin in advance_root on the previous root, replay instead and
3171 : retry next iteration. */
3172 0 : if( FD_UNLIKELY( fd_accdb_cmd_pending( ctx->accdb ) ) ) return 0;
3173 :
3174 : /* Let a waiting block start take the idle accdb first, else it would
3175 : wait out the whole drain. */
3176 0 : if( FD_UNLIKELY( fd_sched_block_start_pending( ctx->sched ) ) ) return 0;
3177 :
3178 : /* If the new root is not available because the bank is/has been
3179 : evicted, we can't advance the root. Try again later. */
3180 :
3181 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_ele_query( ctx, &ctx->consensus_root, ctx->consensus_root_slot );
3182 0 : if( FD_UNLIKELY( !block_id_ele ) ) return 0;
3183 0 : fd_bank_t * target_bank = fd_banks_bank_query( ctx->banks, fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele ) );
3184 0 : if( FD_UNLIKELY( !target_bank ||
3185 0 : target_bank->bank_seq!=block_id_ele->bank_seq ||
3186 0 : !fd_hash_eq( &target_bank->f.block_id, &ctx->consensus_root ) ||
3187 0 : target_bank->state==FD_BANK_STATE_PRUNABLE ) ) {
3188 0 : return 0;
3189 0 : }
3190 :
3191 0 : fd_bank_t * published_root_bank = fd_banks_bank_query( ctx->banks, ctx->published_root_bank_idx );
3192 0 : FD_TEST( published_root_bank );
3193 0 : ulong published_root_block_height = published_root_bank->f.block_height;
3194 0 : ulong consensus_root_block_height = target_bank->f.block_height;
3195 :
3196 : /* If the identity vote has been seen on a bank that should be rooted,
3197 : then we are now ready to produce blocks. */
3198 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
3199 0 : if( target_bank->f.identity_vote_idx==ctx->identity_idx ) ctx->identity_vote_rooted = 1;
3200 0 : }
3201 :
3202 0 : ulong advanceable_root_idx = ULONG_MAX;
3203 0 : if( FD_UNLIKELY( !fd_banks_advance_root_prepare( ctx->banks, target_bank->idx, &advanceable_root_idx ) ) ) {
3204 0 : ctx->metrics.storage_root_behind++;
3205 0 : return 0;
3206 0 : }
3207 :
3208 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, advanceable_root_idx );
3209 0 : FD_TEST( bank );
3210 :
3211 0 : if( FD_UNLIKELY( advanceable_root_idx>=ctx->block_id_len ) ) {
3212 0 : FD_LOG_CRIT(( "invariant violation: advanceable root ele out of bounds [0, %lu) index %lu", ctx->block_id_len, advanceable_root_idx ));
3213 0 : }
3214 : /* Telemetry: banks still mid-replay (started but never completed) on
3215 : forks not descending from the new root are about to be cancelled by
3216 : fd_banks_advance_root without ever emitting a block_completed row.
3217 : Emit them as abandoned with reason `pruned`: the block was not
3218 : invalid, it lost the fork race. Every live bank is linked in the
3219 : bank tree under the current root, so one preorder pass that skips
3220 : the new root's subtree visits exactly the banks about to go. */
3221 0 : if( FD_UNLIKELY( fd_event_tl ) ) {
3222 0 : ulong root_idx = fd_banks_root( ctx->banks )->idx;
3223 0 : ulong s = root_idx;
3224 0 : fd_bank_t * b = fd_banks_root( ctx->banks );
3225 0 : for(;;) {
3226 0 : if( FD_LIKELY( s!=advanceable_root_idx ) ) {
3227 0 : if( FD_UNLIKELY( b->state==FD_BANK_STATE_INIT || b->state==FD_BANK_STATE_REPLAYABLE ) ) {
3228 0 : fd_block_id_ele_t * ele = &ctx->block_id_arr[ s ];
3229 0 : report_block_incomplete( ctx, ele->slot, ele->block_id_seen ? &ele->dmr : &ele->latest_mr, b, FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_PRUNED );
3230 0 : }
3231 0 : if( FD_LIKELY( b->child_idx!=ULONG_MAX ) ) { s = b->child_idx; b = fd_banks_bank_query( ctx->banks, s ); continue; }
3232 0 : }
3233 0 : while( s!=root_idx && b->sibling_idx==ULONG_MAX ) { s = b->parent_idx; b = fd_banks_bank_query( ctx->banks, s ); }
3234 0 : if( s==root_idx ) break;
3235 0 : s = b->sibling_idx; b = fd_banks_bank_query( ctx->banks, s );
3236 0 : }
3237 0 : }
3238 :
3239 0 : ulong advanceable_root_slot = bank->f.slot;
3240 0 : int root_epoch_changed = bank->f.epoch!=published_root_bank->f.epoch;
3241 0 : fd_txncache_advance_root( ctx->txncache, bank->txncache_fork_id );
3242 0 : fd_progcache_advance_root( ctx->progcache, bank->progcache_fork_id );
3243 0 : fd_accdb_advance_root( ctx->accdb, bank->accdb_fork_id );
3244 0 : fd_sched_advance_root( ctx->sched, advanceable_root_idx );
3245 0 : fd_banks_advance_root( ctx->banks, advanceable_root_idx );
3246 0 : if( FD_UNLIKELY( root_epoch_changed ) ) refresh_vote_account_staked( ctx, bank );
3247 :
3248 0 : for( ulong b=0UL; b<ctx->max_live_slots; b++ ) {
3249 0 : if( FD_UNLIKELY( ctx->timing_slot_of_bank[ b ]!=fd_timing_slot_pool_idx_null( ctx->timing_slot_pool ) && !fd_banks_bank_query( ctx->banks, b ) ) ) timing_slot_release( ctx, b );
3250 0 : }
3251 :
3252 0 : int snap_incremental;
3253 0 : int snap_due = snapshot_due_for_root( ctx,
3254 0 : published_root_block_height,
3255 0 : advanceable_root_slot,
3256 0 : bank->f.block_height,
3257 0 : consensus_root_block_height,
3258 0 : &snap_incremental );
3259 :
3260 0 : ctx->published_root_slot = advanceable_root_slot;
3261 0 : ctx->published_root_bank_idx = advanceable_root_idx;
3262 :
3263 0 : if( FD_UNLIKELY( snap_due ) ) {
3264 0 : snapmk_start( ctx, stem, snap_incremental );
3265 0 : if( FD_UNLIKELY( !snap_incremental ) ) {
3266 0 : if( FD_UNLIKELY( ctx->snapmk.full_interval_blocks &&
3267 0 : bank->f.block_height>=ctx->snapmk.next_full_block_height ) ) {
3268 0 : ulong interval = ctx->snapmk.full_interval_blocks;
3269 0 : ctx->snapmk.next_full_block_height = ((bank->f.block_height/interval)+1UL)*interval;
3270 0 : }
3271 0 : if( FD_UNLIKELY( advanceable_root_slot>=ctx->snapmk.scheduled_at_slot ) ) {
3272 0 : ctx->snapmk.scheduled_at_slot = ULONG_MAX;
3273 0 : }
3274 0 : }
3275 0 : if( FD_UNLIKELY( ctx->snapmk.incremental_interval_blocks ) ) {
3276 0 : ulong interval = ctx->snapmk.incremental_interval_blocks;
3277 0 : ctx->snapmk.next_incremental_block_height = ((bank->f.block_height/interval)+1UL)*interval;
3278 0 : }
3279 0 : }
3280 :
3281 0 : return 1;
3282 0 : }
3283 :
3284 : static int
3285 0 : try_prune_sched( fd_replay_tile_t * ctx ) {
3286 0 : ulong bank_idx;
3287 0 : int pruned = 0;
3288 0 : while( (bank_idx=fd_sched_pruned_block_next( ctx->sched ) )!=ULONG_MAX ) {
3289 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
3290 0 : FD_TEST( bank );
3291 : /* An evicted bank drains here with its verdict final: every
3292 : in-flight task has resolved, so still PRUNABLE means none ruled
3293 : it invalid (that would have converted it to DEAD and reported a
3294 : dead row), and this is the block's only row. A frozen victim
3295 : already reported a completed row. */
3296 0 : if( FD_UNLIKELY( bank->state==FD_BANK_STATE_PRUNABLE && !bank->block_completed_nanos ) ) {
3297 0 : fd_block_id_ele_t * ele = &ctx->block_id_arr[ bank_idx ];
3298 0 : report_block_incomplete( ctx, ele->slot, ele->block_id_seen ? &ele->dmr : &ele->latest_mr, bank,
3299 0 : FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD,
3300 0 : FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_EVICTED );
3301 0 : }
3302 0 : bank->refcnt--;
3303 0 : FD_LOG_DEBUG(( "bank (idx=%lu) refcnt decremented to %lu for sched", bank->idx, bank->refcnt ));
3304 0 : pruned = 1;
3305 0 : }
3306 0 : return pruned;
3307 0 : }
3308 :
3309 : static int
3310 0 : try_prune_bank( fd_replay_tile_t * ctx ) {
3311 : /* A cancellation purges accdb, which would spin on a pending root.
3312 : Check before popping the bank so the cancel info isn't lost. */
3313 0 : if( FD_UNLIKELY( fd_accdb_cmd_pending( ctx->accdb ) ) ) return 0;
3314 :
3315 0 : fd_banks_prune_cancel_info_t cancel_info[ 1 ];
3316 :
3317 0 : int pruned = fd_banks_prune_one_bank( ctx->banks, cancel_info );
3318 0 : switch( pruned ) {
3319 0 : case 2: { /* pruning bank + cancellation is needed */
3320 0 : fd_txncache_cancel_fork( ctx->txncache, cancel_info->txncache_fork_id );
3321 0 : fd_progcache_cancel_fork( ctx->progcache, cancel_info->progcache_fork_id );
3322 0 : fd_accdb_purge( ctx->accdb, cancel_info->accdb_fork_id );
3323 0 : __attribute__((fallthrough));
3324 0 : }
3325 0 : case 1: { /* pruning bank + no cancellation is needed */
3326 : /* A sched block exists, and can be marked dead, for a bank as
3327 : soon as its first FEC has been ingested, which can happen
3328 : before the bank ever set itself up for actual execution (e.g. a
3329 : block that parses as bad on its very first FEC). So always
3330 : instruct sched to prune the block whenever banks prunes the
3331 : bank. The txncache/progcache/accdb forks, on the other hand,
3332 : are only created once the bank started actual execution. */
3333 0 : fd_sched_cancel( ctx->sched, cancel_info->bank_idx );
3334 0 : timing_slot_release( ctx, cancel_info->bank_idx );
3335 0 : return 1;
3336 0 : }
3337 0 : case 0: /* no bank to prune */
3338 0 : return 0;
3339 0 : default:
3340 0 : FD_LOG_ERR(( "unreachable" ));
3341 0 : }
3342 0 : }
3343 :
3344 : static void
3345 : after_credit( fd_replay_tile_t * ctx,
3346 : fd_stem_context_t * stem,
3347 : int * opt_poll_in,
3348 0 : int * charge_busy ) {
3349 0 : if( FD_UNLIKELY( ctx->halt_replay && !ctx->is_leader ) ) return;
3350 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->wfs_complete ) ) return;
3351 :
3352 : /* The overall priority for the replay tile in order is:
3353 : 1. Make sure replay has room to progress:
3354 : a. evicting pending FECs from the reassembler
3355 : b. queueing up evictable banks for pruning if needed
3356 : c. clearing any pending bank eviction victims.
3357 : 2. Drain outstanding bank references from the scheduler. This
3358 : happens after a block gets completed or a fork gets pruned.
3359 : 3. Notify sched and bank consumers of a new consensus root, then
3360 : advance the storage root once old references drain.
3361 : 4. Replay. If there is work to do for replay, do it. This is
3362 : more important than ingesting more FEC sets.
3363 : 5. If replay has nothing to do, ingest more FEC sets.
3364 : WARNING: The ordering here is VERY load bearing and it should not
3365 : be changed without extreme caution. */
3366 :
3367 0 : if( FD_UNLIKELY( try_prune_sched( ctx ) ) ) {
3368 0 : *charge_busy = 1;
3369 0 : *opt_poll_in = 0;
3370 0 : return;
3371 0 : }
3372 :
3373 0 : if( FD_UNLIKELY( try_notify_consensus_root( ctx, stem ) ) ) {
3374 0 : *charge_busy = 1;
3375 0 : *opt_poll_in = 0;
3376 0 : return;
3377 0 : }
3378 :
3379 0 : if( FD_UNLIKELY( try_prune_bank( ctx ) ) ) {
3380 0 : *charge_busy = 1;
3381 0 : *opt_poll_in = 0;
3382 0 : return;
3383 0 : }
3384 :
3385 0 : if( FD_UNLIKELY( try_become_leader_ag( ctx, stem, FD_METRICS_SAMPLE_REASON_LEADER_START, 0UL ) ) ) {
3386 0 : *charge_busy = 1;
3387 0 : *opt_poll_in = 0;
3388 0 : return;
3389 0 : }
3390 :
3391 0 : if( FD_UNLIKELY( try_fini_leader( ctx, stem ) ) ) {
3392 0 : *charge_busy = 1;
3393 0 : *opt_poll_in = 0;
3394 0 : return;
3395 0 : }
3396 :
3397 0 : if( FD_UNLIKELY( try_advance_published_root( ctx, stem ) ) ) {
3398 0 : *charge_busy = 1;
3399 0 : *opt_poll_in = 0;
3400 0 : return;
3401 0 : }
3402 :
3403 0 : if( FD_LIKELY( try_replay( ctx, stem ) ) ) {
3404 0 : *charge_busy = 1;
3405 0 : *opt_poll_in = 0;
3406 0 : return;
3407 0 : }
3408 :
3409 0 : ctx->execrp_idle_cnt++;
3410 0 : }
3411 :
3412 : static int
3413 : before_frag( fd_replay_tile_t * ctx,
3414 : ulong in_idx,
3415 : ulong seq FD_PARAM_UNUSED,
3416 0 : ulong sig ) {
3417 :
3418 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_GOSSIP_OUT && sig!=FD_GOSSIP_UPDATE_TAG_WFS_DONE ) ) return 1;
3419 0 : return 0;
3420 0 : }
3421 :
3422 : static void
3423 : process_exec_task_done( fd_replay_tile_t * ctx,
3424 : fd_stem_context_t * stem,
3425 : fd_execrp_task_done_msg_t * msg,
3426 0 : ulong sig ) {
3427 :
3428 0 : ulong exec_tile_idx = sig&0xFFFFFFFFUL;
3429 :
3430 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, msg->bank_idx );
3431 0 : FD_TEST( bank );
3432 0 : bank->refcnt--;
3433 :
3434 0 : switch( sig>>32 ) {
3435 0 : case FD_EXECRP_TT_TXN_EXEC: {
3436 0 : ulong txn_idx = msg->txn_exec->txn_idx;
3437 0 : ulong max_compute_units = fd_cost_tracker_store_peek( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id )->block_cost_limit;
3438 0 : fd_cost_tracker_store_unpin( fd_bank_cost_tracker( bank ), bank->cost_tracker_fork_id );
3439 :
3440 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
3441 : /* Query the txn signature against our recently generated vote
3442 : txn signatures. If the query is successful, then we have
3443 : seen our own vote transaction land and this should be marked
3444 : in the bank. We go through this exercise until we've seen
3445 : our vote rooted. */
3446 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, txn_idx );
3447 :
3448 0 : fd_pubkey_t * identity_pubkey_out = NULL;
3449 0 : if( fd_vote_tracker_query_sig( ctx->vote_tracker, fd_type_pun_const( txn_p->payload+TXN( txn_p )->signature_off ), &identity_pubkey_out ) && fd_pubkey_eq( identity_pubkey_out, ctx->identity_pubkey ) ) {
3450 0 : bank->f.identity_vote_idx = ctx->identity_idx;
3451 0 : }
3452 0 : }
3453 0 : if( FD_UNLIKELY( !msg->txn_exec->is_committable && bank->state!=FD_BANK_STATE_DEAD) ) {
3454 : /* Every transaction in a valid block has to execute.
3455 : Otherwise, we should mark the block as dead. Non-committable
3456 : means the txn failed before account loading or failed the cost
3457 : tracker; txn_err distinguishes the cases for telemetry. */
3458 0 : int dead_reason;
3459 0 : switch( msg->txn_exec->txn_err ) {
3460 0 : case FD_RUNTIME_TXN_ERR_WOULD_EXCEED_MAX_BLOCK_COST_LIMIT:
3461 0 : dead_reason = FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_BLOCK_COST_LIMIT;
3462 0 : break;
3463 0 : case FD_RUNTIME_TXN_ERR_WOULD_EXCEED_MAX_ACCOUNT_COST_LIMIT:
3464 0 : dead_reason = FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ACCOUNT_COST_LIMIT;
3465 0 : break;
3466 0 : case FD_RUNTIME_TXN_ERR_WOULD_EXCEED_MAX_VOTE_COST_LIMIT:
3467 0 : dead_reason = FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_VOTE_COST_LIMIT;
3468 0 : break;
3469 0 : case FD_RUNTIME_TXN_ERR_WOULD_EXCEED_ACCOUNT_DATA_BLOCK_LIMIT:
3470 0 : case FD_RUNTIME_TXN_ERR_WOULD_EXCEED_ACCOUNT_DATA_TOTAL_LIMIT:
3471 0 : dead_reason = FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_ACCOUNT_DATA_LIMIT;
3472 0 : break;
3473 0 : case FD_RUNTIME_TXN_ERR_ACCOUNT_LOADED_TWICE: /* exec-path duplicate, same taxonomy as the chkdup ruling */
3474 0 : dead_reason = FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_DUPLICATE_ACCOUNT;
3475 0 : break;
3476 0 : default:
3477 0 : dead_reason = FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_TXN_FAILED_TO_LOAD;
3478 0 : }
3479 0 : mark_bank_dead( ctx, stem, bank->idx, dead_reason, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED, NULL );
3480 0 : fd_sched_block_abandon( ctx->sched, bank->idx, FD_SCHED_ABANDON_INVALID );
3481 0 : }
3482 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, txn_idx, exec_tile_idx, NULL );
3483 0 : FD_TEST( res==0 );
3484 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
3485 0 : txn_info->flags |= FD_SCHED_TXN_EXEC_DONE;
3486 0 : txn_info->is_simple_vote = msg->txn_exec->is_simple_vote;
3487 0 : txn_info->bank_seq = msg->txn_exec->bank_seq;
3488 :
3489 0 : txn_info->tick_load_start = msg->txn_exec->tick_load_start;
3490 0 : txn_info->tick_check_start = msg->txn_exec->tick_check_start;
3491 0 : txn_info->tick_exec_start = msg->txn_exec->tick_exec_start;
3492 0 : txn_info->tick_commit_start = msg->txn_exec->tick_commit_start;
3493 0 : txn_info->tick_commit_end = msg->txn_exec->tick_commit_end;
3494 :
3495 0 : txn_info->compute_units_consumed = msg->txn_exec->compute_units_consumed;
3496 0 : txn_info->max_compute_units = max_compute_units;
3497 0 : txn_info->transaction_fee = msg->txn_exec->transaction_fee;
3498 0 : txn_info->priority_fee = msg->txn_exec->priority_fee;
3499 0 : txn_info->tips = msg->txn_exec->tips;
3500 0 : if( FD_LIKELY( !(txn_info->flags&FD_SCHED_TXN_SIGVERIFY_DONE)||!txn_info->txn_err ) ) { /* Set execution status if sigverify hasn't happened yet or if sigverify was a success. */
3501 0 : txn_info->txn_err = msg->txn_exec->txn_err;
3502 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_committable, FD_SCHED_TXN_IS_COMMITTABLE, 0UL );
3503 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_fees_only, FD_SCHED_TXN_IS_FEES_ONLY, 0UL );
3504 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_noop, FD_SCHED_TXN_IS_NOOP, 0UL );
3505 0 : }
3506 0 : if( FD_UNLIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) { /* UNLIKELY because generally exec happens before sigverify. */
3507 0 : publish_txn_executed( ctx, stem, bank->idx, txn_idx );
3508 0 : }
3509 0 : break;
3510 0 : }
3511 0 : case FD_EXECRP_TT_TXN_SIGVERIFY: {
3512 0 : ulong txn_idx = msg->txn_sigverify->txn_idx;
3513 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
3514 0 : txn_info->flags |= FD_SCHED_TXN_SIGVERIFY_DONE;
3515 0 : if( FD_UNLIKELY( msg->txn_sigverify->err ) ) {
3516 0 : txn_info->txn_err = FD_RUNTIME_TXN_ERR_SIGNATURE_FAILURE;
3517 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_COMMITTABLE;
3518 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_FEES_ONLY;
3519 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_NOOP;
3520 0 : }
3521 0 : if( FD_UNLIKELY( msg->txn_sigverify->err && bank->state!=FD_BANK_STATE_DEAD ) ) {
3522 : /* Every transaction in a valid block has to sigverify.
3523 : Otherwise, we should mark the block as dead. Also freeze the
3524 : bank if possible. */
3525 0 : mark_bank_dead( ctx, stem, bank->idx, FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_SIGVERIFY_FAILED, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED, NULL );
3526 0 : fd_sched_block_abandon( ctx->sched, bank->idx, FD_SCHED_ABANDON_INVALID );
3527 0 : }
3528 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, txn_idx, exec_tile_idx, NULL );
3529 0 : FD_TEST( res==0 );
3530 0 : if( FD_LIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) {
3531 0 : publish_txn_executed( ctx, stem, bank->idx, txn_idx );
3532 0 : }
3533 0 : break;
3534 0 : }
3535 0 : case FD_EXECRP_TT_POH_HASH: {
3536 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_POH_HASH, ULONG_MAX, exec_tile_idx, msg->poh_hash );
3537 0 : if( FD_UNLIKELY( res && bank->state!=FD_BANK_STATE_DEAD ) ) {
3538 0 : mark_bank_dead( ctx, stem, bank->idx, sched_dead_reason_to_event( res ), FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED, NULL );
3539 0 : }
3540 0 : break;
3541 0 : }
3542 0 : case FD_EXECRP_TT_CERT_VERIFY: {
3543 0 : FD_TEST( bank->cert_task_cnt );
3544 0 : bank->cert_task_cnt--;
3545 0 : if( msg->cert_verify->has_hm ) fd_footer_cache_hash_insert( ctx->footer_cache, ctx->cert_payload[ exec_tile_idx ], ctx->cert_payload_sz[ exec_tile_idx ], &msg->cert_verify->hm );
3546 0 : if( FD_UNLIKELY( msg->cert_verify->err && bank->state!=FD_BANK_STATE_DEAD ) ) {
3547 0 : mark_bank_dead( ctx, stem, bank->idx, FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_BAD_FOOTER, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED, NULL );
3548 0 : fd_sched_block_abandon( ctx->sched, bank->idx, FD_SCHED_ABANDON_INVALID );
3549 0 : }
3550 : /* Last one: freeze and publish, before sched can start its child */
3551 0 : if( !bank->cert_task_cnt && bank->state==FD_BANK_STATE_REPLAYABLE && !fd_sched_block_is_discarded( ctx->sched, bank->idx ) ) {
3552 0 : if( FD_UNLIKELY( replay_block_complete( ctx, stem, bank, fd_sched_get_footer( ctx->sched, bank->idx ), bank->execution_fees_pre_settle, bank->priority_fees_pre_settle ) ) ) {
3553 0 : fd_sched_block_abandon( ctx->sched, bank->idx, FD_SCHED_ABANDON_INVALID );
3554 0 : }
3555 0 : }
3556 0 : FD_TEST( !fd_sched_task_done( ctx->sched, FD_SCHED_TT_CERT_VERIFY, ULONG_MAX, exec_tile_idx, NULL ) );
3557 0 : break;
3558 0 : }
3559 0 : default: FD_LOG_CRIT(( "unexpected sig 0x%lx", sig ));
3560 0 : }
3561 :
3562 : /* Reference counter just decreased, and an exec tile just got freed
3563 : up. If there's a need to be more aggressively pruning, we could
3564 : check here if more slots just became publishable and publish. Not
3565 : publishing here shouldn't bloat the fork tree too much though. We
3566 : mark minority forks dead as soon as we can, and execution dispatch
3567 : stops on dead blocks. So shortly afterwards, dead blocks should be
3568 : eligible for pruning as in-flight transactions retire from the
3569 : execution pipeline. */
3570 :
3571 0 : }
3572 :
3573 : /* Essentially an inlined version of try_process_fec that reads off the
3574 : dcache directly, instead of the reasm out_queue.
3575 : All FECs processed by this function must be safe to be forever
3576 : removed from the dcache. */
3577 : static void
3578 : process_rotor_fec( fd_replay_tile_t * ctx,
3579 : fd_stem_context_t * stem,
3580 0 : fd_rotor_replay_fec_t * fec ) {
3581 0 : if( FD_LIKELY( fec->metrics.highest_fec_complete_slot && ( ctx->catch_up_max_fec_slot==ULONG_MAX || fec->metrics.highest_fec_complete_slot>ctx->catch_up_max_fec_slot ) ) ) {
3582 0 : ctx->catch_up_max_fec_slot = fec->metrics.highest_fec_complete_slot;
3583 0 : ctx->catch_up_tip_advance_cnt++;
3584 0 : }
3585 :
3586 0 : ulong parent_bank_idx = ULONG_MAX;
3587 0 : if( FD_UNLIKELY( fec->fec_set_idx==0 ) ) {
3588 0 : ag_block_id_t parent_key = ag_block_id( fec->parent_slot, fec->parent_block_id.uc );
3589 0 : fd_block_id_ele_t * parent = fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &parent_key, NULL, ctx->block_id_arr );
3590 0 : FD_TEST( parent );
3591 0 : parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent );
3592 0 : } else {
3593 : /* mid-slot FEC: rotor tile promises to populate block_id
3594 : correctly for blocks with known block id. If we are processing
3595 : the last FEC set of a slot received through turbine though, the
3596 : block_id_map will still be keyed with {0}, but the block_id on
3597 : the FEC will have the computed DMR. We first try to replay
3598 : this FEC on a bank keyed with the computed DMR - if it doesn't
3599 : exist, it must belong to the original turbine version. */
3600 0 : ag_block_id_t key = { .slot = fec->slot };
3601 0 : if( FD_LIKELY( fec->known_id ) ) {
3602 0 : key = ag_block_id( fec->slot, fec->block_id.uc );
3603 0 : }
3604 :
3605 0 : fd_block_id_ele_t * parent = fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &key, NULL, ctx->block_id_arr );
3606 0 : FD_TEST( parent );
3607 0 : parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent );
3608 0 : }
3609 :
3610 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
3611 0 : FD_TEST( parent_bank );
3612 :
3613 0 : if( FD_UNLIKELY( parent_bank->state==FD_BANK_STATE_DEAD ) ) {
3614 0 : FD_LOG_WARNING(( "parent bank is dead for slot %lu, fec set idx %u, parent slot %lu, dropping", fec->slot, fec->fec_set_idx, fec->parent_slot ));
3615 0 : return;
3616 0 : }
3617 :
3618 0 : long work = fd_tickcount();
3619 :
3620 0 : fd_store_fec_data_view_t data_view[1];
3621 0 : fd_rwlock_read( &ctx->store->fec_lock );
3622 0 : fd_store_fec_t * store_fec = fd_store_query( ctx->map_join, &fec->mr );
3623 0 : ctx->metrics.store_query_cnt++;
3624 0 : int view_err = store_fec ? fd_store_fec_data_view( ctx->store, ctx->store_disk_fd, store_fec, data_view ) : -1;
3625 0 : fd_rwlock_unread( &ctx->store->fec_lock );
3626 :
3627 : /* A missing store entry is expected: rotor (the store publisher)
3628 : removes FEC sets on publish, so a FEC delivered for a slice that has
3629 : since been pruned/rooted is no longer in the store. The lookup and
3630 : payload pin are one lifetime transaction under fec_lock: remove may
3631 : unlink the key only after the view has pinned the pool element. */
3632 0 : if( FD_UNLIKELY( !store_fec || view_err ) ) {
3633 0 : ctx->metrics.store_query_missing_cnt++;
3634 0 : FD_BASE58_ENCODE_32_BYTES( fec->mr.key, key_b58 );
3635 0 : FD_LOG_INFO(( "store fec for slot: %lu unavailable (pruned by publish); abandoning slice. root: %lu. merkle: %s", fec->slot, ctx->consensus_root_slot, key_b58 ));
3636 0 : return;
3637 0 : }
3638 :
3639 0 : long now = fd_log_wallclock();
3640 0 : fd_block_id_ele_t * block_id_ele;
3641 0 : fd_bank_t * bank;
3642 0 : if( FD_UNLIKELY( fec->fec_set_idx==0U ) ) {
3643 0 : bank = fec->is_leader ? ctx->leader_bank : fd_banks_new_bank( ctx->banks, parent_bank_idx, now, 0 );
3644 0 : if( FD_UNLIKELY( fec->is_leader && ctx->leader_stats.slot==fec->slot ) ) {
3645 0 : ctx->leader_stats.first_fec_returned_nanos = now;
3646 0 : }
3647 :
3648 0 : block_id_ele = &ctx->block_id_arr[ bank->idx ];
3649 : /* Clear the previous occupant of this reused bank idx */
3650 0 : if( FD_LIKELY( fd_ag_block_id_map_ele_query( ctx->ag_block_id_map, &block_id_ele->block_info, NULL, ctx->block_id_arr )==block_id_ele ) ) {
3651 0 : FD_TEST( fd_ag_block_id_map_ele_remove( ctx->ag_block_id_map, &block_id_ele->block_info, NULL, ctx->block_id_arr ) );
3652 0 : }
3653 :
3654 0 : block_id_ele->block_id_seen = 0;
3655 0 : block_id_ele->bank_seq = bank->bank_seq;
3656 0 : block_id_ele->slot = fec->slot;
3657 0 : block_id_ele->latest_fec_idx = 0U;
3658 0 : block_id_ele->fec_cnt = 0U;
3659 0 : block_id_ele->block_info = fec->known_id ? ag_block_id( fec->slot, fec->block_id.uc ) : (ag_block_id_t){ .slot = fec->slot };
3660 :
3661 : /* If this fec 0 is rebuilding an evicted block (recovery redelivery),
3662 : the evicted bank's stale map entry may still hold this key at a
3663 : DIFFERENT idx. */
3664 0 : fd_ag_block_id_map_ele_remove( ctx->ag_block_id_map, &block_id_ele->block_info, NULL, ctx->block_id_arr ); // evicted / stale entry
3665 0 : FD_TEST( fd_ag_block_id_map_ele_insert( ctx->ag_block_id_map, block_id_ele, ctx->block_id_arr ) );
3666 0 : } else { /* FEC for the middle or end of a block */
3667 : /* Assign bank idx + seqno to the FEC. Update block id pool ele.
3668 : The block stays keyed by {slot, 0} in the compound map. */
3669 0 : block_id_ele = &ctx->block_id_arr[ parent_bank_idx ];
3670 0 : block_id_ele->latest_fec_idx = fec->fec_set_idx;
3671 0 : bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
3672 0 : }
3673 :
3674 0 : block_id_ele->latest_mr = fec->mr;
3675 :
3676 0 : if( FD_UNLIKELY( ctx->report_runtime_diffs && block_id_ele->fec_cnt<FD_FEC_BLK_MAX ) ) {
3677 0 : ctx->fec_chain[ bank->idx*FD_FEC_BLK_MAX + block_id_ele->fec_cnt ] = fec->mr;
3678 0 : }
3679 0 : block_id_ele->fec_cnt++;
3680 :
3681 0 : fd_reception_stats_t * stats = &ctx->reception_stats[ bank->idx ];
3682 0 : stats->bank_seq = bank->bank_seq;
3683 0 : stats->fec_set_idx = fec->fec_set_idx;
3684 0 : stats->metrics.rotor = fec->metrics; /* votor_repaired always, the counters only if stats_valid */
3685 :
3686 0 : if( FD_UNLIKELY( fec->slot_complete ) ) {
3687 0 : FD_BASE58_ENCODE_32_BYTES( fec->block_id.uc, block_id_b58 );
3688 0 : FD_LOG_INFO(( "slot %lu fec set idx %u slot_bid %s is complete, inserted to map", fec->slot, fec->fec_set_idx, block_id_b58 ));
3689 0 : block_id_ele->block_id_seen = 1;
3690 0 : block_id_ele->dmr = fec->block_id;
3691 :
3692 : /* block is complete and DMR block id is now known so re-key. The
3693 : map does not support duplicate keys: rotor guarantees at most one
3694 : delivery stream per block (a turbine copy of a slot with a
3695 : votor-driven version is abandoned in the rotor, see
3696 : fd_rotor_tile.h), so {slot, block_id} can never already be
3697 : occupied by another LIVE bank. It can however be occupied by a
3698 : STALE entry: a previously-completed copy of this block whose bank
3699 : was evicted (eviction does not remove map entries), now being
3700 : rebuilt through recovery redelivery. Evict the stale entry before
3701 : inserting. */
3702 0 : FD_TEST( fd_ag_block_id_map_ele_remove( ctx->ag_block_id_map, &block_id_ele->block_info, NULL, ctx->block_id_arr )==block_id_ele );
3703 0 : ag_block_id_t key = ag_block_id( fec->slot, fec->block_id.uc );
3704 0 : block_id_ele->block_info = key;
3705 0 : fd_ag_block_id_map_ele_remove( ctx->ag_block_id_map, &key, NULL, ctx->block_id_arr ); // evicted / stale entry
3706 0 : FD_TEST( fd_ag_block_id_map_ele_insert( ctx->ag_block_id_map, block_id_ele, ctx->block_id_arr ) );
3707 0 : }
3708 :
3709 : /* For leader FECs, don't insert the FEC into the scheduler. */
3710 0 : if( FD_UNLIKELY( fec->is_leader ) ) {
3711 0 : fd_store_fec_data_view_release( ctx->store, data_view );
3712 0 : return;
3713 0 : }
3714 :
3715 : /* Forks form a partial ordering over FEC sets. The Rotor tile
3716 : delivers FEC sets in-order per fork, but FEC set ordering across
3717 : forks is arbitrary */
3718 0 : fd_sched_fec_t sched_fec[ 1 ];
3719 0 : sched_fec->shred_cnt = FD_FEC_SHRED_CNT;
3720 0 : sched_fec->is_last_in_batch = !!fec->data_complete;
3721 0 : sched_fec->is_last_in_block = !!fec->slot_complete;
3722 0 : sched_fec->bank_idx = bank->idx;
3723 0 : sched_fec->parent_bank_idx = bank->parent_idx;
3724 0 : sched_fec->slot = fec->slot;
3725 0 : sched_fec->parent_slot = fec->parent_slot;
3726 0 : sched_fec->is_first_in_block = fec->fec_set_idx==0U;
3727 0 : sched_fec->fec = store_fec;
3728 0 : sched_fec->data = data_view->data;
3729 0 : sched_fec->alut_ctx->fork_id = fd_banks_bank_query( ctx->banks, ctx->published_root_bank_idx )->accdb_fork_id;
3730 0 : sched_fec->alut_ctx->accdb = ctx->accdb;
3731 0 : sched_fec->alut_ctx->els = ctx->published_root_slot;
3732 0 : sched_fec->completed_ns = fec->metrics.fec_completed_ts_nanos;
3733 :
3734 0 : if( sched_fec->is_first_in_block ) {
3735 0 : bank->refcnt++;
3736 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for sched", bank->idx, sched_fec->slot, bank->refcnt ));
3737 0 : }
3738 :
3739 0 : int ingested = fd_sched_fec_ingest( ctx->sched, sched_fec );
3740 0 : fd_store_fec_data_view_release( ctx->store, data_view );
3741 :
3742 0 : if( FD_UNLIKELY( !ingested ) ) {
3743 0 : int dr = sched_block_dead_reason_to_event( ctx, sched_fec->bank_idx );
3744 0 : int ar = dr==FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD ? FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_PRUNED
3745 0 : : FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_NOT_ABANDONED;
3746 0 : mark_bank_dead( ctx, stem, sched_fec->bank_idx, dr, ar, NULL );
3747 0 : return;
3748 0 : }
3749 :
3750 0 : fd_histf_sample( ctx->metrics.store_query_work, (ulong)( fd_tickcount() - work ) );
3751 0 : ctx->execrp_idle_cnt = 0UL;
3752 0 : return;
3753 0 : }
3754 :
3755 : static void
3756 0 : process_resolv_slot_completed( fd_replay_tile_t * ctx, ulong bank_idx ) {
3757 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
3758 0 : FD_TEST( bank );
3759 0 : bank->refcnt--;
3760 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for resolv", bank->idx, bank->f.slot, bank->refcnt ));
3761 0 : }
3762 :
3763 : static void
3764 : process_vote_txn_sent( fd_replay_tile_t * ctx,
3765 0 : fd_txn_m_t * txnm ) {
3766 : /* The send tile has signed and sent a vote. Add this vote to the
3767 : vote tracker. We go through this exercise until the client has
3768 : seen a vote corresponding to the current identity rooted. */
3769 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
3770 0 : uchar * payload = (uchar *)txnm + sizeof(fd_txn_m_t);
3771 0 : uchar txn_mem[ FD_TXN_MAX_SZ ] __attribute__((aligned(alignof(fd_txn_t))));
3772 0 : fd_txn_t * txn = (fd_txn_t *)txn_mem;
3773 0 : if( FD_UNLIKELY( !fd_txn_parse( payload, txnm->payload_sz, txn_mem, NULL ) ) ) {
3774 0 : FD_LOG_CRIT(( "Could not parse txn from send tile" ));
3775 0 : }
3776 : /* The identity of the validator that the signed the vote will
3777 : always be the first signer in the vote transaction. */
3778 0 : fd_pubkey_t * vote_identity = fd_type_pun( payload+txn->acct_addr_off );
3779 0 : fd_vote_tracker_insert( ctx->vote_tracker, vote_identity, fd_type_pun_const( payload+txn->signature_off ) );
3780 0 : }
3781 0 : }
3782 :
3783 : static inline void
3784 0 : maybe_verify_shred_version( fd_replay_tile_t * ctx ) {
3785 0 : if( FD_LIKELY( ctx->has_genesis_hash && ctx->hard_fork_cnt!=ULONG_MAX ) ) {
3786 0 : ctx->shred_version = compute_shred_version( ctx->genesis_hash->uc, ctx->hard_forks, ctx->hard_fork_cnt );
3787 0 : }
3788 :
3789 0 : if( FD_LIKELY( ctx->expected_shred_version && ctx->ipecho_shred_version ) ) {
3790 0 : if( FD_UNLIKELY( ctx->expected_shred_version!=ctx->ipecho_shred_version ) ) {
3791 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from ipecho", ctx->expected_shred_version, ctx->ipecho_shred_version ) );
3792 0 : }
3793 0 : }
3794 :
3795 : /* The configured shred version is the post-restart one. Compare it
3796 : only on the boot bank: before boot the hard forks are the full
3797 : snapshot's, predating the restart. */
3798 0 : if( FD_UNLIKELY( ctx->wfs_enabled && !ctx->is_booted && ctx->expected_shred_version ) ) return;
3799 :
3800 0 : if( FD_LIKELY( ctx->shred_version && (ctx->expected_shred_version || ctx->ipecho_shred_version) ) ) {
3801 0 : ushort expected_shred_version = ctx->expected_shred_version ? ctx->expected_shred_version : ctx->ipecho_shred_version;
3802 0 : ushort actual_shred_version = ctx->shred_version;
3803 :
3804 0 : if( FD_UNLIKELY( expected_shred_version!=actual_shred_version ) ) {
3805 0 : char const * expected_source = ctx->expected_shred_version ?
3806 0 : "[consensus.expected_shred_version]" : "the boot entrypoint";
3807 0 : FD_LOG_ERR(( "genesis.bin at `%s` plus the loaded snapshot's hard_forks give shred version %hu, "
3808 0 : "but %s reports %hu. Delete genesis.bin to re-download it, or the snapshot if it "
3809 0 : "predates the latest hard fork, and restart.",
3810 0 : ctx->genesis_path, actual_shred_version, expected_source, expected_shred_version ));
3811 0 : }
3812 0 : }
3813 0 : }
3814 :
3815 : static inline void
3816 0 : maybe_verify_genesis_timestamp( fd_replay_tile_t * ctx ) {
3817 0 : if( FD_LIKELY( !ctx->has_expected_genesis_timestamp || !ctx->has_genesis_timestamp ) ) return;
3818 0 : if( FD_LIKELY( ctx->genesis_timestamp==ctx->expected_genesis_timestamp ) ) return;
3819 :
3820 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis timestamp of %lu but the snapshot you loaded has a genesis "
3821 0 : "timestamp of %lu. This either means that the genesis.bin file you have is for a different cluster than "
3822 0 : "the one you are trying to connect to, or you have loaded a snapshot for the wrong cluster. In either "
3823 0 : "case, you can delete the problematic file and restart the node to download the correct one automatically.",
3824 0 : ctx->genesis_path, ctx->genesis_timestamp, ctx->expected_genesis_timestamp ));
3825 0 : }
3826 :
3827 : /* snapmk_start instructs the snapmk tile to start producing a snapshot. */
3828 :
3829 : static void
3830 : snapmk_start( fd_replay_tile_t * ctx,
3831 : fd_stem_context_t * stem,
3832 0 : int incremental ) {
3833 :
3834 0 : FD_CHECK_CRIT( !ctx->snapmk.active, "snapshot creation already in progress" );
3835 :
3836 : /* pin current produced bank */
3837 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, ctx->published_root_bank_idx );
3838 0 : FD_CHECK_CRIT( bank, "invalid published_root_bank_idx" );
3839 :
3840 0 : if( FD_UNLIKELY( incremental ) ) {
3841 0 : FD_CHECK_CRIT( ctx->snapmk.base_slot!=ULONG_MAX, "incremental snapshot without a base full snapshot" );
3842 0 : FD_CHECK_CRIT( bank->f.slot>ctx->snapmk.base_slot, "incremental snapshot at or below its base slot" );
3843 0 : }
3844 :
3845 0 : bank->refcnt++;
3846 0 : ctx->snapmk.bank_idx = bank->idx;
3847 0 : ctx->snapmk.incremental = !!incremental;
3848 :
3849 : /* Snapmk reads the bank's epoch credits in place, so they stay
3850 : pinned in memory until snapmk_done. */
3851 0 : FD_TEST( fd_epoch_credits_view_init( ctx->snapmk.epoch_credits_view, fd_bank_epoch_credits( bank ), bank->epoch_credits_fork_id ) );
3852 :
3853 : /* Send SNAP_START message to snapmk. */
3854 0 : fd_replay_snap_start_t * msg = fd_chunk_to_laddr( ctx->snapmk_out->mem, ctx->snapmk_out->chunk );
3855 0 : *msg = (fd_replay_snap_start_t) {
3856 0 : .bank_idx = ctx->published_root_bank_idx,
3857 0 : .base_slot = incremental ? ctx->snapmk.base_slot : bank->f.slot,
3858 0 : .slot = bank->f.slot,
3859 0 : .leader = bank->leader
3860 0 : };
3861 0 : ulong out_idx = ctx->snapmk_out->idx;
3862 0 : ulong sig = REPLAY_SIG_SNAP_START;
3863 0 : ulong chunk = ctx->snapmk_out->chunk;
3864 0 : ulong tspub = fd_frag_meta_ts_comp( fd_tickcount() );
3865 0 : ulong sz = sizeof(fd_replay_snap_start_t);
3866 0 : ulong seq = fd_stem_publish( stem, out_idx, sig, chunk, sz, 0UL, 0UL, tspub );
3867 0 : ctx->snapmk_out->chunk = fd_dcache_compact_next( ctx->snapmk_out->chunk, sz, ctx->snapmk_out->chunk0, ctx->snapmk_out->wmark );
3868 :
3869 : /* wake up the snapmk tile */
3870 0 : fd_frag_meta_t * replay_snapmk = stem->mcaches[ out_idx ];
3871 0 : ulong * snap_sync = fd_mcache_seq_laddr( replay_snapmk );
3872 0 : fd_mcache_seq_update( snap_sync, fd_seq_inc( seq, 1UL ) );
3873 0 : long ret = syscall( SYS_futex, snap_sync, FUTEX_WAKE, 1 );
3874 0 : if( FD_UNLIKELY( ret<0 ) ) {
3875 0 : FD_LOG_ERR(( "FUTEX_WAKE(snap_sync,seq=%u) failed (%i-%s)", (uint)seq, errno, fd_io_strerror( errno ) ));
3876 0 : }
3877 :
3878 : /* update internal state */
3879 0 : ctx->snapmk.active = 1;
3880 0 : }
3881 :
3882 : /* snapmk_done reacts to the snapmk tile reporting completion. */
3883 :
3884 : static void
3885 : snapmk_done( fd_replay_tile_t * ctx,
3886 : fd_stem_context_t * stem,
3887 0 : int success ) {
3888 0 : (void)stem;
3889 :
3890 0 : FD_CHECK_CRIT( ctx->snapmk.active, "spurious snap complete msg (not creating snapshot)" );
3891 :
3892 : /* release bank */
3893 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, ctx->snapmk.bank_idx );
3894 0 : FD_CHECK_CRIT( bank, "invalid snapmk.bank_idx" );
3895 0 : FD_CHECK_CRIT( bank->refcnt > 0UL, "invalid snapmk.bank_idx refcnt" );
3896 :
3897 : /* A completed full snapshot becomes the base of later incrementals. */
3898 0 : if( FD_LIKELY( success && !ctx->snapmk.incremental ) ) {
3899 0 : ctx->snapmk.base_slot = bank->f.slot;
3900 0 : }
3901 :
3902 0 : if( FD_LIKELY( success ) ) {
3903 0 : if( FD_UNLIKELY( ctx->snapmk.incremental ) ) {
3904 0 : ctx->snapmk.snap_finished_incr = bank->f.slot;
3905 0 : ctx->snapmk.snap_produced_incr_cnt++;
3906 0 : } else {
3907 0 : ctx->snapmk.snap_finished_full = bank->f.slot;
3908 0 : }
3909 0 : }
3910 :
3911 0 : fd_epoch_credits_view_fini( ctx->snapmk.epoch_credits_view );
3912 0 : bank->refcnt--;
3913 0 : ctx->snapmk.active = 0;
3914 0 : }
3915 :
3916 : static void
3917 : msg_snapmk( fd_replay_tile_t * ctx,
3918 : fd_stem_context_t * stem,
3919 : ulong msg_type,
3920 : ulong in_idx,
3921 0 : ulong chunk ) {
3922 0 : switch( msg_type ) {
3923 0 : case FD_SNAPMK_MSG_FOUND: {
3924 0 : fd_snapmk_msg_found_t const * found = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
3925 0 : if( FD_UNLIKELY( found->base_slot==ULONG_MAX ) ) {
3926 0 : if( found->slot > ctx->snapmk.snap_finished_full ) ctx->snapmk.snap_finished_full = found->slot;
3927 0 : } else {
3928 0 : if( found->slot > ctx->snapmk.snap_finished_incr ) ctx->snapmk.snap_finished_incr = found->slot;
3929 0 : }
3930 0 : break;
3931 0 : }
3932 0 : case FD_SNAPMK_MSG_DELETED: {
3933 : /* Zeroing loses older surviving snapshots, but keeping a stale
3934 : watermark would claim one that no longer exists. */
3935 0 : fd_snapmk_msg_deleted_t const * deleted = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
3936 0 : if( FD_UNLIKELY( deleted->base_slot==ULONG_MAX ) ) {
3937 0 : if( deleted->slot==ctx->snapmk.snap_finished_full ) ctx->snapmk.snap_finished_full = 0UL;
3938 0 : } else {
3939 0 : if( deleted->slot==ctx->snapmk.snap_finished_incr ) ctx->snapmk.snap_finished_incr = 0UL;
3940 0 : }
3941 0 : break;
3942 0 : }
3943 0 : case FD_SNAPMK_MSG_CREATED:
3944 0 : snapmk_done( ctx, stem, 1 );
3945 0 : break;
3946 0 : case FD_SNAPMK_MSG_FAILED:
3947 0 : snapmk_done( ctx, stem, 0 );
3948 0 : break;
3949 0 : default:
3950 0 : break;
3951 0 : }
3952 0 : }
3953 :
3954 : /* admin command handlers
3955 : every admin command must trigger one response frag */
3956 :
3957 : static void
3958 : admin_respond( fd_replay_tile_t * ctx,
3959 : fd_stem_context_t * stem,
3960 : ulong orig,
3961 0 : ulong err ) {
3962 0 : ulong ctl = fd_frag_meta_ctl( orig, 0, 0, !!err );
3963 0 : ulong tspub = fd_frag_meta_ts_comp( fd_tickcount() );
3964 0 : fd_stem_publish( stem, ctx->admin_out_idx, err, 0UL, 0UL, ctl, 0UL, tspub );
3965 0 : }
3966 :
3967 : static void
3968 : admin_snap_create( fd_replay_tile_t * ctx,
3969 : fd_stem_context_t * stem,
3970 0 : ulong sig ) {
3971 0 : ulong target_slot = sig;
3972 :
3973 0 : if( FD_UNLIKELY( !ctx->snapmk.supported ) ) {
3974 0 : FD_LOG_WARNING(( "admin requested snapshot creation, but current config cannot create snapshots. increase [layout.snapzp_tile_count]?" ));
3975 0 : admin_respond( ctx, stem, FD_ADMINCTL_CMD_SNAP_CREATE, FD_ADMINCTL_RESULT_UNSUPPORTED );
3976 0 : return;
3977 0 : }
3978 :
3979 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) {
3980 0 : FD_LOG_WARNING(( "admin requested snapshot creation, but client has not yet started" ));
3981 0 : admin_respond( ctx, stem, FD_ADMINCTL_CMD_SNAP_CREATE, FD_SNAPSHOT_CREATE_RESULT_NOT_READY );
3982 0 : return;
3983 0 : }
3984 :
3985 0 : if( FD_UNLIKELY( target_slot ) ) {
3986 0 : if( FD_UNLIKELY( target_slot<=ctx->published_root_slot ) ) {
3987 0 : FD_LOG_WARNING(( "admin requested snapshot creation at slot %lu, but rooting is already past it (published root slot %lu)", target_slot, ctx->published_root_slot ));
3988 0 : admin_respond( ctx, stem, FD_ADMINCTL_CMD_SNAP_CREATE, FD_SNAPSHOT_CREATE_RESULT_SLOT_IN_PAST );
3989 0 : return;
3990 0 : }
3991 :
3992 :
3993 0 : if( FD_UNLIKELY( ctx->snapmk.scheduled_at_slot!=ULONG_MAX &&
3994 0 : ctx->snapmk.scheduled_at_slot!=target_slot ) ) {
3995 0 : FD_LOG_WARNING(( "admin requested snapshot creation at slot %lu, but a snapshot is already scheduled at slot %lu. ignoring ...", target_slot, ctx->snapmk.scheduled_at_slot ));
3996 0 : admin_respond( ctx, stem, FD_ADMINCTL_CMD_SNAP_CREATE, FD_SNAPSHOT_CREATE_RESULT_BUSY );
3997 0 : return;
3998 0 : }
3999 :
4000 0 : ctx->snapmk.scheduled_at_slot = target_slot;
4001 0 : FD_LOG_NOTICE(( "snapshot creation scheduled at slot %lu", target_slot ));
4002 0 : admin_respond( ctx, stem, FD_ADMINCTL_CMD_SNAP_CREATE, FD_ADMINCTL_RESULT_SUCCESS );
4003 0 : return;
4004 0 : }
4005 :
4006 0 : if( FD_UNLIKELY( ctx->snapmk.active ) ) {
4007 0 : FD_LOG_WARNING(( "admin requested snapshot creation, but currently busy creating another snapshot. ignoring ..." ));
4008 0 : admin_respond( ctx, stem, FD_ADMINCTL_CMD_SNAP_CREATE, FD_SNAPSHOT_CREATE_RESULT_BUSY );
4009 0 : return;
4010 0 : }
4011 :
4012 0 : snapmk_start( ctx, stem, 0 );
4013 0 : admin_respond( ctx, stem, FD_ADMINCTL_CMD_SNAP_CREATE, FD_ADMINCTL_RESULT_SUCCESS );
4014 0 : }
4015 :
4016 : static void
4017 : msg_admin( fd_replay_tile_t * ctx,
4018 : fd_stem_context_t * stem,
4019 : ulong orig,
4020 0 : ulong sig ) {
4021 0 : switch( orig ) {
4022 0 : case FD_ADMINCTL_CMD_SNAP_CREATE:
4023 0 : admin_snap_create( ctx, stem, sig );
4024 0 : break;
4025 0 : default:
4026 0 : FD_LOG_CRIT(( "unknown admin cmd (orig=%lu, sig=%lu)", orig, sig ));
4027 0 : }
4028 0 : }
4029 :
4030 : static inline int
4031 : returnable_frag( fd_replay_tile_t * ctx,
4032 : ulong in_idx,
4033 : ulong seq,
4034 : ulong sig,
4035 : ulong chunk,
4036 : ulong sz,
4037 : ulong ctl,
4038 : ulong tsorig,
4039 : ulong tspub,
4040 0 : fd_stem_context_t * stem ) {
4041 0 : (void)seq;
4042 0 : (void)ctl;
4043 0 : (void)tsorig;
4044 0 : (void)tspub;
4045 :
4046 0 : if( FD_UNLIKELY( sz!=0UL && (chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark || sz>ctx->in[ in_idx ].mtu ) ) )
4047 0 : FD_LOG_CRIT(( "chunk %lu %lu from in %d corrupt, not in range [%lu,%lu]", chunk, sz, ctx->in_kind[ in_idx ], ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
4048 :
4049 0 : switch( ctx->in_kind[in_idx] ) {
4050 0 : case IN_KIND_GENESIS: {
4051 0 : fd_genesis_meta_t const * meta = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
4052 0 : ctx->has_genesis_hash = 1;
4053 0 : ctx->has_genesis_timestamp = 1;
4054 0 : ctx->genesis_timestamp = meta->creation_time_seconds;
4055 0 : *ctx->genesis_hash = meta->genesis_hash;
4056 0 : fd_node_info_write_begin( ctx->node_info );
4057 0 : ctx->node_info->info.genesis_hash = *ctx->genesis_hash;
4058 0 : fd_node_info_write_end( ctx->node_info );
4059 0 : if( FD_LIKELY( meta->bootstrap ) ) {
4060 0 : boot_genesis( ctx, stem, meta );
4061 0 : } else {
4062 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
4063 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
4064 0 : }
4065 0 : ctx->has_genesis_timestamp = 1;
4066 0 : ctx->genesis_timestamp = ctx->genesis->creation_time;
4067 :
4068 0 : maybe_verify_cluster_type( ctx );
4069 0 : maybe_verify_shred_version( ctx );
4070 0 : maybe_verify_genesis_timestamp( ctx );
4071 0 : break;
4072 0 : }
4073 0 : case IN_KIND_IPECHO: {
4074 0 : FD_TEST( sig && sig<=USHORT_MAX );
4075 0 : ctx->ipecho_shred_version = (ushort)sig;
4076 0 : maybe_verify_shred_version( ctx );
4077 0 : break;
4078 0 : }
4079 0 : case IN_KIND_SNAP: {
4080 0 : on_snapshot_message( ctx, stem, in_idx, chunk, sig );
4081 0 : maybe_verify_cluster_type( ctx );
4082 0 : maybe_verify_shred_version( ctx );
4083 0 : maybe_verify_genesis_timestamp( ctx );
4084 0 : break;
4085 0 : }
4086 0 : case IN_KIND_EXECRP: {
4087 0 : process_exec_task_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), sig );
4088 0 : ctx->execrp_idle_cnt = 0UL;
4089 0 : break;
4090 0 : }
4091 0 : case IN_KIND_POH: {
4092 0 : process_poh_message( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
4093 0 : break;
4094 0 : }
4095 0 : case IN_KIND_RESOLV: {
4096 0 : fd_resolv_slot_exchanged_t * exchanged_slot = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
4097 0 : process_resolv_slot_completed( ctx, exchanged_slot->bank_idx );
4098 0 : break;
4099 0 : }
4100 0 : case IN_KIND_VOTOR: {
4101 0 : if( FD_UNLIKELY( sig==FD_VOTOR_SIG_LEADER ) ) {
4102 0 : fd_votor_leader_t const * leader = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
4103 0 : *ctx->votor_leader = *leader;
4104 0 : ctx->next_leader_slot = leader->slot;
4105 0 : try_become_leader_ag( ctx, stem, FD_METRICS_SAMPLE_REASON_LEADER_START, 0UL );
4106 0 : } else if( FD_LIKELY( sig==FD_VOTOR_SIG_CERTED ) ) {
4107 0 : fd_votor_certed_t const * certed = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
4108 0 : fd_votor_certed_t * fin = ctx->votor_final;
4109 0 : switch( certed->kind ) {
4110 0 : case AG_CERT_KIND_FINAL:
4111 0 : if( fin->slot==ULONG_MAX || certed->slot>fin->slot ) *fin = *certed;
4112 0 : try_advance_root_ag( ctx, ag_block_id( certed->slot, certed->block_id.uc ) );
4113 0 : break;
4114 0 : case AG_CERT_KIND_FAST_FINAL: /* fast beats slow at the same slot */
4115 0 : if( fin->slot==ULONG_MAX || certed->slot>fin->slot || ( certed->slot==fin->slot && fin->kind==AG_CERT_KIND_FINAL ) ) *fin = *certed;
4116 0 : try_advance_root_ag( ctx, ag_block_id( certed->slot, certed->block_id.uc ) );
4117 0 : break;
4118 0 : default: break;
4119 0 : }
4120 0 : } else if( FD_UNLIKELY( sig==FD_VOTOR_SIG_PARENT_READY ) ) {
4121 0 : fd_votor_parent_ready_t const * pr = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
4122 0 : ctx->votor_parent_ready_slot = fd_ulong_max( ctx->votor_parent_ready_slot, pr->slot );
4123 0 : } else if( FD_UNLIKELY( sig==FD_VOTOR_SIG_REWARD ) ) {
4124 0 : fd_votor_reward_t const * reward = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
4125 0 : fd_votor_reward_t * ring = &ctx->votor_reward[ reward->slot%(FD_NUM_SLOTS_FOR_REWARD+AG_SLOTS_PER_WINDOW+1UL) ];
4126 0 : if( ring->slot==ULONG_MAX || reward->slot>=ring->slot ) *ring = *reward;
4127 0 : }
4128 0 : break;
4129 0 : }
4130 0 : case IN_KIND_REPAIR: {
4131 0 : if( FD_UNLIKELY( sig!=ROTOR_SIG_FEC_REPLAY ) ) break;
4132 :
4133 : /* process rotor incoming FECs. 1 to keep frag for retry, returning 0 consumes it. */
4134 :
4135 0 : fd_rotor_replay_fec_t * fec = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
4136 0 : if( FD_UNLIKELY( !( ctx->execrp_idle_cnt>=2UL*ctx->in_cnt || ctx->is_leader ) ) ) return 1; /* pace behind exec notifications, retry */
4137 :
4138 0 : int evict_banks = 0;
4139 0 : int res = can_process_rotor_fec( ctx, fec, &evict_banks );
4140 :
4141 : /* drain_rotor_fecs: a bank eviction broke the replayable chain of
4142 : FECs delivered from rotor (we lost a parent we needed to
4143 : replay off of). While draining, ignore DROP/SKIP FECs until
4144 : one is replayable again (PROCESS_FEC_OK, i.e. its parent
4145 : context is present), then resume normal processing. WAIT must
4146 : fall through to the retry/eviction logic below: WAIT's contract
4147 : is keep-and-retry, and the redelivered FEC that rebuilds the
4148 : evicted block can itself return WAIT+evict_banks (banks full) --
4149 : swallowing it would consume the frag, never queue the eviction,
4150 : and wedge recovery. */
4151 0 : if( FD_UNLIKELY( ctx->drain_rotor_fecs ) ) {
4152 0 : if( FD_LIKELY( res==PROCESS_FEC_OK ) ) {
4153 0 : ctx->drain_rotor_fecs = 0; /* chain re-established, resume */
4154 0 : }
4155 0 : else if( res!=PROCESS_FEC_WAIT ) return 0; /* swallow DROPs/SKIPs, do evict_banks if needed */
4156 0 : }
4157 :
4158 0 : switch( res ) {
4159 0 : case PROCESS_FEC_OK: {
4160 0 : process_rotor_fec( ctx, stem, fec );
4161 0 : return 0;
4162 0 : }
4163 0 : case PROCESS_FEC_DROP: {
4164 : /* enter drain state */
4165 0 : ctx->drain_rotor_fecs = 1;
4166 0 : publish_replay_out( ctx, stem, REPLAY_SIG_MISSING_FEC, 0 );
4167 0 : return 0;
4168 0 : }
4169 0 : case PROCESS_FEC_SKIP: { return 0; }
4170 0 : case PROCESS_FEC_WAIT: {
4171 : /* queue an eviction, then retry the frag. */
4172 0 : if( FD_UNLIKELY( evict_banks ) ) {
4173 0 : ulong evictable_bank_idx = fd_banks_get_evictable_bank( ctx->banks, ctx->notified_root_bank );
4174 0 : if( FD_UNLIKELY( evictable_bank_idx==ULONG_MAX ) ) {
4175 0 : FD_LOG_DEBUG(( "replay has no banks to mark as prunable, it's possible that there is one bank already marked as prunable" ));
4176 0 : return 1;
4177 0 : }
4178 :
4179 0 : FD_LOG_WARNING(( "banks full, evicting bank (idx=%lu)", evictable_bank_idx ));
4180 :
4181 0 : timing_slot_release( ctx, evictable_bank_idx );
4182 :
4183 0 : if( FD_UNLIKELY( fd_sched_block_is_discarded( ctx->sched, evictable_bank_idx ) ) ) {
4184 0 : fd_block_id_ele_t * ele = &ctx->block_id_arr[ evictable_bank_idx ];
4185 0 : report_block_incomplete( ctx, ele->slot, ele->block_id_seen ? &ele->dmr : &ele->latest_mr, fd_banks_bank_query( ctx->banks, evictable_bank_idx ),
4186 0 : FD_EVENT_BLOCK_COMPLETED_DEAD_REASON_NOT_DEAD, FD_EVENT_BLOCK_COMPLETED_ABANDONED_REASON_PRUNED );
4187 0 : }
4188 :
4189 : /* Send a notification to other tiles to drop a reference to the
4190 : evictable bank. The RPC tile is the only tile which holds onto
4191 : non-rooted banks, non-transiently. */
4192 0 : fd_replay_drop_bank_ref_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
4193 0 : fd_sched_block_abandon( ctx->sched, evictable_bank_idx, FD_SCHED_ABANDON_DISCARDED );
4194 0 : msg->bank_idx = evictable_bank_idx;
4195 0 : publish_replay_out( ctx, stem, REPLAY_SIG_DROP_BANK_REF, sizeof(fd_replay_drop_bank_ref_t) );
4196 0 : }
4197 0 : return 1;
4198 0 : }
4199 0 : default:
4200 0 : FD_LOG_CRIT(( "unhandled process_rotor_fec result: %d", res ));
4201 0 : }
4202 0 : break;
4203 0 : }
4204 0 : case IN_KIND_TXSEND: {
4205 0 : process_vote_txn_sent( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
4206 0 : break;
4207 0 : }
4208 0 : case IN_KIND_GOSSIP_OUT: {
4209 0 : FD_TEST( sig==FD_GOSSIP_UPDATE_TAG_WFS_DONE );
4210 0 : if( FD_UNLIKELY( ctx->wfs_complete ) ) break;
4211 0 : ctx->wfs_complete = 1;
4212 0 : maybe_verify_shred_version( ctx );
4213 :
4214 : /* Recalculate next_leader_tickcount relative to now. The
4215 : original value was computed at boot time (in boot_genesis or
4216 : on_snapshot_message). */
4217 0 : ctx->next_leader_tickcount = LONG_MAX;
4218 0 : if( FD_LIKELY( ctx->next_leader_slot!=ULONG_MAX ) ) {
4219 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_ele_query( ctx, &ctx->reset_dmr, ctx->reset_slot );
4220 0 : if( FD_LIKELY( block_id_ele ) ) {
4221 0 : fd_bank_t * reset_bank = fd_banks_bank_query( ctx->banks, fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele ) );
4222 0 : if( FD_LIKELY( reset_bank && reset_bank->bank_seq==block_id_ele->bank_seq && reset_bank->state!=FD_BANK_STATE_PRUNABLE ) ) {
4223 0 : double slot_duration_ticks = (double)reset_bank->f.slot_params.ns_per_slot_adjusted*ctx->tick_per_ns;
4224 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*slot_duration_ticks) + fd_tickcount();
4225 0 : }
4226 0 : }
4227 0 : }
4228 :
4229 0 : FD_LOG_NOTICE(( "Done waiting for supermajority. More than 80 percent of cluster stake has joined." ));
4230 :
4231 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
4232 0 : publish_replay_out( ctx, stem, REPLAY_SIG_WFS_DONE, 0UL );
4233 0 : }
4234 0 : break;
4235 0 : }
4236 0 : case IN_KIND_RPC: {
4237 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, sig );
4238 0 : FD_TEST( bank );
4239 0 : bank->refcnt--;
4240 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for %s", bank->idx, bank->f.slot, bank->refcnt, ctx->in_kind[ in_idx ]==IN_KIND_RPC ? "rpc" : "gui" ));
4241 0 : break;
4242 0 : }
4243 0 : case IN_KIND_SNAPMK:
4244 0 : msg_snapmk( ctx, stem, sig, in_idx, chunk );
4245 0 : break;
4246 0 : case IN_KIND_ADMIN:
4247 0 : msg_admin( ctx, stem, fd_frag_meta_ctl_orig( ctl ), sig );
4248 0 : break;
4249 0 : default:
4250 0 : FD_LOG_ERR(( "unhandled kind %d", ctx->in_kind[ in_idx ] ));
4251 0 : }
4252 :
4253 0 : return 0;
4254 0 : }
4255 :
4256 : #undef PROCESS_FEC_DROP
4257 : #undef PROCESS_FEC_WAIT
4258 : #undef PROCESS_FEC_SKIP
4259 : #undef PROCESS_FEC_OK
4260 :
4261 : static inline fd_replay_out_link_t
4262 : out1( fd_topo_t const * topo,
4263 : fd_topo_tile_t const * tile,
4264 0 : char const * name ) {
4265 0 : ulong idx = ULONG_MAX;
4266 :
4267 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
4268 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
4269 0 : if( !strcmp( link->name, name ) ) {
4270 0 : if( FD_UNLIKELY( idx!=ULONG_MAX ) ) FD_LOG_ERR(( "tile %s:%lu had multiple output links named %s but expected one", tile->name, tile->kind_id, name ));
4271 0 : idx = i;
4272 0 : }
4273 0 : }
4274 :
4275 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) return (fd_replay_out_link_t){ .idx = ULONG_MAX, .mem = NULL, .chunk0 = 0, .wmark = 0, .chunk = 0 };
4276 :
4277 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
4278 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
4279 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, topo->links[ tile->out_link_id[ idx ] ].mtu );
4280 :
4281 0 : return (fd_replay_out_link_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0 };
4282 0 : }
4283 :
4284 : static void
4285 : privileged_init( fd_topo_t const * topo,
4286 0 : fd_topo_tile_t const * tile ) {
4287 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
4288 :
4289 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
4290 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
4291 :
4292 0 : if( FD_UNLIKELY( !strcmp( tile->replay.identity_key_path, "" ) ) ) FD_LOG_ERR(( "identity_key_path not set" ));
4293 :
4294 0 : ctx->identity_pubkey[ 0 ] = *(fd_pubkey_t const *)fd_type_pun_const( fd_keyload_load( tile->replay.identity_key_path, /* pubkey only: */ 1 ) );
4295 0 : ctx->identity_idx = 0UL;
4296 0 : ctx->identity_dirty = 0;
4297 :
4298 0 : ctx->vote_account_staked = 0;
4299 0 : ctx->vote_account_inadmissible = 0;
4300 :
4301 0 : ctx->metrics.voted_slot = ULONG_MAX;
4302 :
4303 0 : ctx->has_vote_account = !!tile->replay.vote_account_path[ 0 ];
4304 0 : if( FD_LIKELY( ctx->has_vote_account ) ) {
4305 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->replay.vote_account_path, ctx->vote_account->uc ) ) ) {
4306 0 : uchar const * vote_key = fd_keyload_load( tile->replay.vote_account_path, /* pubkey only: */ 1 );
4307 0 : fd_memcpy( ctx->vote_account->uc, vote_key, sizeof(fd_pubkey_t) );
4308 0 : }
4309 0 : }
4310 :
4311 0 : ctx->bundle.enabled = tile->replay.bundle.enabled;
4312 0 : if( FD_UNLIKELY( !tile->replay.bundle.vote_account_path[0] ) ) {
4313 0 : ctx->bundle.enabled = 0;
4314 0 : }
4315 :
4316 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
4317 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->replay.bundle.vote_account_path, ctx->bundle.vote_account.uc ) ) ) {
4318 0 : const uchar * vote_key = fd_keyload_load( tile->replay.bundle.vote_account_path, /* pubkey only: */ 1 );
4319 0 : fd_memcpy( ctx->bundle.vote_account.uc, vote_key, 32UL );
4320 0 : }
4321 0 : }
4322 :
4323 0 : FD_TEST( fd_rng_secure( &ctx->rng_seed, sizeof(ctx->rng_seed) ) );
4324 0 : FD_TEST( fd_rng_secure( &ctx->blockhash_seed, sizeof(ulong) ) );
4325 0 : FD_TEST( fd_rng_secure( &ctx->vote_tracker_seed, sizeof(ulong) ) );
4326 0 : FD_TEST( fd_rng_secure( &ctx->ag_block_id_map_seed, sizeof(ulong) ) );
4327 0 : FD_TEST( fd_rng_secure( &ctx->initial_block_id, sizeof(fd_hash_t) ) );
4328 0 : FD_TEST( fd_rng_secure( &ctx->runtime_stack_seed, sizeof(ulong) ) );
4329 :
4330 0 : ctx->store_disk_fd = -1;
4331 0 : ulong store_obj_id = fd_pod_query_ulong( topo->props, "store", ULONG_MAX );
4332 0 : if( FD_LIKELY( store_obj_id!=ULONG_MAX ) ) {
4333 0 : fd_store_t * store = fd_store_join( fd_topo_obj_laddr( topo, store_obj_id ) );
4334 0 : FD_TEST( store && store->magic==FD_STORE_MAGIC );
4335 0 : ctx->store_disk_fd = FD_STORE_FD_RO;
4336 0 : if( FD_UNLIKELY( fcntl( ctx->store_disk_fd, F_GETFD )<0 ) )
4337 0 : FD_LOG_ERR(( "store file descriptor was not inherited (%i-%s)", errno, fd_io_strerror( errno ) ));
4338 0 : }
4339 0 : }
4340 :
4341 : static void
4342 : unprivileged_init( fd_topo_t const * topo,
4343 0 : fd_topo_tile_t const * tile ) {
4344 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
4345 0 : void * block_id_map_mem = NULL;
4346 0 : ulong chain_cnt = fd_ag_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
4347 :
4348 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
4349 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
4350 0 : void * genesis_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_genesis_align(), fd_genesis_footprint( fd_genesis_account_max( tile->replay.genesis_max_message_size ) ) );
4351 0 : void * runtime_stack_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_runtime_stack_align(), fd_runtime_stack_footprint( FD_RUNTIME_MAX_VAT_VOTE_ACCOUNTS, FD_RUNTIME_MAX_STAKED_VOTE_ACCOUNTS, FD_RUNTIME_MAX_STAKE_ACCOUNTS ) );
4352 0 : void * block_id_arr_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
4353 0 : void * dead_idxs_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(ulong), sizeof(ulong) * tile->replay.max_live_slots );
4354 0 : void * fec_chain_mem = tile->replay.report_runtime_diffs ?
4355 0 : FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_hash_t), sizeof(fd_hash_t) * FD_FEC_BLK_MAX * tile->replay.max_live_slots ) : NULL;
4356 0 : void * slot_diffs_mem = tile->replay.report_runtime_diffs ?
4357 0 : FD_SCRATCH_ALLOC_APPEND( l, 8UL, FD_EVENT_RUNTIME_SLOT_DIFFS_FOOTPRINT * tile->replay.max_live_slots ) : NULL;
4358 0 : block_id_map_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_ag_block_id_map_align(), fd_ag_block_id_map_footprint( chain_cnt ) );
4359 0 : void * _txncache = FD_SCRATCH_ALLOC_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
4360 0 : void * _accdb = FD_SCRATCH_ALLOC_APPEND( l, fd_accdb_align(), fd_accdb_footprint( tile->replay.max_live_slots, 0 ) );
4361 0 : void * recp_stats_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_reception_stats_t), sizeof(fd_reception_stats_t)*tile->replay.max_live_slots );
4362 0 : void * sched_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots, tile->replay.max_shreds_per_block, tile->replay.max_txn_per_slot ) );
4363 0 : void * vote_tracker_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
4364 0 : void * _capture_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
4365 0 : void * dump_proto_ctx_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
4366 0 : void * block_completed_ev = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_event_block_completed_t), sizeof(fd_event_block_completed_t) );
4367 0 : void * timing_pool_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_timing_slot_pool_align(), fd_timing_slot_pool_footprint( FD_REPLAY_TXN_TIMING_SLOTS ) );
4368 0 : void * timing_rec_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_txn_timing_t), FD_REPLAY_TXN_TIMING_SLOTS*tile->replay.max_txn_per_slot*sizeof(fd_replay_txn_timing_t) );
4369 0 : void * timing_of_bank_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(ulong), tile->replay.max_live_slots*sizeof(ulong) );
4370 0 : void * block_dump_ctx = NULL;
4371 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
4372 0 : block_dump_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
4373 0 : }
4374 :
4375 0 : ctx->genesis = fd_genesis_new( genesis_mem, fd_genesis_account_max( tile->replay.genesis_max_message_size ) );
4376 0 : FD_TEST( ctx->genesis );
4377 :
4378 0 : ctx->runtime_stack = fd_runtime_stack_join( fd_runtime_stack_new( runtime_stack_mem, FD_RUNTIME_MAX_VAT_VOTE_ACCOUNTS, FD_RUNTIME_MAX_STAKED_VOTE_ACCOUNTS, FD_RUNTIME_MAX_STAKE_ACCOUNTS, ctx->runtime_stack_seed ) );
4379 0 : FD_TEST( ctx->runtime_stack );
4380 :
4381 0 : ctx->wksp = topo->workspaces[ topo->objs[ tile->tile_obj_id ].wksp_id ].wksp;
4382 :
4383 0 : ulong store_obj_id = fd_pod_query_ulong( topo->props, "store", ULONG_MAX );
4384 0 : FD_TEST( store_obj_id!=ULONG_MAX );
4385 0 : ctx->store = fd_store_join( fd_topo_obj_laddr( topo, store_obj_id ) );
4386 0 : FD_TEST( ctx->store );
4387 0 : FD_TEST( fd_store_map_ljoin( ctx->store, ctx->map_join ) );
4388 :
4389 0 : ulong banks_obj_id = fd_pod_query_ulong( topo->props, "banks", ULONG_MAX );
4390 0 : FD_TEST( banks_obj_id!=ULONG_MAX );
4391 :
4392 0 : ctx->banks = fd_banks_join( fd_topo_obj_laddr( topo, banks_obj_id ) );
4393 0 : FD_TEST( ctx->banks );
4394 :
4395 0 : ulong footer_cache_obj_id = fd_pod_query_ulong( topo->props, "footer_cache", ULONG_MAX );
4396 0 : ctx->footer_cache = NULL;
4397 0 : if( footer_cache_obj_id!=ULONG_MAX ) {
4398 0 : ctx->footer_cache = fd_footer_cache_join( fd_topo_obj_laddr( topo, footer_cache_obj_id ) );
4399 0 : FD_TEST( ctx->footer_cache );
4400 0 : }
4401 :
4402 0 : ctx->leader_txn_timing = NULL;
4403 0 : ulong ldr_tt_obj_id = fd_pod_query_ulong( topo->props, "ldr_tt", ULONG_MAX );
4404 0 : if( FD_LIKELY( ldr_tt_obj_id!=ULONG_MAX ) ) ctx->leader_txn_timing = fd_topo_obj_laddr( topo, ldr_tt_obj_id );
4405 :
4406 0 : ulong node_info_obj_id = fd_pod_query_ulong( topo->props, "node_info", ULONG_MAX );
4407 0 : FD_TEST( node_info_obj_id!=ULONG_MAX );
4408 0 : ctx->node_info = fd_node_info_box_join( fd_topo_obj_laddr( topo, node_info_obj_id ) );
4409 0 : FD_TEST( ctx->node_info );
4410 0 : fd_node_info_write_begin( ctx->node_info );
4411 0 : ctx->node_info->info.identity = *ctx->identity_pubkey;
4412 0 : if( FD_LIKELY( ctx->has_vote_account ) ) ctx->node_info->info.vote_account = *ctx->vote_account;
4413 0 : fd_node_info_write_end( ctx->node_info );
4414 :
4415 0 : ulong wait_info_obj_id = fd_pod_query_ulong( topo->props, "wait_info", ULONG_MAX );
4416 0 : if( FD_LIKELY( wait_info_obj_id!=ULONG_MAX ) ) {
4417 0 : ctx->wait_info = fd_wait_info_box_join( fd_topo_obj_laddr( topo, wait_info_obj_id ) );
4418 0 : FD_TEST( ctx->wait_info );
4419 0 : } else {
4420 0 : ctx->wait_info = NULL;
4421 0 : }
4422 :
4423 0 : FD_MGAUGE_SET( REPLAY, BANK_CAPACITY, fd_banks_pool_max_cnt( ctx->banks ) );
4424 :
4425 0 : ctx->delinquent_known = 0;
4426 0 : ctx->delinquent_sample_slot = 0UL;
4427 0 : ctx->delinquent_stake_lamports = 0UL;
4428 0 : ctx->cluster_active_stake_lamports = 0UL;
4429 0 : ctx->ag_last_voted_epoch = ULONG_MAX;
4430 0 : ctx->ag_last_settled = 0UL;
4431 0 : ctx->ag_stamp_since = ULONG_MAX;
4432 0 : ctx->consensus_root_slot = ULONG_MAX;
4433 0 : ctx->consensus_root = ctx->initial_block_id;
4434 0 : ctx->finalized_block_id_lo = ag_block_id( ULONG_MAX, ctx->initial_block_id.uc );
4435 0 : ctx->finalized_block_id_hi = ctx->finalized_block_id_lo;
4436 0 : ctx->notified_root_slot = ULONG_MAX;
4437 0 : ctx->notified_root = ctx->initial_block_id;
4438 0 : ctx->notified_root_bank = NULL;
4439 0 : ctx->published_root_slot = ULONG_MAX;
4440 :
4441 0 : ctx->expected_shred_version = tile->replay.expected_shred_version;
4442 0 : ctx->ipecho_shred_version = 0;
4443 0 : ctx->shred_version = 0;
4444 0 : fd_memcpy( ctx->genesis_path, tile->replay.genesis_path, sizeof(ctx->genesis_path) );
4445 0 : ctx->has_genesis_hash = 0;
4446 0 : ctx->has_cluster_type = 0;
4447 0 : ctx->has_genesis_timestamp = 0;
4448 0 : ctx->has_expected_genesis_timestamp = 0;
4449 0 : ctx->cluster_type = FD_CLUSTER_UNKNOWN;
4450 0 : ctx->hard_fork_cnt = ULONG_MAX;
4451 0 : ctx->has_manifest_block_id = 0;
4452 :
4453 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
4454 0 : if( FD_UNLIKELY( !fd_bundle_crank_gen_init( ctx->bundle.gen,
4455 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_distribution_program_addr,
4456 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_payment_program_addr,
4457 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc,
4458 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc, "NAN", 0UL ) ) ) {
4459 0 : FD_LOG_ERR(( "failed to initialize bundle crank gen" ));
4460 0 : }
4461 0 : }
4462 :
4463 0 : FD_TEST( tile->replay.enable_features_cnt<=sizeof(ctx->enable_features)/sizeof(ctx->enable_features[0]) );
4464 0 : ctx->enable_features_cnt = tile->replay.enable_features_cnt;
4465 0 : for( ulong i=0UL; i<tile->replay.enable_features_cnt; i++ ) {
4466 0 : fd_memcpy( ctx->enable_features[ i ], tile->replay.enable_features[ i ], FD_BASE58_ENCODED_32_SZ );
4467 0 : }
4468 :
4469 0 : ulong progcache_obj_id; FD_TEST( (progcache_obj_id = fd_pod_query_ulong( topo->props, "progcache", ULONG_MAX ) )!=ULONG_MAX );
4470 0 : FD_TEST( fd_progcache_shmem_join( ctx->progcache, fd_topo_obj_laddr( topo, progcache_obj_id ) ) );
4471 :
4472 0 : void * _txncache_shmem = fd_topo_obj_laddr( topo, tile->replay.txncache_obj_id );
4473 0 : fd_txncache_shmem_t * txncache_shmem = fd_txncache_shmem_join( _txncache_shmem );
4474 0 : FD_TEST( txncache_shmem );
4475 0 : ctx->txncache = fd_txncache_join( fd_txncache_new( _txncache, txncache_shmem ) );
4476 0 : FD_TEST( ctx->txncache );
4477 :
4478 0 : void * _accdb_shmem = fd_topo_obj_laddr( topo, tile->replay.accdb_obj_id );
4479 0 : fd_accdb_shmem_t * accdb_shmem = fd_accdb_shmem_join( _accdb_shmem );
4480 0 : FD_TEST( accdb_shmem );
4481 0 : fd_sleep_t * accdb_sleep = topo->sleep_obj_id!=ULONG_MAX ? fd_sleep_join( fd_topo_obj_laddr( topo, topo->sleep_obj_id ) ) : NULL;
4482 0 : ctx->accdb = fd_accdb_join( fd_accdb_new( _accdb, accdb_shmem, FD_ACCDB_FD_RW, 0UL, NULL, accdb_sleep, fd_topo_find_tile( topo, "accdb", 0UL ), 0 ) );
4483 0 : FD_TEST( ctx->accdb );
4484 :
4485 0 : ctx->capture_ctx = NULL;
4486 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
4487 0 : ctx->capture_ctx = fd_capture_ctx_join( fd_capture_ctx_new( _capture_ctx ) );
4488 0 : ctx->capture_ctx->solcap_start_slot = tile->replay.capture_start_slot;
4489 0 : ctx->capture_ctx->capture_solcap = 1;
4490 0 : }
4491 :
4492 0 : ctx->block_completed_event = block_completed_ev;
4493 :
4494 0 : ctx->timing_slot_pool = fd_timing_slot_pool_join( fd_timing_slot_pool_new( timing_pool_mem, FD_REPLAY_TXN_TIMING_SLOTS ) );
4495 0 : FD_TEST( ctx->timing_slot_pool );
4496 0 : ctx->timing_rec = timing_rec_mem;
4497 0 : ctx->timing_slot_of_bank = timing_of_bank_mem;
4498 0 : for( ulong i=0UL; i<tile->replay.max_live_slots; i++ ) ctx->timing_slot_of_bank[ i ] = fd_timing_slot_pool_idx_null( ctx->timing_slot_pool );
4499 :
4500 0 : ctx->dump_proto_ctx = NULL;
4501 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.dump_proto_dir ) ) ) {
4502 0 : ctx->dump_proto_ctx = dump_proto_ctx_mem;
4503 0 : ctx->dump_proto_ctx->dump_proto_output_dir = tile->replay.dump_proto_dir;
4504 0 : if( FD_LIKELY( tile->replay.dump_block_to_pb ) ) {
4505 0 : ctx->dump_proto_ctx->dump_block_to_pb = !!tile->replay.dump_block_to_pb;
4506 0 : }
4507 0 : }
4508 :
4509 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
4510 0 : ctx->block_dump_ctx = fd_block_dump_context_join( fd_block_dump_context_new( block_dump_ctx ) );
4511 0 : } else {
4512 0 : ctx->block_dump_ctx = NULL;
4513 0 : }
4514 :
4515 0 : ctx->is_booted = 0;
4516 :
4517 0 : ctx->tick_per_ns = fd_tempo_tick_per_ns( NULL );
4518 :
4519 0 : fd_clock_tile_init( ctx->clock );
4520 :
4521 0 : ctx->max_txn_per_slot = tile->replay.max_txn_per_slot;
4522 0 : ctx->max_shreds_per_block = tile->replay.max_shreds_per_block;
4523 :
4524 0 : FD_TEST( fd_rng_new( ctx->rng, ctx->rng_seed, 0UL ) );
4525 :
4526 0 : ctx->reception_stats = recp_stats_mem;
4527 0 : ctx->reception_stats_cnt = tile->replay.max_live_slots;
4528 0 : FD_TEST( ctx->reception_stats_cnt );
4529 0 : memset( ctx->reception_stats, 0, sizeof(fd_reception_stats_t)*ctx->reception_stats_cnt );
4530 :
4531 0 : ctx->leader_stats.slot = ULONG_MAX;
4532 0 : ctx->sched = fd_sched_join( fd_sched_new( sched_mem, ctx->rng, tile->replay.sched_depth, tile->replay.max_live_slots, ctx->max_shreds_per_block, ctx->max_txn_per_slot, fd_topo_tile_name_cnt( topo, "execrp" ), 1 ) );
4533 0 : FD_TEST( ctx->sched );
4534 :
4535 0 : ctx->in_cnt = tile->in_cnt;
4536 0 : ctx->execrp_idle_cnt = 0UL;
4537 :
4538 0 : ctx->vote_tracker = fd_vote_tracker_join( fd_vote_tracker_new( vote_tracker_mem, ctx->vote_tracker_seed ) );
4539 0 : FD_TEST( ctx->vote_tracker );
4540 :
4541 0 : ctx->identity_vote_rooted = 0;
4542 :
4543 0 : ctx->wait_for_vote_to_start_leader = tile->replay.wait_for_vote_to_start_leader;
4544 :
4545 0 : ctx->wfs_hash_is_zero = !memcmp( tile->replay.wait_for_supermajority_with_bank_hash.uc, ((fd_hash_t){0}).uc, sizeof(fd_hash_t) );
4546 0 : ctx->wfs_enabled = fd_wfs_configured( tile->replay.wait_for_supermajority_at_slot,
4547 0 : ctx->wfs_hash_is_zero,
4548 0 : (ulong)tile->replay.expected_shred_version );
4549 0 : ctx->wait_for_supermajority_at_slot = tile->replay.wait_for_supermajority_at_slot;
4550 0 : ctx->expected_bank_hash = tile->replay.wait_for_supermajority_with_bank_hash;
4551 0 : ctx->wfs_complete = !ctx->wfs_enabled;
4552 :
4553 : /* With WFS, auto is resolved at boot instead (on_snapshot_message). */
4554 0 : if( FD_UNLIKELY( !ctx->wfs_enabled ) ) {
4555 0 : ctx->wait_for_vote_to_start_leader = wfs_resolve_leader_gate_auto( ctx->wait_for_vote_to_start_leader );
4556 0 : }
4557 :
4558 0 : ctx->mleaders = fd_multi_epoch_leaders_join( fd_multi_epoch_leaders_new( ctx->mleaders_mem ) );
4559 0 : FD_TEST( ctx->mleaders );
4560 0 : ctx->next_leader_query_start = ULONG_MAX;
4561 :
4562 0 : ctx->is_leader = 0;
4563 0 : ctx->drain_rotor_fecs = 0;
4564 0 : ctx->supports_leader = fd_topo_find_tile( topo, "pack", 0UL )!=ULONG_MAX;
4565 0 : ctx->snapmk.active = 0;
4566 0 : ctx->snapmk.supported = fd_topo_find_tile( topo, "snapmk", 0UL )!=ULONG_MAX;
4567 0 : ctx->snapmk.scheduled_at_slot = ULONG_MAX;
4568 0 : ctx->snapmk.full_interval_blocks = tile->replay.full_snapshot_interval_blocks;
4569 0 : ctx->snapmk.next_full_block_height = ULONG_MAX;
4570 0 : ctx->snapmk.incremental_interval_blocks = tile->replay.incremental_snapshot_interval_blocks;
4571 0 : ctx->snapmk.next_incremental_block_height = ULONG_MAX;
4572 0 : ctx->snapmk.base_slot = ULONG_MAX;
4573 0 : ctx->snapmk.snap_finished_full = 0UL;
4574 0 : ctx->snapmk.snap_finished_incr = 0UL;
4575 0 : ctx->snapmk.snap_produced_incr_cnt = 0UL;
4576 0 : if( FD_UNLIKELY( !ctx->snapmk.supported ) ) {
4577 0 : ctx->snapmk.full_interval_blocks = 0UL;
4578 0 : ctx->snapmk.incremental_interval_blocks = 0UL;
4579 0 : }
4580 0 : ctx->reset_slot = 0UL;
4581 0 : ctx->reset_dmr = ctx->initial_block_id;
4582 0 : ctx->epoch_end_slot = 0UL;
4583 0 : ctx->slots_per_epoch = 0UL;
4584 0 : ctx->ns_per_slot = 0UL;
4585 0 : ctx->reset_timestamp_nanos = 0UL;
4586 0 : ctx->next_leader_slot = ULONG_MAX;
4587 0 : ctx->next_leader_tickcount = LONG_MAX;
4588 0 : ctx->highwater_leader_slot = ULONG_MAX;
4589 :
4590 0 : ctx->votor_final->slot = ULONG_MAX;
4591 0 : ctx->votor_parent_ready_slot = 0UL;
4592 0 : for( ulong i=0UL; i<FD_NUM_SLOTS_FOR_REWARD+AG_SLOTS_PER_WINDOW+1UL; i++ ) ctx->votor_reward[ i ].slot = ULONG_MAX;
4593 :
4594 0 : ctx->caught_up = 0;
4595 0 : ctx->catch_up_max_fec_slot = ULONG_MAX;
4596 0 : ctx->catch_up_tip_advance_cnt = 0UL;
4597 0 : ctx->boot_timestamp_nanos = tile->replay.boot_timestamp_nanos;
4598 0 : ctx->leader_bank = NULL;
4599 :
4600 0 : ctx->block_id_len = tile->replay.max_live_slots;
4601 0 : ctx->max_live_slots = tile->replay.max_live_slots;
4602 0 : ctx->block_id_arr = (fd_block_id_ele_t *)block_id_arr_mem;
4603 0 : ctx->dead_idxs = (ulong *)dead_idxs_mem;
4604 :
4605 0 : ctx->fec_chain = (fd_hash_t *)fec_chain_mem;
4606 0 : if( FD_UNLIKELY( slot_diffs_mem ) ) fd_event_runtime_slot_diffs_init( slot_diffs_mem, tile->replay.max_live_slots );
4607 :
4608 0 : ctx->ag_block_id_map = fd_ag_block_id_map_join( fd_ag_block_id_map_new( block_id_map_mem, chain_cnt, ctx->ag_block_id_map_seed ) );
4609 0 : FD_TEST( ctx->ag_block_id_map );
4610 :
4611 0 : for( ulong i=0UL; i<tile->replay.max_live_slots; i++ ) {
4612 0 : ctx->block_id_arr[ i ].block_id_seen = 0;
4613 0 : ctx->block_id_arr[ i ].fec_cnt = 0U;
4614 0 : memset( &ctx->block_id_arr[ i ].block_info, 0, sizeof(ag_block_id_t) );
4615 0 : }
4616 :
4617 0 : ctx->resolv_tile_cnt = fd_topo_tile_name_cnt( topo, "resolv" );
4618 :
4619 0 : ctx->keyswitch = fd_keyswitch_join( fd_topo_obj_laddr( topo, tile->id_keyswitch_obj_id ) );
4620 0 : FD_TEST( ctx->keyswitch );
4621 0 : ctx->halt_replay = 0;
4622 :
4623 0 : FD_TEST( tile->in_cnt<=sizeof(ctx->in)/sizeof(ctx->in[0]) );
4624 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
4625 0 : fd_topo_link_t const * link = &topo->links[ tile->in_link_id[ i ] ];
4626 0 : fd_topo_wksp_t const * link_wksp = &topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ];
4627 :
4628 0 : if( FD_LIKELY( link->dcache ) ) {
4629 0 : ctx->in[ i ].mem = link_wksp->wksp;
4630 0 : ctx->in[ i ].chunk0 = fd_dcache_compact_chunk0( ctx->in[ i ].mem, link->dcache );
4631 0 : ctx->in[ i ].wmark = fd_dcache_compact_wmark ( ctx->in[ i ].mem, link->dcache, link->mtu );
4632 0 : ctx->in[ i ].mtu = link->mtu;
4633 0 : }
4634 :
4635 0 : if( !strcmp( link->name, "genesi_out" ) ) ctx->in_kind[ i ] = IN_KIND_GENESIS;
4636 0 : else if( !strcmp( link->name, "ipecho_out" ) ) ctx->in_kind[ i ] = IN_KIND_IPECHO;
4637 0 : else if( !strcmp( link->name, "snapin_manif" ) ) ctx->in_kind[ i ] = IN_KIND_SNAP;
4638 0 : else if( !strcmp( link->name, "execrp_replay" ) ) ctx->in_kind[ i ] = IN_KIND_EXECRP;
4639 0 : else if( !strcmp( link->name, "poh_replay" ) ) ctx->in_kind[ i ] = IN_KIND_POH;
4640 0 : else if( !strcmp( link->name, "resolv_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RESOLV;
4641 0 : else if( !strcmp( link->name, "shred_out" ) ) ctx->in_kind[ i ] = IN_KIND_REPAIR;
4642 0 : else if( !strcmp( link->name, "repair_out" ) ) ctx->in_kind[ i ] = IN_KIND_REPAIR;
4643 0 : else if( !strcmp( link->name, "txsend_out" ) ) ctx->in_kind[ i ] = IN_KIND_TXSEND;
4644 0 : else if( !strcmp( link->name, "rpc_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RPC;
4645 0 : else if( !strcmp( link->name, "gossip_misc" ) ) ctx->in_kind[ i ] = IN_KIND_GOSSIP_OUT;
4646 0 : else if( !strcmp( link->name, "snapmk_out" ) ) ctx->in_kind[ i ] = IN_KIND_SNAPMK;
4647 0 : else if( !strcmp( link->name, "admin_replay" ) ) ctx->in_kind[ i ] = IN_KIND_ADMIN;
4648 0 : else if( !strcmp( link->name, "votor_out" ) ) ctx->in_kind[ i ] = IN_KIND_VOTOR;
4649 0 : else FD_LOG_ERR(( "unexpected input link name %s", link->name ));
4650 :
4651 0 : if( ctx->in_kind[ i ]==IN_KIND_ADMIN ) {
4652 0 : FD_TEST( ( ctx->admin_out_idx = fd_topo_find_tile_out_link( topo, tile, "replay_admin", 0UL ) )!=ULONG_MAX );
4653 0 : }
4654 0 : }
4655 :
4656 0 : *ctx->epoch_out = out1( topo, tile, "replay_epoch" ); FD_TEST( ctx->epoch_out->idx!=ULONG_MAX );
4657 0 : *ctx->replay_out = out1( topo, tile, "replay_out" ); FD_TEST( ctx->replay_out->idx!=ULONG_MAX );
4658 0 : *ctx->slot_out = out1( topo, tile, "replay_slot" ); FD_TEST( ctx->slot_out->idx!=ULONG_MAX );
4659 0 : *ctx->rotor_out = out1( topo, tile, "replay_rotor" );
4660 0 : *ctx->snapmk_out = out1( topo, tile, "replay_snapmk" ); FD_TEST( ctx->snapmk.supported == (ctx->snapmk_out->idx!=ULONG_MAX) );
4661 :
4662 0 : ctx->exec_cnt = 0UL;
4663 0 : for( ulong i=0UL; i<FD_SCHED_MAX_EXEC_TILE_CNT; i++ ) ctx->exec_out[ i ].idx = ULONG_MAX;
4664 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
4665 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
4666 0 : if( strcmp( link->name, "replay_execrp" ) ) continue;
4667 0 : FD_TEST( link->kind_id<FD_SCHED_MAX_EXEC_TILE_CNT && ctx->exec_out[ link->kind_id ].idx==ULONG_MAX );
4668 0 : void * mem = topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ].wksp;
4669 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, link->dcache );
4670 0 : ctx->exec_out[ link->kind_id ] = (fd_replay_out_link_t){
4671 0 : .idx = i,
4672 0 : .mem = mem,
4673 0 : .chunk0 = chunk0,
4674 0 : .wmark = fd_dcache_compact_wmark( mem, link->dcache, link->mtu ),
4675 0 : .chunk = chunk0,
4676 0 : };
4677 0 : ctx->exec_cnt++;
4678 0 : }
4679 0 : FD_TEST( ctx->exec_cnt && ctx->exec_cnt==fd_topo_tile_name_cnt( topo, "execrp" ) );
4680 0 : for( ulong i=0UL; i<ctx->exec_cnt; i++ ) FD_TEST( ctx->exec_out[ i ].idx!=ULONG_MAX );
4681 :
4682 0 : ctx->slot_out_seq = fd_mcache_seq_laddr_const( topo->links[ tile->out_link_id[ ctx->slot_out->idx ] ].mcache );
4683 :
4684 0 : ctx->rpc_enabled = fd_topo_find_tile( topo, "rpc", 0UL )!=ULONG_MAX;
4685 :
4686 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
4687 0 : ulong idx = fd_topo_find_tile_out_link( topo, tile, "cap_repl", 0UL );
4688 0 : FD_TEST( idx!=ULONG_MAX );
4689 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ idx ] ];
4690 :
4691 :
4692 0 : fd_capture_link_buf_t * cap_repl_out = ctx->cap_repl_out;
4693 0 : cap_repl_out->base.vt = &fd_capture_link_buf_vt;
4694 0 : cap_repl_out->idx = idx;
4695 0 : cap_repl_out->mem = topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ].wksp;
4696 0 : cap_repl_out->chunk0 = fd_dcache_compact_chunk0( cap_repl_out->mem, link->dcache );
4697 0 : cap_repl_out->wmark = fd_dcache_compact_wmark( cap_repl_out->mem, link->dcache, link->mtu );
4698 0 : cap_repl_out->chunk = cap_repl_out->chunk0;
4699 0 : cap_repl_out->mcache = link->mcache;
4700 0 : cap_repl_out->depth = fd_mcache_depth( link->mcache );
4701 0 : cap_repl_out->seq = 0UL;
4702 :
4703 0 : ctx->capture_ctx->capctx_type.buf = cap_repl_out;
4704 0 : ctx->capture_ctx->capture_link = &cap_repl_out->base;
4705 0 : ctx->capture_ctx->current_txn_idx = 0UL;
4706 :
4707 :
4708 0 : ulong consumer_tile_idx = fd_topo_find_tile( topo, "solcap", 0UL );
4709 0 : fd_topo_tile_t const * consumer_tile = &topo->tiles[ consumer_tile_idx ];
4710 0 : cap_repl_out->fseq = NULL;
4711 0 : for( ulong j = 0UL; j < consumer_tile->in_cnt; j++ ) {
4712 0 : if( FD_UNLIKELY( consumer_tile->in_link_id[ j ] == link->id ) ) {
4713 0 : cap_repl_out->fseq = fd_fseq_join( fd_topo_obj_laddr( topo, consumer_tile->in_link_fseq_obj_id[ j ] ) );
4714 0 : FD_TEST( cap_repl_out->fseq );
4715 0 : break;
4716 0 : }
4717 0 : }
4718 0 : }
4719 :
4720 0 : fd_memset( &ctx->metrics, 0, sizeof(ctx->metrics) );
4721 :
4722 0 : fd_histf_join( fd_metrics_histf_new_ns( ctx->metrics.store_query_work, FD_MHIST_EDGES_NS( REPLAY, STORE_QUERY_DURATION_NANOS ) ) );
4723 :
4724 : /* Ensure precompiles are available, crash fast otherwise */
4725 0 : fd_precompiles();
4726 :
4727 0 : ctx->report_runtime_diffs = tile->replay.report_runtime_diffs;
4728 0 : ctx->banks->report_runtime_diffs = tile->replay.report_runtime_diffs;
4729 :
4730 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, scratch_align() );
4731 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
4732 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
4733 0 : }
4734 :
4735 : static ulong
4736 : populate_allowed_seccomp( fd_topo_t const * topo,
4737 : fd_topo_tile_t const * tile,
4738 : ulong out_cnt,
4739 0 : struct sock_filter * out ) {
4740 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
4741 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
4742 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
4743 0 : populate_sock_filter_policy_fd_replay_tile( out_cnt, out, (uint)fd_log_private_logfile_fd(), FD_ACCDB_FD_RW, (uint)ctx->store_disk_fd, FD_STAKE_DELEGATIONS_FD, FD_EPOCH_CREDITS_FD, FD_COST_TRACKER_FD, FD_VOTE_STAKES_FD );
4744 0 : return sock_filter_policy_fd_replay_tile_instr_cnt;
4745 0 : }
4746 :
4747 : static ulong
4748 : populate_allowed_fds( fd_topo_t const * topo,
4749 : fd_topo_tile_t const * tile,
4750 : ulong out_fds_cnt,
4751 0 : int * out_fds ) {
4752 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
4753 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
4754 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
4755 0 : if( FD_UNLIKELY( out_fds_cnt<8UL ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
4756 :
4757 0 : ulong out_cnt = 0UL;
4758 0 : out_fds[ out_cnt++ ] = 2; /* stderr */
4759 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) )
4760 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
4761 0 : out_fds[ out_cnt++ ] = FD_ACCDB_FD_RW; /* accounts db */
4762 0 : out_fds[ out_cnt++ ] = FD_STAKE_DELEGATIONS_FD; /* stake delegation disk spill */
4763 0 : out_fds[ out_cnt++ ] = FD_EPOCH_CREDITS_FD; /* epoch credits disk spill */
4764 0 : out_fds[ out_cnt++ ] = FD_COST_TRACKER_FD; /* cost tracker disk spill */
4765 0 : out_fds[ out_cnt++ ] = FD_VOTE_STAKES_FD; /* vote stakes disk spill */
4766 0 : if( FD_LIKELY( ctx->store_disk_fd>=0 ) )
4767 0 : out_fds[ out_cnt++ ] = ctx->store_disk_fd;
4768 :
4769 0 : return out_cnt;
4770 0 : }
4771 :
4772 : static inline void
4773 0 : during_housekeeping( fd_replay_tile_t * ctx ) {
4774 0 : wait_info_publish( ctx );
4775 :
4776 0 : if( FD_UNLIKELY( fd_clock_tile_recal_due( ctx->clock ) ) ) fd_clock_tile_recal( ctx->clock );
4777 :
4778 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_UNHALT_PENDING ) ) {
4779 0 : FD_CHECK_CRIT( ctx->halt_replay, "state machine corruption" );
4780 0 : FD_LOG_DEBUG(( "keyswitch: unhalting replay" ));
4781 0 : ctx->halt_replay = 0;
4782 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
4783 0 : }
4784 :
4785 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) {
4786 0 : if( FD_LIKELY( !ctx->halt_replay ) ) FD_LOG_DEBUG(( "keyswitch: halting replay" ));
4787 0 : ctx->halt_replay = 1;
4788 0 : if( !ctx->is_leader && ctx->is_booted ) maybe_switch_identity( ctx );
4789 0 : }
4790 0 : }
4791 :
4792 : #undef DEBUG_LOGGING
4793 :
4794 : /* Credits are checked against the out link with the fewest, so the
4795 : burst is the most frags one stem iteration publishes to any one out
4796 : link: 4, from boot (slot completed, next leader, root advanced,
4797 : reset) or try_fini_leader chaining into the next leader slot (slot
4798 : completed, next leader, reset, became leader). after_credit and
4799 : returnable_frag never both publish in one iteration: every
4800 : after_credit path that publishes returns nonzero, which skips the
4801 : input poll. */
4802 0 : #define STEM_BURST (4UL)
4803 :
4804 : /* fd_tempo_lazy_default( 16384 ) where 16384 is the minimum out-link
4805 : depth (i.e. cr_max) but excludes replay_epoch and replay_slot, which
4806 : are so infrequent credit availability is a non-issue, and the
4807 : per-tile replay_execrp links, whose credits execrp returns per
4808 : frag. */
4809 0 : #define STEM_LAZY ((long)36865)
4810 :
4811 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_replay_tile_t
4812 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_replay_tile_t)
4813 :
4814 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
4815 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
4816 0 : #define STEM_CALLBACK_BEFORE_FRAG before_frag
4817 0 : #define STEM_CALLBACK_RETURNABLE_FRAG returnable_frag
4818 0 : #define STEM_CALLBACK_DURING_HOUSEKEEPING during_housekeeping
4819 :
4820 : #include "../../disco/stem/fd_stem.c"
4821 :
4822 : static ulong
4823 0 : max_event_sz( fd_topo_tile_t const * tile ) {
4824 : /* replay always emits block_completed; runtime_block (the largest
4825 : runtime event) is added when runtime diffs are on. */
4826 0 : ulong sz = fd_ulong_max( sizeof(fd_event_block_completed_t), sizeof(fd_event_metrics_trigger_t) );
4827 0 : if( tile->replay.report_runtime_diffs && sizeof(fd_event_runtime_block_t)>sz ) sz = sizeof(fd_event_runtime_block_t);
4828 0 : return sz;
4829 0 : }
4830 :
4831 : fd_topo_run_tile_t fd_tile_replay = {
4832 : .name = "replay",
4833 : .max_event_sz = max_event_sz,
4834 : .populate_allowed_seccomp = populate_allowed_seccomp,
4835 : .populate_allowed_fds = populate_allowed_fds,
4836 : .scratch_align = scratch_align,
4837 : .scratch_footprint = scratch_footprint,
4838 : .privileged_init = privileged_init,
4839 : .unprivileged_init = unprivileged_init,
4840 : .run = stem_run,
4841 : };
|
