/src/postgres/src/backend/statistics/extended_stats_funcs.c
Line | Count | Source |
1 | | /*------------------------------------------------------------------------- |
2 | | * |
3 | | * extended_stats_funcs.c |
4 | | * Functions for manipulating extended statistics. |
5 | | * |
6 | | * This file includes the set of facilities required to support the direct |
7 | | * manipulations of extended statistics objects. |
8 | | * |
9 | | * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group |
10 | | * Portions Copyright (c) 1994, Regents of the University of California |
11 | | * |
12 | | * IDENTIFICATION |
13 | | * src/backend/statistics/extended_stats_funcs.c |
14 | | * |
15 | | *------------------------------------------------------------------------- |
16 | | */ |
17 | | #include "postgres.h" |
18 | | |
19 | | #include "access/heapam.h" |
20 | | #include "catalog/indexing.h" |
21 | | #include "catalog/namespace.h" |
22 | | #include "catalog/pg_collation_d.h" |
23 | | #include "catalog/pg_database.h" |
24 | | #include "catalog/pg_operator.h" |
25 | | #include "catalog/pg_statistic_ext.h" |
26 | | #include "catalog/pg_statistic_ext_data.h" |
27 | | #include "miscadmin.h" |
28 | | #include "nodes/makefuncs.h" |
29 | | #include "nodes/nodeFuncs.h" |
30 | | #include "optimizer/optimizer.h" |
31 | | #include "statistics/extended_stats_internal.h" |
32 | | #include "statistics/stat_utils.h" |
33 | | #include "utils/acl.h" |
34 | | #include "utils/array.h" |
35 | | #include "utils/builtins.h" |
36 | | #include "utils/fmgroids.h" |
37 | | #include "utils/jsonb.h" |
38 | | #include "utils/lsyscache.h" |
39 | | #include "utils/syscache.h" |
40 | | #include "utils/typcache.h" |
41 | | |
42 | | |
43 | | /* |
44 | | * Index of the arguments for the SQL functions. |
45 | | */ |
46 | | enum extended_stats_argnum |
47 | | { |
48 | | RELSCHEMA_ARG = 0, |
49 | | RELNAME_ARG, |
50 | | STATSCHEMA_ARG, |
51 | | STATNAME_ARG, |
52 | | INHERITED_ARG, |
53 | | NDISTINCT_ARG, |
54 | | DEPENDENCIES_ARG, |
55 | | MOST_COMMON_VALS_ARG, |
56 | | MOST_COMMON_FREQS_ARG, |
57 | | MOST_COMMON_BASE_FREQS_ARG, |
58 | | EXPRESSIONS_ARG, |
59 | | NUM_EXTENDED_STATS_ARGS, |
60 | | }; |
61 | | |
62 | | /* |
63 | | * The argument names and type OIDs of the arguments for the SQL |
64 | | * functions. |
65 | | */ |
66 | | static struct StatsArgInfo extarginfo[] = |
67 | | { |
68 | | [RELSCHEMA_ARG] = {"schemaname", TEXTOID}, |
69 | | [RELNAME_ARG] = {"relname", TEXTOID}, |
70 | | [STATSCHEMA_ARG] = {"statistics_schemaname", TEXTOID}, |
71 | | [STATNAME_ARG] = {"statistics_name", TEXTOID}, |
72 | | [INHERITED_ARG] = {"inherited", BOOLOID}, |
73 | | [NDISTINCT_ARG] = {"n_distinct", PG_NDISTINCTOID}, |
74 | | [DEPENDENCIES_ARG] = {"dependencies", PG_DEPENDENCIESOID}, |
75 | | [MOST_COMMON_VALS_ARG] = {"most_common_vals", TEXTARRAYOID}, |
76 | | [MOST_COMMON_FREQS_ARG] = {"most_common_freqs", FLOAT8ARRAYOID}, |
77 | | [MOST_COMMON_BASE_FREQS_ARG] = {"most_common_base_freqs", FLOAT8ARRAYOID}, |
78 | | [EXPRESSIONS_ARG] = {"exprs", JSONBOID}, |
79 | | [NUM_EXTENDED_STATS_ARGS] = {0}, |
80 | | }; |
81 | | |
82 | | /* |
83 | | * An index of the elements of a stxdexpr Datum, which repeat for each |
84 | | * expression in the extended statistics object. |
85 | | */ |
86 | | enum extended_stats_exprs_element |
87 | | { |
88 | | NULL_FRAC_ELEM = 0, |
89 | | AVG_WIDTH_ELEM, |
90 | | N_DISTINCT_ELEM, |
91 | | MOST_COMMON_VALS_ELEM, |
92 | | MOST_COMMON_FREQS_ELEM, |
93 | | HISTOGRAM_BOUNDS_ELEM, |
94 | | CORRELATION_ELEM, |
95 | | MOST_COMMON_ELEMS_ELEM, |
96 | | MOST_COMMON_ELEM_FREQS_ELEM, |
97 | | ELEM_COUNT_HISTOGRAM_ELEM, |
98 | | RANGE_LENGTH_HISTOGRAM_ELEM, |
99 | | RANGE_EMPTY_FRAC_ELEM, |
100 | | RANGE_BOUNDS_HISTOGRAM_ELEM, |
101 | | NUM_ATTRIBUTE_STATS_ELEMS |
102 | | }; |
103 | | |
104 | | /* |
105 | | * The argument names of the repeating arguments for stxdexpr. |
106 | | */ |
107 | | static const char *extexprargname[NUM_ATTRIBUTE_STATS_ELEMS] = |
108 | | { |
109 | | "null_frac", |
110 | | "avg_width", |
111 | | "n_distinct", |
112 | | "most_common_vals", |
113 | | "most_common_freqs", |
114 | | "histogram_bounds", |
115 | | "correlation", |
116 | | "most_common_elems", |
117 | | "most_common_elem_freqs", |
118 | | "elem_count_histogram", |
119 | | "range_length_histogram", |
120 | | "range_empty_frac", |
121 | | "range_bounds_histogram" |
122 | | }; |
123 | | |
124 | | static bool extended_statistics_update(FunctionCallInfo fcinfo); |
125 | | |
126 | | static HeapTuple get_pg_statistic_ext(Relation pg_stext, Oid nspoid, |
127 | | const char *stxname); |
128 | | static bool delete_pg_statistic_ext_data(Oid stxoid, bool inherited); |
129 | | |
130 | | /* |
131 | | * Track the extended statistics kinds expected for a pg_statistic_ext |
132 | | * tuple. |
133 | | */ |
134 | | typedef struct |
135 | | { |
136 | | bool ndistinct; |
137 | | bool dependencies; |
138 | | bool mcv; |
139 | | bool expressions; |
140 | | } StakindFlags; |
141 | | |
142 | | static void expand_stxkind(HeapTuple tup, StakindFlags *enabled); |
143 | | static void upsert_pg_statistic_ext_data(const Datum *values, |
144 | | const bool *nulls, |
145 | | const bool *replaces); |
146 | | |
147 | | static bool check_mcvlist_array(const ArrayType *arr, int argindex, |
148 | | int required_ndims, int mcv_length); |
149 | | static Datum import_expressions(Relation pgsd, int numexprs, |
150 | | Oid *atttypids, int32 *atttypmods, |
151 | | Oid *atttypcolls, Jsonb *exprs_jsonb, |
152 | | bool *exprs_is_perfect); |
153 | | static Datum import_mcv(const ArrayType *mcv_arr, |
154 | | const ArrayType *freqs_arr, |
155 | | const ArrayType *base_freqs_arr, |
156 | | Oid *atttypids, int32 *atttypmods, |
157 | | Oid *atttypcolls, int numattrs, |
158 | | bool *ok); |
159 | | |
160 | | static char *jbv_string_get_cstr(JsonbValue *jval); |
161 | | static bool jbv_to_infunc_datum(JsonbValue *jval, PGFunction func, |
162 | | AttrNumber exprnum, const char *argname, |
163 | | Datum *datum); |
164 | | static bool key_in_expr_argnames(JsonbValue *key); |
165 | | static bool check_all_expr_argnames_valid(JsonbContainer *cont, AttrNumber exprnum); |
166 | | static Datum array_in_safe(FmgrInfo *array_in, const char *s, Oid typid, |
167 | | int32 typmod, AttrNumber exprnum, |
168 | | const char *element_name, bool *ok); |
169 | | static Datum import_pg_statistic(Relation pgsd, JsonbContainer *cont, |
170 | | AttrNumber exprnum, FmgrInfo *array_in_fn, |
171 | | Oid typid, int32 typmod, Oid typcoll, |
172 | | bool *pg_statistic_ok); |
173 | | |
174 | | /* |
175 | | * Fetch a pg_statistic_ext row by name and namespace OID. |
176 | | */ |
177 | | static HeapTuple |
178 | | get_pg_statistic_ext(Relation pg_stext, Oid nspoid, const char *stxname) |
179 | 0 | { |
180 | 0 | ScanKeyData key[2]; |
181 | 0 | SysScanDesc scan; |
182 | 0 | HeapTuple tup; |
183 | 0 | Oid stxoid = InvalidOid; |
184 | |
|
185 | 0 | ScanKeyInit(&key[0], |
186 | 0 | Anum_pg_statistic_ext_stxname, |
187 | 0 | BTEqualStrategyNumber, |
188 | 0 | F_NAMEEQ, |
189 | 0 | CStringGetDatum(stxname)); |
190 | 0 | ScanKeyInit(&key[1], |
191 | 0 | Anum_pg_statistic_ext_stxnamespace, |
192 | 0 | BTEqualStrategyNumber, |
193 | 0 | F_OIDEQ, |
194 | 0 | ObjectIdGetDatum(nspoid)); |
195 | | |
196 | | /* |
197 | | * Try to find matching pg_statistic_ext row. |
198 | | */ |
199 | 0 | scan = systable_beginscan(pg_stext, |
200 | 0 | StatisticExtNameIndexId, |
201 | 0 | true, |
202 | 0 | NULL, |
203 | 0 | 2, |
204 | 0 | key); |
205 | | |
206 | | /* Lookup is based on a unique index, so we get either 0 or 1 tuple. */ |
207 | 0 | tup = systable_getnext(scan); |
208 | |
|
209 | 0 | if (HeapTupleIsValid(tup)) |
210 | 0 | stxoid = ((Form_pg_statistic_ext) GETSTRUCT(tup))->oid; |
211 | |
|
212 | 0 | systable_endscan(scan); |
213 | |
|
214 | 0 | if (!OidIsValid(stxoid)) |
215 | 0 | return NULL; |
216 | | |
217 | 0 | return SearchSysCacheCopy1(STATEXTOID, ObjectIdGetDatum(stxoid)); |
218 | 0 | } |
219 | | |
220 | | /* |
221 | | * Decode the stxkind column so that we know which stats types to expect, |
222 | | * returning a StakindFlags set depending on the stats kinds expected by |
223 | | * a pg_statistic_ext tuple. |
224 | | */ |
225 | | static void |
226 | | expand_stxkind(HeapTuple tup, StakindFlags *enabled) |
227 | 0 | { |
228 | 0 | Datum datum; |
229 | 0 | ArrayType *arr; |
230 | 0 | char *kinds; |
231 | |
|
232 | 0 | datum = SysCacheGetAttrNotNull(STATEXTOID, |
233 | 0 | tup, |
234 | 0 | Anum_pg_statistic_ext_stxkind); |
235 | 0 | arr = DatumGetArrayTypeP(datum); |
236 | 0 | if (ARR_NDIM(arr) != 1 || ARR_HASNULL(arr) || ARR_ELEMTYPE(arr) != CHAROID) |
237 | 0 | elog(ERROR, "stxkind is not a one-dimension char array"); |
238 | | |
239 | 0 | kinds = (char *) ARR_DATA_PTR(arr); |
240 | |
|
241 | 0 | for (int i = 0; i < ARR_DIMS(arr)[0]; i++) |
242 | 0 | { |
243 | 0 | switch (kinds[i]) |
244 | 0 | { |
245 | 0 | case STATS_EXT_NDISTINCT: |
246 | 0 | enabled->ndistinct = true; |
247 | 0 | break; |
248 | 0 | case STATS_EXT_DEPENDENCIES: |
249 | 0 | enabled->dependencies = true; |
250 | 0 | break; |
251 | 0 | case STATS_EXT_MCV: |
252 | 0 | enabled->mcv = true; |
253 | 0 | break; |
254 | 0 | case STATS_EXT_EXPRESSIONS: |
255 | 0 | enabled->expressions = true; |
256 | 0 | break; |
257 | 0 | default: |
258 | 0 | elog(ERROR, "incorrect stxkind %c found", kinds[i]); |
259 | 0 | break; |
260 | 0 | } |
261 | 0 | } |
262 | 0 | } |
263 | | |
264 | | /* |
265 | | * Perform the actual storage of a pg_statistic_ext_data tuple. |
266 | | */ |
267 | | static void |
268 | | upsert_pg_statistic_ext_data(const Datum *values, const bool *nulls, |
269 | | const bool *replaces) |
270 | 0 | { |
271 | 0 | Relation pg_stextdata; |
272 | 0 | HeapTuple stxdtup; |
273 | 0 | HeapTuple newtup; |
274 | |
|
275 | 0 | pg_stextdata = table_open(StatisticExtDataRelationId, RowExclusiveLock); |
276 | |
|
277 | 0 | stxdtup = SearchSysCache2(STATEXTDATASTXOID, |
278 | 0 | values[Anum_pg_statistic_ext_data_stxoid - 1], |
279 | 0 | values[Anum_pg_statistic_ext_data_stxdinherit - 1]); |
280 | |
|
281 | 0 | if (HeapTupleIsValid(stxdtup)) |
282 | 0 | { |
283 | 0 | newtup = heap_modify_tuple(stxdtup, |
284 | 0 | RelationGetDescr(pg_stextdata), |
285 | 0 | values, |
286 | 0 | nulls, |
287 | 0 | replaces); |
288 | 0 | CatalogTupleUpdate(pg_stextdata, &newtup->t_self, newtup); |
289 | 0 | ReleaseSysCache(stxdtup); |
290 | 0 | } |
291 | 0 | else |
292 | 0 | { |
293 | 0 | newtup = heap_form_tuple(RelationGetDescr(pg_stextdata), values, nulls); |
294 | 0 | CatalogTupleInsert(pg_stextdata, newtup); |
295 | 0 | } |
296 | |
|
297 | 0 | heap_freetuple(newtup); |
298 | |
|
299 | 0 | CommandCounterIncrement(); |
300 | |
|
301 | 0 | table_close(pg_stextdata, RowExclusiveLock); |
302 | 0 | } |
303 | | |
304 | | /* |
305 | | * Insert or update an extended statistics object. |
306 | | * |
307 | | * Major errors, such as the table not existing or permission errors, are |
308 | | * reported as ERRORs. There are a couple of paths that generate a WARNING, |
309 | | * like when the statistics object or its schema do not exist, a conversion |
310 | | * failure on one statistic kind, or when other statistic kinds may still |
311 | | * be updated. |
312 | | */ |
313 | | static bool |
314 | | extended_statistics_update(FunctionCallInfo fcinfo) |
315 | 0 | { |
316 | 0 | char *relnspname; |
317 | 0 | char *relname; |
318 | 0 | Oid nspoid; |
319 | 0 | char *nspname; |
320 | 0 | char *stxname; |
321 | 0 | bool inherited; |
322 | 0 | Relation pg_stext = NULL; |
323 | 0 | HeapTuple tup = NULL; |
324 | |
|
325 | 0 | StakindFlags enabled = {false, false, false, false}; |
326 | 0 | StakindFlags has = {false, false, false, false}; |
327 | |
|
328 | 0 | Form_pg_statistic_ext stxform; |
329 | |
|
330 | 0 | Datum values[Natts_pg_statistic_ext_data] = {0}; |
331 | 0 | bool nulls[Natts_pg_statistic_ext_data] = {0}; |
332 | 0 | bool replaces[Natts_pg_statistic_ext_data] = {0}; |
333 | 0 | bool success = true; |
334 | 0 | Datum exprdatum; |
335 | 0 | bool isnull; |
336 | 0 | List *exprs = NIL; |
337 | 0 | int numattnums = 0; |
338 | 0 | int numexprs = 0; |
339 | 0 | int numattrs = 0; |
340 | | |
341 | | /* arrays of type info, if we need them */ |
342 | 0 | Oid *atttypids = NULL; |
343 | 0 | int32 *atttypmods = NULL; |
344 | 0 | Oid *atttypcolls = NULL; |
345 | 0 | Oid relid; |
346 | 0 | Oid locked_table = InvalidOid; |
347 | | |
348 | | /* |
349 | | * Fill out the StakindFlags "has" structure based on which parameters |
350 | | * were provided to the function. |
351 | | * |
352 | | * The MCV stats composite value is an array of record type, but this is |
353 | | * externally represented as three arrays that must be interleaved into |
354 | | * the array of records (pg_stats_ext stores four arrays, |
355 | | * most_common_val_nulls is built from the contents of most_common_vals). |
356 | | * Therefore, none of the three array values is meaningful unless the |
357 | | * other two are also present and in sync in terms of array length. |
358 | | */ |
359 | 0 | has.mcv = (!PG_ARGISNULL(MOST_COMMON_VALS_ARG) && |
360 | 0 | !PG_ARGISNULL(MOST_COMMON_FREQS_ARG) && |
361 | 0 | !PG_ARGISNULL(MOST_COMMON_BASE_FREQS_ARG)); |
362 | 0 | has.ndistinct = !PG_ARGISNULL(NDISTINCT_ARG); |
363 | 0 | has.dependencies = !PG_ARGISNULL(DEPENDENCIES_ARG); |
364 | 0 | has.expressions = !PG_ARGISNULL(EXPRESSIONS_ARG); |
365 | |
|
366 | 0 | if (RecoveryInProgress()) |
367 | 0 | { |
368 | 0 | ereport(WARNING, |
369 | 0 | errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
370 | 0 | errmsg("recovery is in progress"), |
371 | 0 | errhint("Statistics cannot be modified during recovery.")); |
372 | 0 | return false; |
373 | 0 | } |
374 | | |
375 | | /* relation arguments */ |
376 | 0 | stats_check_required_arg(fcinfo, extarginfo, RELSCHEMA_ARG); |
377 | 0 | relnspname = TextDatumGetCString(PG_GETARG_DATUM(RELSCHEMA_ARG)); |
378 | 0 | stats_check_required_arg(fcinfo, extarginfo, RELNAME_ARG); |
379 | 0 | relname = TextDatumGetCString(PG_GETARG_DATUM(RELNAME_ARG)); |
380 | | |
381 | | /* extended statistics arguments */ |
382 | 0 | stats_check_required_arg(fcinfo, extarginfo, STATSCHEMA_ARG); |
383 | 0 | nspname = TextDatumGetCString(PG_GETARG_DATUM(STATSCHEMA_ARG)); |
384 | 0 | stats_check_required_arg(fcinfo, extarginfo, STATNAME_ARG); |
385 | 0 | stxname = TextDatumGetCString(PG_GETARG_DATUM(STATNAME_ARG)); |
386 | 0 | stats_check_required_arg(fcinfo, extarginfo, INHERITED_ARG); |
387 | 0 | inherited = PG_GETARG_BOOL(INHERITED_ARG); |
388 | | |
389 | | /* |
390 | | * First open the relation where we expect to find the statistics. This |
391 | | * is similar to relation and attribute statistics, so as ACL checks are |
392 | | * done before any locks are taken, even before any attempts related to |
393 | | * the extended stats object. |
394 | | */ |
395 | 0 | relid = RangeVarGetRelidExtended(makeRangeVar(relnspname, relname, -1), |
396 | 0 | ShareUpdateExclusiveLock, 0, |
397 | 0 | RangeVarCallbackForStats, &locked_table); |
398 | |
|
399 | 0 | nspoid = get_namespace_oid(nspname, true); |
400 | 0 | if (nspoid == InvalidOid) |
401 | 0 | { |
402 | 0 | ereport(WARNING, |
403 | 0 | errcode(ERRCODE_UNDEFINED_OBJECT), |
404 | 0 | errmsg("could not find schema \"%s\"", nspname)); |
405 | 0 | success = false; |
406 | 0 | goto cleanup; |
407 | 0 | } |
408 | | |
409 | 0 | pg_stext = table_open(StatisticExtRelationId, RowExclusiveLock); |
410 | 0 | tup = get_pg_statistic_ext(pg_stext, nspoid, stxname); |
411 | |
|
412 | 0 | if (!HeapTupleIsValid(tup)) |
413 | 0 | { |
414 | 0 | ereport(WARNING, |
415 | 0 | errcode(ERRCODE_UNDEFINED_OBJECT), |
416 | 0 | errmsg("could not find extended statistics object \"%s.%s\"", |
417 | 0 | nspname, stxname)); |
418 | 0 | success = false; |
419 | 0 | goto cleanup; |
420 | 0 | } |
421 | | |
422 | 0 | stxform = (Form_pg_statistic_ext) GETSTRUCT(tup); |
423 | | |
424 | | /* |
425 | | * The relation tracked by the stats object has to match with the relation |
426 | | * we have already locked. |
427 | | */ |
428 | 0 | if (stxform->stxrelid != relid) |
429 | 0 | { |
430 | 0 | ereport(WARNING, |
431 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
432 | 0 | errmsg("could not restore extended statistics object \"%s.%s\": incorrect relation \"%s.%s\" specified", |
433 | 0 | nspname, stxname, |
434 | 0 | relnspname, relname)); |
435 | | |
436 | 0 | success = false; |
437 | 0 | goto cleanup; |
438 | 0 | } |
439 | | |
440 | | /* Find out what extended statistics kinds we should expect. */ |
441 | 0 | expand_stxkind(tup, &enabled); |
442 | 0 | numattnums = stxform->stxkeys.dim1; |
443 | | |
444 | | /* decode expression (if any) */ |
445 | 0 | exprdatum = SysCacheGetAttr(STATEXTOID, |
446 | 0 | tup, |
447 | 0 | Anum_pg_statistic_ext_stxexprs, |
448 | 0 | &isnull); |
449 | 0 | if (!isnull) |
450 | 0 | { |
451 | 0 | char *s; |
452 | |
|
453 | 0 | s = TextDatumGetCString(exprdatum); |
454 | 0 | exprs = (List *) stringToNode(s); |
455 | 0 | pfree(s); |
456 | | |
457 | | /* |
458 | | * Run the expressions through eval_const_expressions(). This is not |
459 | | * just an optimization, but is necessary, because the planner will be |
460 | | * comparing them to similarly-processed qual clauses, and may fail to |
461 | | * detect valid matches without this. |
462 | | * |
463 | | * We must not use canonicalize_qual(), however, since these are not |
464 | | * qual expressions. |
465 | | */ |
466 | 0 | exprs = (List *) eval_const_expressions(NULL, (Node *) exprs); |
467 | | |
468 | | /* May as well fix opfuncids too */ |
469 | 0 | fix_opfuncids((Node *) exprs); |
470 | | |
471 | | /* Compute the number of expression, for input validation. */ |
472 | 0 | numexprs = list_length(exprs); |
473 | 0 | } |
474 | |
|
475 | 0 | numattrs = numattnums + numexprs; |
476 | | |
477 | | /* |
478 | | * If the object cannot support ndistinct, we should not have data for it. |
479 | | */ |
480 | 0 | if (has.ndistinct && !enabled.ndistinct) |
481 | 0 | { |
482 | 0 | ereport(WARNING, |
483 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
484 | 0 | errmsg("cannot specify parameter \"%s\"", |
485 | 0 | extarginfo[NDISTINCT_ARG].argname), |
486 | 0 | errhint("Extended statistics object \"%s.%s\" does not support statistics of this type.", |
487 | 0 | nspname, stxname)); |
488 | | |
489 | 0 | has.ndistinct = false; |
490 | 0 | success = false; |
491 | 0 | } |
492 | | |
493 | | /* |
494 | | * If the object cannot support dependencies, we should not have data for |
495 | | * it. |
496 | | */ |
497 | 0 | if (has.dependencies && !enabled.dependencies) |
498 | 0 | { |
499 | 0 | ereport(WARNING, |
500 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
501 | 0 | errmsg("cannot specify parameter \"%s\"", |
502 | 0 | extarginfo[DEPENDENCIES_ARG].argname), |
503 | 0 | errhint("Extended statistics object \"%s.%s\" does not support statistics of this type.", |
504 | 0 | nspname, stxname)); |
505 | 0 | has.dependencies = false; |
506 | 0 | success = false; |
507 | 0 | } |
508 | | |
509 | | /* |
510 | | * If the object cannot hold an MCV value, but any of the MCV parameters |
511 | | * are set, then issue a WARNING and ensure that we do not try to load MCV |
512 | | * stats later. In pg_stats_ext, most_common_val_nulls, most_common_freqs |
513 | | * and most_common_base_freqs are NULL if most_common_vals is NULL. |
514 | | */ |
515 | 0 | if (!enabled.mcv) |
516 | 0 | { |
517 | 0 | if (!PG_ARGISNULL(MOST_COMMON_VALS_ARG) || |
518 | 0 | !PG_ARGISNULL(MOST_COMMON_FREQS_ARG) || |
519 | 0 | !PG_ARGISNULL(MOST_COMMON_BASE_FREQS_ARG)) |
520 | 0 | { |
521 | 0 | ereport(WARNING, |
522 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
523 | 0 | errmsg("cannot specify parameters \"%s\", \"%s\", or \"%s\"", |
524 | 0 | extarginfo[MOST_COMMON_VALS_ARG].argname, |
525 | 0 | extarginfo[MOST_COMMON_FREQS_ARG].argname, |
526 | 0 | extarginfo[MOST_COMMON_BASE_FREQS_ARG].argname), |
527 | 0 | errhint("Extended statistics object \"%s.%s\" does not support statistics of this type.", |
528 | 0 | nspname, stxname)); |
529 | | |
530 | 0 | has.mcv = false; |
531 | 0 | success = false; |
532 | 0 | } |
533 | 0 | } |
534 | 0 | else if (!has.mcv) |
535 | 0 | { |
536 | | /* |
537 | | * If we do not have all of the MCV arrays set while the extended |
538 | | * statistics object expects something, something is wrong. This |
539 | | * issues a WARNING if a partial input has been provided. |
540 | | */ |
541 | 0 | if (!PG_ARGISNULL(MOST_COMMON_VALS_ARG) || |
542 | 0 | !PG_ARGISNULL(MOST_COMMON_FREQS_ARG) || |
543 | 0 | !PG_ARGISNULL(MOST_COMMON_BASE_FREQS_ARG)) |
544 | 0 | { |
545 | 0 | ereport(WARNING, |
546 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
547 | 0 | errmsg("could not use \"%s\", \"%s\", and \"%s\": missing one or more parameters", |
548 | 0 | extarginfo[MOST_COMMON_VALS_ARG].argname, |
549 | 0 | extarginfo[MOST_COMMON_FREQS_ARG].argname, |
550 | 0 | extarginfo[MOST_COMMON_BASE_FREQS_ARG].argname)); |
551 | 0 | success = false; |
552 | 0 | } |
553 | 0 | } |
554 | | |
555 | | /* |
556 | | * If the object cannot support expressions, we should not have data for |
557 | | * them. |
558 | | */ |
559 | 0 | if (has.expressions && !enabled.expressions) |
560 | 0 | { |
561 | 0 | ereport(WARNING, |
562 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
563 | 0 | errmsg("cannot specify parameter \"%s\"", |
564 | 0 | extarginfo[EXPRESSIONS_ARG].argname), |
565 | 0 | errhint("Extended statistics object \"%s.%s\" does not support statistics of this type.", |
566 | 0 | nspname, stxname)); |
567 | | |
568 | 0 | has.expressions = false; |
569 | 0 | success = false; |
570 | 0 | } |
571 | | |
572 | | /* |
573 | | * Either of these statistic types requires that we supply a semi-filled |
574 | | * VacAttrStatsP array. |
575 | | * |
576 | | * It is not possible to use the existing lookup_var_attr_stats() and |
577 | | * examine_attribute() because these functions will skip attributes where |
578 | | * attstattarget is 0, and we may have statistics data to import for those |
579 | | * attributes. |
580 | | */ |
581 | 0 | if (has.mcv || has.expressions) |
582 | 0 | { |
583 | 0 | atttypids = palloc0_array(Oid, numattrs); |
584 | 0 | atttypmods = palloc0_array(int32, numattrs); |
585 | 0 | atttypcolls = palloc0_array(Oid, numattrs); |
586 | | |
587 | | /* |
588 | | * The leading stxkeys are attribute numbers up through numattnums. |
589 | | * These keys must be in ascending AttrNumber order, but we do not |
590 | | * rely on that. |
591 | | */ |
592 | 0 | for (int i = 0; i < numattnums; i++) |
593 | 0 | { |
594 | 0 | AttrNumber attnum = stxform->stxkeys.values[i]; |
595 | 0 | HeapTuple atup = SearchSysCache2(ATTNUM, |
596 | 0 | ObjectIdGetDatum(relid), |
597 | 0 | Int16GetDatum(attnum)); |
598 | |
|
599 | 0 | Form_pg_attribute attr; |
600 | | |
601 | | /* Attribute not found */ |
602 | 0 | if (!HeapTupleIsValid(atup)) |
603 | 0 | elog(ERROR, "stxkeys references nonexistent attnum %d", attnum); |
604 | | |
605 | 0 | attr = (Form_pg_attribute) GETSTRUCT(atup); |
606 | |
|
607 | 0 | if (attr->attisdropped) |
608 | 0 | elog(ERROR, "stxkeys references dropped attnum %d", attnum); |
609 | | |
610 | 0 | atttypids[i] = attr->atttypid; |
611 | 0 | atttypmods[i] = attr->atttypmod; |
612 | 0 | atttypcolls[i] = attr->attcollation; |
613 | 0 | ReleaseSysCache(atup); |
614 | 0 | } |
615 | | |
616 | | /* |
617 | | * After all the positive number attnums in stxkeys come the negative |
618 | | * numbers (if any) which represent expressions in the order that they |
619 | | * appear in stxdexpr. Because the expressions are always |
620 | | * monotonically decreasing from -1, there is no point in looking at |
621 | | * the values in stxkeys, it's enough to know how many of them there |
622 | | * are. |
623 | | */ |
624 | 0 | for (int i = numattnums; i < numattrs; i++) |
625 | 0 | { |
626 | 0 | Node *expr = list_nth(exprs, i - numattnums); |
627 | |
|
628 | 0 | atttypids[i] = exprType(expr); |
629 | 0 | atttypmods[i] = exprTypmod(expr); |
630 | 0 | atttypcolls[i] = exprCollation(expr); |
631 | 0 | } |
632 | 0 | } |
633 | | |
634 | | /* |
635 | | * Populate the pg_statistic_ext_data result tuple. |
636 | | */ |
637 | | |
638 | | /* Primary Key: cannot be NULL or replaced. */ |
639 | 0 | values[Anum_pg_statistic_ext_data_stxoid - 1] = ObjectIdGetDatum(stxform->oid); |
640 | 0 | nulls[Anum_pg_statistic_ext_data_stxoid - 1] = false; |
641 | 0 | values[Anum_pg_statistic_ext_data_stxdinherit - 1] = BoolGetDatum(inherited); |
642 | 0 | nulls[Anum_pg_statistic_ext_data_stxdinherit - 1] = false; |
643 | | |
644 | | /* All unspecified parameters will be left unmodified */ |
645 | 0 | nulls[Anum_pg_statistic_ext_data_stxdndistinct - 1] = true; |
646 | 0 | nulls[Anum_pg_statistic_ext_data_stxddependencies - 1] = true; |
647 | 0 | nulls[Anum_pg_statistic_ext_data_stxdmcv - 1] = true; |
648 | 0 | nulls[Anum_pg_statistic_ext_data_stxdexpr - 1] = true; |
649 | | |
650 | | /* |
651 | | * For each stats kind, deserialize the data at hand and perform a round |
652 | | * of validation. The resulting tuple is filled with a set of updated |
653 | | * values. |
654 | | */ |
655 | |
|
656 | 0 | if (has.ndistinct) |
657 | 0 | { |
658 | 0 | Datum ndistinct_datum = PG_GETARG_DATUM(NDISTINCT_ARG); |
659 | 0 | bytea *data = DatumGetByteaPP(ndistinct_datum); |
660 | 0 | MVNDistinct *ndistinct = statext_ndistinct_deserialize(data); |
661 | |
|
662 | 0 | if (statext_ndistinct_validate(ndistinct, &stxform->stxkeys, |
663 | 0 | numexprs, WARNING)) |
664 | 0 | { |
665 | 0 | values[Anum_pg_statistic_ext_data_stxdndistinct - 1] = ndistinct_datum; |
666 | 0 | nulls[Anum_pg_statistic_ext_data_stxdndistinct - 1] = false; |
667 | 0 | replaces[Anum_pg_statistic_ext_data_stxdndistinct - 1] = true; |
668 | 0 | } |
669 | 0 | else |
670 | 0 | success = false; |
671 | |
|
672 | 0 | statext_ndistinct_free(ndistinct); |
673 | 0 | } |
674 | |
|
675 | 0 | if (has.dependencies) |
676 | 0 | { |
677 | 0 | Datum dependencies_datum = PG_GETARG_DATUM(DEPENDENCIES_ARG); |
678 | 0 | bytea *data = DatumGetByteaPP(dependencies_datum); |
679 | 0 | MVDependencies *dependencies = statext_dependencies_deserialize(data); |
680 | |
|
681 | 0 | if (statext_dependencies_validate(dependencies, &stxform->stxkeys, |
682 | 0 | numexprs, WARNING)) |
683 | 0 | { |
684 | 0 | values[Anum_pg_statistic_ext_data_stxddependencies - 1] = dependencies_datum; |
685 | 0 | nulls[Anum_pg_statistic_ext_data_stxddependencies - 1] = false; |
686 | 0 | replaces[Anum_pg_statistic_ext_data_stxddependencies - 1] = true; |
687 | 0 | } |
688 | 0 | else |
689 | 0 | success = false; |
690 | |
|
691 | 0 | statext_dependencies_free(dependencies); |
692 | 0 | } |
693 | |
|
694 | 0 | if (has.mcv) |
695 | 0 | { |
696 | 0 | Datum datum; |
697 | 0 | bool val_ok = false; |
698 | |
|
699 | 0 | datum = import_mcv(PG_GETARG_ARRAYTYPE_P(MOST_COMMON_VALS_ARG), |
700 | 0 | PG_GETARG_ARRAYTYPE_P(MOST_COMMON_FREQS_ARG), |
701 | 0 | PG_GETARG_ARRAYTYPE_P(MOST_COMMON_BASE_FREQS_ARG), |
702 | 0 | atttypids, atttypmods, atttypcolls, numattrs, |
703 | 0 | &val_ok); |
704 | |
|
705 | 0 | if (val_ok) |
706 | 0 | { |
707 | 0 | Assert(datum != (Datum) 0); |
708 | 0 | values[Anum_pg_statistic_ext_data_stxdmcv - 1] = datum; |
709 | 0 | nulls[Anum_pg_statistic_ext_data_stxdmcv - 1] = false; |
710 | 0 | replaces[Anum_pg_statistic_ext_data_stxdmcv - 1] = true; |
711 | 0 | } |
712 | 0 | else |
713 | 0 | success = false; |
714 | 0 | } |
715 | |
|
716 | 0 | if (has.expressions) |
717 | 0 | { |
718 | 0 | Datum datum; |
719 | 0 | Relation pgsd; |
720 | 0 | bool ok = false; |
721 | |
|
722 | 0 | pgsd = table_open(StatisticRelationId, RowExclusiveLock); |
723 | | |
724 | | /* |
725 | | * Generate the expressions array. |
726 | | * |
727 | | * The atttypids, atttypmods, and atttypcolls arrays have all the |
728 | | * regular attributes listed first, so we can pass those arrays with a |
729 | | * start point after the last regular attribute. There are numexprs |
730 | | * elements remaining. |
731 | | */ |
732 | 0 | datum = import_expressions(pgsd, numexprs, |
733 | 0 | &atttypids[numattnums], |
734 | 0 | &atttypmods[numattnums], |
735 | 0 | &atttypcolls[numattnums], |
736 | 0 | PG_GETARG_JSONB_P(EXPRESSIONS_ARG), |
737 | 0 | &ok); |
738 | |
|
739 | 0 | table_close(pgsd, RowExclusiveLock); |
740 | |
|
741 | 0 | if (ok) |
742 | 0 | { |
743 | 0 | Assert(datum != (Datum) 0); |
744 | 0 | values[Anum_pg_statistic_ext_data_stxdexpr - 1] = datum; |
745 | 0 | replaces[Anum_pg_statistic_ext_data_stxdexpr - 1] = true; |
746 | 0 | nulls[Anum_pg_statistic_ext_data_stxdexpr - 1] = false; |
747 | 0 | } |
748 | 0 | else |
749 | 0 | success = false; |
750 | 0 | } |
751 | |
|
752 | 0 | upsert_pg_statistic_ext_data(values, nulls, replaces); |
753 | |
|
754 | 0 | cleanup: |
755 | 0 | if (HeapTupleIsValid(tup)) |
756 | 0 | heap_freetuple(tup); |
757 | 0 | if (pg_stext != NULL) |
758 | 0 | table_close(pg_stext, RowExclusiveLock); |
759 | 0 | if (atttypids != NULL) |
760 | 0 | pfree(atttypids); |
761 | 0 | if (atttypmods != NULL) |
762 | 0 | pfree(atttypmods); |
763 | 0 | if (atttypcolls != NULL) |
764 | 0 | pfree(atttypcolls); |
765 | 0 | return success; |
766 | 0 | } |
767 | | |
768 | | /* |
769 | | * Consistency checks to ensure that other mcvlist arrays are in alignment |
770 | | * with the mcv array. |
771 | | */ |
772 | | static bool |
773 | | check_mcvlist_array(const ArrayType *arr, int argindex, int required_ndims, |
774 | | int mcv_length) |
775 | | { |
776 | | if (ARR_NDIM(arr) != required_ndims) |
777 | | { |
778 | | ereport(WARNING, |
779 | | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
780 | | errmsg("could not parse array \"%s\": incorrect number of dimensions (%d required)", |
781 | | extarginfo[argindex].argname, required_ndims)); |
782 | | return false; |
783 | | } |
784 | | |
785 | | if (array_contains_nulls(arr)) |
786 | | { |
787 | | ereport(WARNING, |
788 | | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
789 | | errmsg("could not parse array \"%s\": NULL value found", |
790 | | extarginfo[argindex].argname)); |
791 | | return false; |
792 | | } |
793 | | |
794 | | if (ARR_DIMS(arr)[0] != mcv_length) |
795 | | { |
796 | | ereport(WARNING, |
797 | | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
798 | | errmsg("could not parse array \"%s\": incorrect number of elements (same as \"%s\" required)", |
799 | | extarginfo[argindex].argname, |
800 | | extarginfo[MOST_COMMON_VALS_ARG].argname)); |
801 | | return false; |
802 | | } |
803 | | |
804 | | return true; |
805 | | } |
806 | | |
807 | | /* |
808 | | * Create the stxdmcv datum from the equal-sized arrays of most common values, |
809 | | * their null flags, and the frequency and base frequency associated with |
810 | | * each value. |
811 | | */ |
812 | | static Datum |
813 | | import_mcv(const ArrayType *mcv_arr, const ArrayType *freqs_arr, |
814 | | const ArrayType *base_freqs_arr, Oid *atttypids, int32 *atttypmods, |
815 | | Oid *atttypcolls, int numattrs, bool *ok) |
816 | 0 | { |
817 | 0 | int nitems; |
818 | 0 | Datum *mcv_elems; |
819 | 0 | bool *mcv_nulls; |
820 | 0 | int check_nummcv; |
821 | 0 | Datum mcv = (Datum) 0; |
822 | |
|
823 | 0 | *ok = false; |
824 | | |
825 | | /* |
826 | | * mcv_arr is an array of arrays. Each inner array must have the same |
827 | | * number of elements "numattrs". |
828 | | */ |
829 | 0 | if (ARR_NDIM(mcv_arr) != 2) |
830 | 0 | { |
831 | 0 | ereport(WARNING, |
832 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
833 | 0 | errmsg("could not parse array \"%s\": incorrect number of dimensions (%d required)", |
834 | 0 | extarginfo[MOST_COMMON_VALS_ARG].argname, 2)); |
835 | 0 | goto mcv_error; |
836 | 0 | } |
837 | | |
838 | 0 | if (ARR_DIMS(mcv_arr)[1] != numattrs) |
839 | 0 | { |
840 | 0 | ereport(WARNING, |
841 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
842 | 0 | errmsg("could not parse array \"%s\": found %d attributes but expected %d", |
843 | 0 | extarginfo[MOST_COMMON_VALS_ARG].argname, |
844 | 0 | ARR_DIMS(mcv_arr)[1], numattrs)); |
845 | 0 | goto mcv_error; |
846 | 0 | } |
847 | | |
848 | | /* |
849 | | * "most_common_freqs" and "most_common_base_freqs" arrays must be of the |
850 | | * same length, one-dimension and cannot contain NULLs. We use mcv_arr as |
851 | | * the reference array for determining their length. |
852 | | */ |
853 | 0 | nitems = ARR_DIMS(mcv_arr)[0]; |
854 | | |
855 | | /* |
856 | | * Reject a MCV list larger than what statext_mcv_deserialize() is able to |
857 | | * accept. |
858 | | */ |
859 | 0 | if (nitems > STATS_MCVLIST_MAX_ITEMS) |
860 | 0 | { |
861 | 0 | ereport(WARNING, |
862 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
863 | 0 | errmsg("could not parse array \"%s\": number of items (%d) exceeds maximum (%d)", |
864 | 0 | extarginfo[MOST_COMMON_VALS_ARG].argname, |
865 | 0 | nitems, STATS_MCVLIST_MAX_ITEMS)); |
866 | 0 | goto mcv_error; |
867 | 0 | } |
868 | | |
869 | 0 | if (!check_mcvlist_array(freqs_arr, MOST_COMMON_FREQS_ARG, 1, nitems) || |
870 | 0 | !check_mcvlist_array(base_freqs_arr, MOST_COMMON_BASE_FREQS_ARG, 1, nitems)) |
871 | 0 | { |
872 | | /* inconsistent input arrays found */ |
873 | 0 | goto mcv_error; |
874 | 0 | } |
875 | | |
876 | | /* |
877 | | * This part builds the contents for "most_common_val_nulls", based on the |
878 | | * values from "most_common_vals". |
879 | | */ |
880 | 0 | deconstruct_array_builtin(mcv_arr, TEXTOID, &mcv_elems, |
881 | 0 | &mcv_nulls, &check_nummcv); |
882 | |
|
883 | 0 | mcv = statext_mcv_import(WARNING, numattrs, |
884 | 0 | atttypids, atttypmods, atttypcolls, |
885 | 0 | nitems, mcv_elems, mcv_nulls, |
886 | 0 | (float8 *) ARR_DATA_PTR(freqs_arr), |
887 | 0 | (float8 *) ARR_DATA_PTR(base_freqs_arr)); |
888 | |
|
889 | 0 | *ok = (mcv != (Datum) 0); |
890 | |
|
891 | 0 | mcv_error: |
892 | 0 | return mcv; |
893 | 0 | } |
894 | | |
895 | | /* |
896 | | * Check if key is found in the list of expression argnames. |
897 | | */ |
898 | | static bool |
899 | | key_in_expr_argnames(JsonbValue *key) |
900 | 0 | { |
901 | 0 | Assert(key->type == jbvString); |
902 | 0 | for (int i = 0; i < NUM_ATTRIBUTE_STATS_ELEMS; i++) |
903 | 0 | { |
904 | 0 | if (strlen(extexprargname[i]) == key->val.string.len && |
905 | 0 | strncmp(extexprargname[i], key->val.string.val, key->val.string.len) == 0) |
906 | 0 | return true; |
907 | 0 | } |
908 | 0 | return false; |
909 | 0 | } |
910 | | |
911 | | /* |
912 | | * Verify that all of the keys in the object are valid argnames. |
913 | | */ |
914 | | static bool |
915 | | check_all_expr_argnames_valid(JsonbContainer *cont, AttrNumber exprnum) |
916 | | { |
917 | | bool all_keys_valid = true; |
918 | | |
919 | | JsonbIterator *jbit; |
920 | | JsonbIteratorToken jitok; |
921 | | JsonbValue jkey; |
922 | | |
923 | | Assert(JsonContainerIsObject(cont)); |
924 | | |
925 | | jbit = JsonbIteratorInit(cont); |
926 | | |
927 | | /* We always start off with a BEGIN OBJECT */ |
928 | | jitok = JsonbIteratorNext(&jbit, &jkey, false); |
929 | | Assert(jitok == WJB_BEGIN_OBJECT); |
930 | | |
931 | | while (true) |
932 | | { |
933 | | JsonbValue jval; |
934 | | |
935 | | jitok = JsonbIteratorNext(&jbit, &jkey, false); |
936 | | |
937 | | /* |
938 | | * We have run of keys. This is the only condition where it is |
939 | | * memory-safe to break out of the loop. |
940 | | */ |
941 | | if (jitok == WJB_END_OBJECT) |
942 | | break; |
943 | | |
944 | | /* We can only find keys inside an object */ |
945 | | Assert(jitok == WJB_KEY); |
946 | | Assert(jkey.type == jbvString); |
947 | | |
948 | | /* A value must follow the key */ |
949 | | jitok = JsonbIteratorNext(&jbit, &jval, false); |
950 | | Assert(jitok == WJB_VALUE); |
951 | | |
952 | | /* |
953 | | * If we have already found an invalid key, there is no point in |
954 | | * looking for more, because additional WARNINGs are just clutter. We |
955 | | * must continue iterating over the json to ensure that we clean up |
956 | | * all allocated memory. |
957 | | */ |
958 | | if (!all_keys_valid) |
959 | | continue; |
960 | | |
961 | | if (!key_in_expr_argnames(&jkey)) |
962 | | { |
963 | | char *bad_element_name = jbv_string_get_cstr(&jkey); |
964 | | |
965 | | ereport(WARNING, |
966 | | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
967 | | errmsg("could not import element in expression %d: invalid key name", |
968 | | exprnum)); |
969 | | |
970 | | pfree(bad_element_name); |
971 | | all_keys_valid = false; |
972 | | } |
973 | | } |
974 | | return all_keys_valid; |
975 | | } |
976 | | |
977 | | /* |
978 | | * Simple conversion of jbvString to cstring |
979 | | */ |
980 | | static char * |
981 | | jbv_string_get_cstr(JsonbValue *jval) |
982 | 0 | { |
983 | 0 | char *s; |
984 | |
|
985 | 0 | Assert(jval->type == jbvString); |
986 | |
|
987 | 0 | s = palloc0(jval->val.string.len + 1); |
988 | 0 | memcpy(s, jval->val.string.val, jval->val.string.len); |
989 | |
|
990 | 0 | return s; |
991 | 0 | } |
992 | | |
993 | | /* |
994 | | * Apply a jbvString value to a safe scalar input function. |
995 | | */ |
996 | | static bool |
997 | | jbv_to_infunc_datum(JsonbValue *jval, PGFunction func, AttrNumber exprnum, |
998 | | const char *argname, Datum *datum) |
999 | 0 | { |
1000 | 0 | ErrorSaveContext escontext = { |
1001 | 0 | .type = T_ErrorSaveContext, |
1002 | 0 | .details_wanted = true |
1003 | 0 | }; |
1004 | |
|
1005 | 0 | char *s = jbv_string_get_cstr(jval); |
1006 | 0 | bool ok; |
1007 | |
|
1008 | 0 | ok = DirectInputFunctionCallSafe(func, s, InvalidOid, -1, |
1009 | 0 | (Node *) &escontext, datum); |
1010 | | |
1011 | | /* |
1012 | | * If we got a type import error, use the report generated and add an |
1013 | | * error hint before throwing a warning. |
1014 | | */ |
1015 | 0 | if (!ok) |
1016 | 0 | { |
1017 | 0 | StringInfoData hint_str; |
1018 | |
|
1019 | 0 | initStringInfo(&hint_str); |
1020 | 0 | appendStringInfo(&hint_str, |
1021 | 0 | "Element \"%s\" in expression %d could not be parsed.", |
1022 | 0 | argname, exprnum); |
1023 | |
|
1024 | 0 | escontext.error_data->elevel = WARNING; |
1025 | 0 | escontext.error_data->hint = hint_str.data; |
1026 | |
|
1027 | 0 | ThrowErrorData(escontext.error_data); |
1028 | 0 | pfree(hint_str.data); |
1029 | 0 | } |
1030 | |
|
1031 | 0 | pfree(s); |
1032 | 0 | return ok; |
1033 | 0 | } |
1034 | | |
1035 | | /* |
1036 | | * Build an array datum with element type typid from a text datum, used as |
1037 | | * value of an attribute in a pg_statistic tuple. |
1038 | | * |
1039 | | * If an error is encountered, capture it, and reduce the elevel to WARNING. |
1040 | | * |
1041 | | * This is an adaptation of statatt_build_stavalues(). |
1042 | | */ |
1043 | | static Datum |
1044 | | array_in_safe(FmgrInfo *array_in, const char *s, Oid typid, int32 typmod, |
1045 | | AttrNumber exprnum, const char *element_name, bool *ok) |
1046 | | { |
1047 | | Datum result; |
1048 | | |
1049 | | ErrorSaveContext escontext = { |
1050 | | .type = T_ErrorSaveContext, |
1051 | | .details_wanted = true |
1052 | | }; |
1053 | | |
1054 | | *ok = false; |
1055 | | |
1056 | | /* |
1057 | | * If the array_in function returned an error, we will want to report that |
1058 | | * ERROR as a WARNING, and add some location context to the error message. |
1059 | | * Overwriting the existing hint (if any) is not ideal, and an error |
1060 | | * context would only work for level >= ERROR. |
1061 | | */ |
1062 | | if (!InputFunctionCallSafe(array_in, (char *) s, typid, typmod, |
1063 | | (Node *) &escontext, &result)) |
1064 | | { |
1065 | | StringInfoData hint_str; |
1066 | | |
1067 | | initStringInfo(&hint_str); |
1068 | | appendStringInfo(&hint_str, |
1069 | | "Element \"%s\" in expression %d could not be parsed.", |
1070 | | element_name, exprnum); |
1071 | | escontext.error_data->elevel = WARNING; |
1072 | | escontext.error_data->hint = hint_str.data; |
1073 | | ThrowErrorData(escontext.error_data); |
1074 | | pfree(hint_str.data); |
1075 | | return (Datum) 0; |
1076 | | } |
1077 | | |
1078 | | if (ARR_NDIM(DatumGetArrayTypeP(result)) != 1) |
1079 | | { |
1080 | | ereport(WARNING, |
1081 | | (errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1082 | | errmsg("could not import element \"%s\" in expression %d: must be a one-dimensional array", |
1083 | | element_name, exprnum))); |
1084 | | return (Datum) 0; |
1085 | | } |
1086 | | |
1087 | | if (array_contains_nulls(DatumGetArrayTypeP(result))) |
1088 | | { |
1089 | | ereport(WARNING, |
1090 | | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1091 | | errmsg("could not import element \"%s\" in expression %d: null value found", |
1092 | | element_name, exprnum)); |
1093 | | return (Datum) 0; |
1094 | | } |
1095 | | |
1096 | | *ok = true; |
1097 | | return result; |
1098 | | } |
1099 | | |
1100 | | /* |
1101 | | * Create a pg_statistic tuple from an expression JSONB container. |
1102 | | * |
1103 | | * The pg_statistic tuple is pre-populated with acceptable defaults, therefore |
1104 | | * even if there is an issue with all of the keys in the container, we can |
1105 | | * still return a legit tuple datum. |
1106 | | * |
1107 | | * Set pg_statistic_ok to true if all of the values found in the container |
1108 | | * were imported without issue. pg_statistic_ok is switched to "true" once |
1109 | | * the full pg_statistic tuple has been built and validated. |
1110 | | */ |
1111 | | static Datum |
1112 | | import_pg_statistic(Relation pgsd, JsonbContainer *cont, |
1113 | | AttrNumber exprnum, FmgrInfo *array_in_fn, |
1114 | | Oid typid, int32 typmod, Oid typcoll, |
1115 | | bool *pg_statistic_ok) |
1116 | 0 | { |
1117 | 0 | const char *argname = extarginfo[EXPRESSIONS_ARG].argname; |
1118 | 0 | TypeCacheEntry *typcache; |
1119 | 0 | Datum values[Natts_pg_statistic]; |
1120 | 0 | bool nulls[Natts_pg_statistic]; |
1121 | 0 | bool replaces[Natts_pg_statistic]; |
1122 | 0 | HeapTuple pgstup = NULL; |
1123 | 0 | Datum pgstdat = (Datum) 0; |
1124 | 0 | Oid elemtypid = InvalidOid; |
1125 | 0 | Oid elemeqopr = InvalidOid; |
1126 | 0 | bool found[NUM_ATTRIBUTE_STATS_ELEMS] = {0}; |
1127 | 0 | JsonbValue val[NUM_ATTRIBUTE_STATS_ELEMS] = {0}; |
1128 | | |
1129 | | /* Assume the worst by default. */ |
1130 | 0 | *pg_statistic_ok = false; |
1131 | |
|
1132 | 0 | if (!JsonContainerIsObject(cont)) |
1133 | 0 | { |
1134 | 0 | ereport(WARNING, |
1135 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1136 | 0 | errmsg("could not parse \"%s\": invalid element in expression %d", |
1137 | 0 | argname, exprnum)); |
1138 | 0 | goto pg_statistic_error; |
1139 | 0 | } |
1140 | | |
1141 | | /* |
1142 | | * Loop through all keys that we need to look up. If any value found is |
1143 | | * neither a string nor a NULL, there is not much we can do, so just give |
1144 | | * on the entire tuple for this expression. |
1145 | | */ |
1146 | 0 | for (int i = 0; i < NUM_ATTRIBUTE_STATS_ELEMS; i++) |
1147 | 0 | { |
1148 | 0 | const char *s = extexprargname[i]; |
1149 | 0 | int len = strlen(s); |
1150 | |
|
1151 | 0 | if (getKeyJsonValueFromContainer(cont, s, len, &val[i]) == NULL) |
1152 | 0 | continue; |
1153 | | |
1154 | 0 | switch (val[i].type) |
1155 | 0 | { |
1156 | 0 | case jbvString: |
1157 | 0 | found[i] = true; |
1158 | 0 | break; |
1159 | | |
1160 | 0 | case jbvNull: |
1161 | 0 | break; |
1162 | | |
1163 | 0 | default: |
1164 | 0 | ereport(WARNING, |
1165 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1166 | 0 | errmsg("could not parse \"%s\": invalid element in expression %d", argname, exprnum), |
1167 | 0 | errhint("Value of element \"%s\" must be a null or a string.", s)); |
1168 | 0 | goto pg_statistic_error; |
1169 | 0 | } |
1170 | 0 | } |
1171 | | |
1172 | | /* Look for invalid keys */ |
1173 | 0 | if (!check_all_expr_argnames_valid(cont, exprnum)) |
1174 | 0 | goto pg_statistic_error; |
1175 | | |
1176 | | /* |
1177 | | * There are two arg pairs, MCV+MCF and MCEV+MCEF. Both values must |
1178 | | * either be found or not be found. Any disagreement is a warning. Once |
1179 | | * we have ruled out disagreeing pairs, we can use either found flag as a |
1180 | | * proxy for the other. |
1181 | | */ |
1182 | 0 | if (found[MOST_COMMON_VALS_ELEM] != found[MOST_COMMON_FREQS_ELEM]) |
1183 | 0 | { |
1184 | 0 | ereport(WARNING, |
1185 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1186 | 0 | errmsg("could not parse \"%s\": invalid element in expression %d", |
1187 | 0 | argname, exprnum), |
1188 | 0 | errhint("\"%s\" and \"%s\" must be both either strings or nulls.", |
1189 | 0 | extexprargname[MOST_COMMON_VALS_ELEM], |
1190 | 0 | extexprargname[MOST_COMMON_FREQS_ELEM])); |
1191 | 0 | goto pg_statistic_error; |
1192 | 0 | } |
1193 | 0 | if (found[MOST_COMMON_ELEMS_ELEM] != found[MOST_COMMON_ELEM_FREQS_ELEM]) |
1194 | 0 | { |
1195 | 0 | ereport(WARNING, |
1196 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1197 | 0 | errmsg("could not parse \"%s\": invalid element in expression %d", |
1198 | 0 | argname, exprnum), |
1199 | 0 | errhint("\"%s\" and \"%s\" must be both either strings or nulls.", |
1200 | 0 | extexprargname[MOST_COMMON_ELEMS_ELEM], |
1201 | 0 | extexprargname[MOST_COMMON_ELEM_FREQS_ELEM])); |
1202 | 0 | goto pg_statistic_error; |
1203 | 0 | } |
1204 | | |
1205 | | /* |
1206 | | * Range types may expect three values to be set. All three of them must |
1207 | | * either be found or not be found. Any disagreement is a warning. |
1208 | | */ |
1209 | 0 | if (found[RANGE_LENGTH_HISTOGRAM_ELEM] != found[RANGE_EMPTY_FRAC_ELEM] || |
1210 | 0 | found[RANGE_LENGTH_HISTOGRAM_ELEM] != found[RANGE_BOUNDS_HISTOGRAM_ELEM]) |
1211 | 0 | { |
1212 | 0 | ereport(WARNING, |
1213 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1214 | 0 | errmsg("could not parse \"%s\": invalid element in expression %d", |
1215 | 0 | argname, exprnum), |
1216 | 0 | errhint("\"%s\", \"%s\", and \"%s\" must be all either strings or all nulls.", |
1217 | 0 | extexprargname[RANGE_LENGTH_HISTOGRAM_ELEM], |
1218 | 0 | extexprargname[RANGE_EMPTY_FRAC_ELEM], |
1219 | 0 | extexprargname[RANGE_BOUNDS_HISTOGRAM_ELEM])); |
1220 | 0 | goto pg_statistic_error; |
1221 | 0 | } |
1222 | | |
1223 | | /* This finds the right operators even if atttypid is a domain */ |
1224 | 0 | typcache = lookup_type_cache(typid, TYPECACHE_LT_OPR | TYPECACHE_EQ_OPR); |
1225 | |
|
1226 | 0 | statatt_init_empty_tuple(InvalidOid, InvalidAttrNumber, false, |
1227 | 0 | values, nulls, replaces); |
1228 | | |
1229 | | /* |
1230 | | * Special case: collation for tsvector is DEFAULT_COLLATION_OID. See |
1231 | | * compute_tsvector_stats(). |
1232 | | */ |
1233 | 0 | if (typid == TSVECTOROID) |
1234 | 0 | typcoll = DEFAULT_COLLATION_OID; |
1235 | | |
1236 | | /* |
1237 | | * We only need to fetch element type and eq operator if we have a stat of |
1238 | | * type MCELEM or DECHIST, otherwise the values are unnecessary and not |
1239 | | * meaningful. |
1240 | | */ |
1241 | 0 | if (found[MOST_COMMON_ELEMS_ELEM] || found[ELEM_COUNT_HISTOGRAM_ELEM]) |
1242 | 0 | { |
1243 | 0 | if (!statatt_get_elem_type(typid, typcache->typtype, |
1244 | 0 | &elemtypid, &elemeqopr)) |
1245 | 0 | { |
1246 | 0 | ereport(WARNING, |
1247 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1248 | 0 | errmsg("could not parse \"%s\": invalid element type in expression %d", |
1249 | 0 | argname, exprnum)); |
1250 | 0 | goto pg_statistic_error; |
1251 | 0 | } |
1252 | 0 | } |
1253 | | |
1254 | | /* |
1255 | | * These three fields can only be set if dealing with a range or |
1256 | | * multi-range type. |
1257 | | */ |
1258 | 0 | if (found[RANGE_LENGTH_HISTOGRAM_ELEM] || |
1259 | 0 | found[RANGE_EMPTY_FRAC_ELEM] || |
1260 | 0 | found[RANGE_BOUNDS_HISTOGRAM_ELEM]) |
1261 | 0 | { |
1262 | 0 | if (typcache->typtype != TYPTYPE_RANGE && |
1263 | 0 | typcache->typtype != TYPTYPE_MULTIRANGE) |
1264 | 0 | { |
1265 | 0 | ereport(WARNING, |
1266 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1267 | 0 | errmsg("could not parse \"%s\": invalid data in expression %d", |
1268 | 0 | argname, exprnum), |
1269 | 0 | errhint("\"%s\", \"%s\", and \"%s\" can only be set for a range type.", |
1270 | 0 | extexprargname[RANGE_LENGTH_HISTOGRAM_ELEM], |
1271 | 0 | extexprargname[RANGE_EMPTY_FRAC_ELEM], |
1272 | 0 | extexprargname[RANGE_BOUNDS_HISTOGRAM_ELEM])); |
1273 | 0 | goto pg_statistic_error; |
1274 | 0 | } |
1275 | 0 | } |
1276 | | |
1277 | | /* null_frac */ |
1278 | 0 | if (found[NULL_FRAC_ELEM]) |
1279 | 0 | { |
1280 | 0 | Datum datum; |
1281 | |
|
1282 | 0 | if (jbv_to_infunc_datum(&val[NULL_FRAC_ELEM], float4in, exprnum, |
1283 | 0 | extexprargname[NULL_FRAC_ELEM], &datum)) |
1284 | 0 | values[Anum_pg_statistic_stanullfrac - 1] = datum; |
1285 | 0 | else |
1286 | 0 | goto pg_statistic_error; |
1287 | 0 | } |
1288 | | |
1289 | | /* avg_width */ |
1290 | 0 | if (found[AVG_WIDTH_ELEM]) |
1291 | 0 | { |
1292 | 0 | Datum datum; |
1293 | |
|
1294 | 0 | if (jbv_to_infunc_datum(&val[AVG_WIDTH_ELEM], int4in, exprnum, |
1295 | 0 | extexprargname[AVG_WIDTH_ELEM], &datum)) |
1296 | 0 | values[Anum_pg_statistic_stawidth - 1] = datum; |
1297 | 0 | else |
1298 | 0 | goto pg_statistic_error; |
1299 | 0 | } |
1300 | | |
1301 | | /* n_distinct */ |
1302 | 0 | if (found[N_DISTINCT_ELEM]) |
1303 | 0 | { |
1304 | 0 | Datum datum; |
1305 | |
|
1306 | 0 | if (jbv_to_infunc_datum(&val[N_DISTINCT_ELEM], float4in, exprnum, |
1307 | 0 | extexprargname[N_DISTINCT_ELEM], &datum)) |
1308 | 0 | values[Anum_pg_statistic_stadistinct - 1] = datum; |
1309 | 0 | else |
1310 | 0 | goto pg_statistic_error; |
1311 | 0 | } |
1312 | | |
1313 | | /* |
1314 | | * The STAKIND statistics are the same as the ones found in attribute |
1315 | | * stats. However, these are all derived from json strings, whereas the |
1316 | | * ones derived for attribute stats are a mix of datatypes. This limits |
1317 | | * the opportunities for code sharing between the two. |
1318 | | * |
1319 | | * Some statistic kinds have both a stanumbers and a stavalues components. |
1320 | | * In those cases, both values must either be NOT NULL or both NULL, and |
1321 | | * if they aren't then we need to reject that stakind completely. |
1322 | | * Currently we go a step further and reject the expression array |
1323 | | * completely. |
1324 | | */ |
1325 | | |
1326 | 0 | if (found[MOST_COMMON_VALS_ELEM]) |
1327 | 0 | { |
1328 | 0 | Datum stavalues; |
1329 | 0 | Datum stanumbers; |
1330 | 0 | bool val_ok = false; |
1331 | 0 | bool num_ok = false; |
1332 | 0 | char *s; |
1333 | |
|
1334 | 0 | s = jbv_string_get_cstr(&val[MOST_COMMON_VALS_ELEM]); |
1335 | 0 | stavalues = array_in_safe(array_in_fn, s, typid, typmod, exprnum, |
1336 | 0 | extexprargname[MOST_COMMON_VALS_ELEM], |
1337 | 0 | &val_ok); |
1338 | |
|
1339 | 0 | pfree(s); |
1340 | |
|
1341 | 0 | s = jbv_string_get_cstr(&val[MOST_COMMON_FREQS_ELEM]); |
1342 | 0 | stanumbers = array_in_safe(array_in_fn, s, FLOAT4OID, -1, exprnum, |
1343 | 0 | extexprargname[MOST_COMMON_FREQS_ELEM], |
1344 | 0 | &num_ok); |
1345 | 0 | pfree(s); |
1346 | | |
1347 | | /* Only set the slot if both datums have been built */ |
1348 | 0 | if (val_ok && num_ok) |
1349 | 0 | { |
1350 | 0 | ArrayType *vals_arr = DatumGetArrayTypeP(stavalues); |
1351 | 0 | ArrayType *nums_arr = DatumGetArrayTypeP(stanumbers); |
1352 | 0 | int nvals = ARR_DIMS(vals_arr)[0]; |
1353 | 0 | int nnums = ARR_DIMS(nums_arr)[0]; |
1354 | |
|
1355 | 0 | if (nvals != nnums) |
1356 | 0 | { |
1357 | 0 | ereport(WARNING, |
1358 | 0 | (errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1359 | 0 | errmsg("could not parse \"%s\": incorrect number of elements (same as \"%s\" required)", |
1360 | 0 | "most_common_vals", |
1361 | 0 | "most_common_freqs"))); |
1362 | 0 | goto pg_statistic_error; |
1363 | 0 | } |
1364 | | |
1365 | 0 | statatt_set_slot(values, nulls, replaces, |
1366 | 0 | STATISTIC_KIND_MCV, |
1367 | 0 | typcache->eq_opr, typcoll, |
1368 | 0 | stanumbers, false, stavalues, false); |
1369 | 0 | } |
1370 | 0 | else |
1371 | 0 | goto pg_statistic_error; |
1372 | 0 | } |
1373 | | |
1374 | | /* STATISTIC_KIND_HISTOGRAM */ |
1375 | 0 | if (found[HISTOGRAM_BOUNDS_ELEM]) |
1376 | 0 | { |
1377 | 0 | Datum stavalues; |
1378 | 0 | bool val_ok = false; |
1379 | 0 | char *s = jbv_string_get_cstr(&val[HISTOGRAM_BOUNDS_ELEM]); |
1380 | |
|
1381 | 0 | stavalues = array_in_safe(array_in_fn, s, typid, typmod, exprnum, |
1382 | 0 | extexprargname[HISTOGRAM_BOUNDS_ELEM], |
1383 | 0 | &val_ok); |
1384 | 0 | pfree(s); |
1385 | |
|
1386 | 0 | if (val_ok) |
1387 | 0 | statatt_set_slot(values, nulls, replaces, |
1388 | 0 | STATISTIC_KIND_HISTOGRAM, |
1389 | 0 | typcache->lt_opr, typcoll, |
1390 | 0 | 0, true, stavalues, false); |
1391 | 0 | else |
1392 | 0 | goto pg_statistic_error; |
1393 | 0 | } |
1394 | | |
1395 | | /* STATISTIC_KIND_CORRELATION */ |
1396 | 0 | if (found[CORRELATION_ELEM]) |
1397 | 0 | { |
1398 | 0 | Datum corr[] = {(Datum) 0}; |
1399 | |
|
1400 | 0 | if (jbv_to_infunc_datum(&val[CORRELATION_ELEM], float4in, exprnum, |
1401 | 0 | extexprargname[CORRELATION_ELEM], &corr[0])) |
1402 | 0 | { |
1403 | 0 | ArrayType *arry = construct_array_builtin(corr, 1, FLOAT4OID); |
1404 | 0 | Datum stanumbers = PointerGetDatum(arry); |
1405 | |
|
1406 | 0 | statatt_set_slot(values, nulls, replaces, |
1407 | 0 | STATISTIC_KIND_CORRELATION, |
1408 | 0 | typcache->lt_opr, typcoll, |
1409 | 0 | stanumbers, false, 0, true); |
1410 | 0 | } |
1411 | 0 | else |
1412 | 0 | goto pg_statistic_error; |
1413 | 0 | } |
1414 | | |
1415 | | /* STATISTIC_KIND_MCELEM */ |
1416 | 0 | if (found[MOST_COMMON_ELEMS_ELEM]) |
1417 | 0 | { |
1418 | 0 | Datum stavalues; |
1419 | 0 | Datum stanumbers; |
1420 | 0 | bool val_ok = false; |
1421 | 0 | bool num_ok = false; |
1422 | 0 | char *s; |
1423 | |
|
1424 | 0 | s = jbv_string_get_cstr(&val[MOST_COMMON_ELEMS_ELEM]); |
1425 | 0 | stavalues = array_in_safe(array_in_fn, s, elemtypid, typmod, exprnum, |
1426 | 0 | extexprargname[MOST_COMMON_ELEMS_ELEM], |
1427 | 0 | &val_ok); |
1428 | 0 | pfree(s); |
1429 | | |
1430 | |
|
1431 | 0 | s = jbv_string_get_cstr(&val[MOST_COMMON_ELEM_FREQS_ELEM]); |
1432 | 0 | stanumbers = array_in_safe(array_in_fn, s, FLOAT4OID, -1, exprnum, |
1433 | 0 | extexprargname[MOST_COMMON_ELEM_FREQS_ELEM], |
1434 | 0 | &num_ok); |
1435 | 0 | pfree(s); |
1436 | | |
1437 | | /* Only set the slot if both datums have been built */ |
1438 | 0 | if (val_ok && num_ok) |
1439 | 0 | statatt_set_slot(values, nulls, replaces, |
1440 | 0 | STATISTIC_KIND_MCELEM, |
1441 | 0 | elemeqopr, typcoll, |
1442 | 0 | stanumbers, false, stavalues, false); |
1443 | 0 | else |
1444 | 0 | goto pg_statistic_error; |
1445 | 0 | } |
1446 | | |
1447 | | /* STATISTIC_KIND_DECHIST */ |
1448 | 0 | if (found[ELEM_COUNT_HISTOGRAM_ELEM]) |
1449 | 0 | { |
1450 | 0 | Datum stanumbers; |
1451 | 0 | bool num_ok = false; |
1452 | 0 | char *s; |
1453 | |
|
1454 | 0 | s = jbv_string_get_cstr(&val[ELEM_COUNT_HISTOGRAM_ELEM]); |
1455 | 0 | stanumbers = array_in_safe(array_in_fn, s, FLOAT4OID, -1, exprnum, |
1456 | 0 | extexprargname[ELEM_COUNT_HISTOGRAM_ELEM], |
1457 | 0 | &num_ok); |
1458 | 0 | pfree(s); |
1459 | |
|
1460 | 0 | if (num_ok) |
1461 | 0 | statatt_set_slot(values, nulls, replaces, STATISTIC_KIND_DECHIST, |
1462 | 0 | elemeqopr, typcoll, stanumbers, false, 0, true); |
1463 | 0 | else |
1464 | 0 | goto pg_statistic_error; |
1465 | 0 | } |
1466 | | |
1467 | | /* |
1468 | | * STATISTIC_KIND_BOUNDS_HISTOGRAM |
1469 | | * |
1470 | | * This stakind appears before STATISTIC_KIND_RANGE_LENGTH_HISTOGRAM even |
1471 | | * though it is numerically greater, and all other stakinds appear in |
1472 | | * numerical order. |
1473 | | */ |
1474 | 0 | if (found[RANGE_BOUNDS_HISTOGRAM_ELEM]) |
1475 | 0 | { |
1476 | 0 | Datum stavalues; |
1477 | 0 | bool val_ok = false; |
1478 | 0 | char *s; |
1479 | 0 | Oid rtypid = typid; |
1480 | | |
1481 | | /* |
1482 | | * If it's a multirange, step down to the range type, as is done by |
1483 | | * multirange_typanalyze(). |
1484 | | */ |
1485 | 0 | if (type_is_multirange(typid)) |
1486 | 0 | rtypid = get_multirange_range(typid); |
1487 | |
|
1488 | 0 | s = jbv_string_get_cstr(&val[RANGE_BOUNDS_HISTOGRAM_ELEM]); |
1489 | |
|
1490 | 0 | stavalues = array_in_safe(array_in_fn, s, rtypid, typmod, exprnum, |
1491 | 0 | extexprargname[RANGE_BOUNDS_HISTOGRAM_ELEM], |
1492 | 0 | &val_ok); |
1493 | |
|
1494 | 0 | if (val_ok && statatt_check_bounds_histogram(stavalues)) |
1495 | 0 | statatt_set_slot(values, nulls, replaces, |
1496 | 0 | STATISTIC_KIND_BOUNDS_HISTOGRAM, |
1497 | 0 | InvalidOid, InvalidOid, |
1498 | 0 | 0, true, stavalues, false); |
1499 | 0 | else |
1500 | 0 | goto pg_statistic_error; |
1501 | 0 | } |
1502 | | |
1503 | | /* STATISTIC_KIND_RANGE_LENGTH_HISTOGRAM */ |
1504 | 0 | if (found[RANGE_LENGTH_HISTOGRAM_ELEM]) |
1505 | 0 | { |
1506 | 0 | Datum empty_frac[] = {(Datum) 0}; |
1507 | 0 | Datum stavalues; |
1508 | 0 | Datum stanumbers; |
1509 | 0 | bool val_ok = false; |
1510 | 0 | char *s; |
1511 | |
|
1512 | 0 | if (jbv_to_infunc_datum(&val[RANGE_EMPTY_FRAC_ELEM], float4in, exprnum, |
1513 | 0 | extexprargname[RANGE_EMPTY_FRAC_ELEM], &empty_frac[0])) |
1514 | 0 | { |
1515 | 0 | ArrayType *arry = construct_array_builtin(empty_frac, 1, FLOAT4OID); |
1516 | |
|
1517 | 0 | stanumbers = PointerGetDatum(arry); |
1518 | 0 | } |
1519 | 0 | else |
1520 | 0 | goto pg_statistic_error; |
1521 | | |
1522 | 0 | s = jbv_string_get_cstr(&val[RANGE_LENGTH_HISTOGRAM_ELEM]); |
1523 | 0 | stavalues = array_in_safe(array_in_fn, s, FLOAT8OID, -1, exprnum, |
1524 | 0 | extexprargname[RANGE_LENGTH_HISTOGRAM_ELEM], |
1525 | 0 | &val_ok); |
1526 | |
|
1527 | 0 | if (val_ok) |
1528 | 0 | statatt_set_slot(values, nulls, replaces, |
1529 | 0 | STATISTIC_KIND_RANGE_LENGTH_HISTOGRAM, |
1530 | 0 | Float8LessOperator, InvalidOid, |
1531 | 0 | stanumbers, false, stavalues, false); |
1532 | 0 | else |
1533 | 0 | goto pg_statistic_error; |
1534 | 0 | } |
1535 | | |
1536 | 0 | pgstup = heap_form_tuple(RelationGetDescr(pgsd), values, nulls); |
1537 | 0 | pgstdat = heap_copy_tuple_as_datum(pgstup, RelationGetDescr(pgsd)); |
1538 | |
|
1539 | 0 | heap_freetuple(pgstup); |
1540 | |
|
1541 | 0 | *pg_statistic_ok = true; |
1542 | |
|
1543 | 0 | return pgstdat; |
1544 | | |
1545 | 0 | pg_statistic_error: |
1546 | 0 | return (Datum) 0; |
1547 | 0 | } |
1548 | | |
1549 | | /* |
1550 | | * Create the stxdexpr datum, which is an array of pg_statistic rows with all |
1551 | | * of the object identification fields left at defaults, using the json array |
1552 | | * of objects/nulls referenced against the datatypes for the expressions. |
1553 | | * |
1554 | | * The exprs_is_perfect will be set to true if all pg_statistic rows were |
1555 | | * imported cleanly. If any of them experienced a problem (and thus were |
1556 | | * set as if they were null), then the expression is kept but exprs_is_perfect |
1557 | | * will be marked as false. |
1558 | | * |
1559 | | * This datum is needed to fill out a complete pg_statistic_ext_data tuple. |
1560 | | */ |
1561 | | static Datum |
1562 | | import_expressions(Relation pgsd, int numexprs, |
1563 | | Oid *atttypids, int32 *atttypmods, |
1564 | | Oid *atttypcolls, Jsonb *exprs_jsonb, |
1565 | | bool *exprs_is_perfect) |
1566 | 0 | { |
1567 | 0 | const char *argname = extarginfo[EXPRESSIONS_ARG].argname; |
1568 | 0 | Oid pgstypoid = get_rel_type_id(StatisticRelationId); |
1569 | 0 | ArrayBuildState *astate = NULL; |
1570 | 0 | Datum result = (Datum) 0; |
1571 | 0 | int num_import_ok = 0; |
1572 | 0 | JsonbContainer *root; |
1573 | 0 | int num_root_elements; |
1574 | |
|
1575 | 0 | FmgrInfo array_in_fn; |
1576 | |
|
1577 | 0 | *exprs_is_perfect = false; |
1578 | | |
1579 | | /* Json schema must be [{expr},...] */ |
1580 | 0 | if (!JB_ROOT_IS_ARRAY(exprs_jsonb)) |
1581 | 0 | { |
1582 | 0 | ereport(WARNING, |
1583 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1584 | 0 | errmsg("could not parse \"%s\": root-level array required", argname)); |
1585 | 0 | goto exprs_error; |
1586 | 0 | } |
1587 | | |
1588 | 0 | root = &exprs_jsonb->root; |
1589 | | |
1590 | | /* |
1591 | | * The number of elements in the array must match the number of |
1592 | | * expressions in the stats object definition. |
1593 | | */ |
1594 | 0 | num_root_elements = JsonContainerSize(root); |
1595 | 0 | if (numexprs != num_root_elements) |
1596 | 0 | { |
1597 | 0 | ereport(WARNING, |
1598 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1599 | 0 | errmsg("could not parse \"%s\": incorrect number of elements (%d required)", |
1600 | 0 | argname, numexprs)); |
1601 | 0 | goto exprs_error; |
1602 | 0 | } |
1603 | | |
1604 | 0 | fmgr_info(F_ARRAY_IN, &array_in_fn); |
1605 | | |
1606 | | /* |
1607 | | * Iterate over each expected expression object in the array. Some of |
1608 | | * them could be null. If the element is a completely wrong data type, |
1609 | | * give a WARNING and then treat the element like a NULL element in the |
1610 | | * result array. |
1611 | | * |
1612 | | * Each expression *MUST* have a value appended in the result pg_statistic |
1613 | | * array. |
1614 | | */ |
1615 | 0 | for (int i = 0; i < numexprs; i++) |
1616 | 0 | { |
1617 | 0 | Datum pgstdat = (Datum) 0; |
1618 | 0 | bool isnull = false; |
1619 | 0 | AttrNumber exprattnum = -1 - i; |
1620 | |
|
1621 | 0 | JsonbValue *elem = getIthJsonbValueFromContainer(root, i); |
1622 | |
|
1623 | 0 | switch (elem->type) |
1624 | 0 | { |
1625 | 0 | case jbvBinary: |
1626 | 0 | { |
1627 | 0 | bool sta_ok = false; |
1628 | | |
1629 | | /* a real stats object */ |
1630 | 0 | pgstdat = import_pg_statistic(pgsd, elem->val.binary.data, |
1631 | 0 | exprattnum, &array_in_fn, |
1632 | 0 | atttypids[i], atttypmods[i], |
1633 | 0 | atttypcolls[i], &sta_ok); |
1634 | | |
1635 | | /* |
1636 | | * If some incorrect data has been found, assign NULL for |
1637 | | * this expression as a mean to give up. |
1638 | | */ |
1639 | 0 | if (sta_ok) |
1640 | 0 | num_import_ok++; |
1641 | 0 | else |
1642 | 0 | { |
1643 | 0 | isnull = true; |
1644 | 0 | pgstdat = (Datum) 0; |
1645 | 0 | } |
1646 | 0 | } |
1647 | 0 | break; |
1648 | | |
1649 | 0 | case jbvNull: |
1650 | | /* NULL placeholder for invalid data, still fine */ |
1651 | 0 | isnull = true; |
1652 | 0 | num_import_ok++; |
1653 | 0 | break; |
1654 | | |
1655 | 0 | default: |
1656 | | /* cannot possibly be valid */ |
1657 | 0 | ereport(WARNING, |
1658 | 0 | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1659 | 0 | errmsg("could not parse \"%s\": invalid element in expression %d", |
1660 | 0 | argname, exprattnum)); |
1661 | 0 | goto exprs_error; |
1662 | 0 | } |
1663 | | |
1664 | 0 | astate = accumArrayResult(astate, pgstdat, isnull, pgstypoid, |
1665 | 0 | CurrentMemoryContext); |
1666 | 0 | } |
1667 | | |
1668 | | /* |
1669 | | * The expressions datum is perfect *if and only if* all of the |
1670 | | * pg_statistic elements were also ok, for a number of elements equal to |
1671 | | * the number of expressions. Anything else means a failure in restoring |
1672 | | * the data of this statistics object. |
1673 | | */ |
1674 | 0 | *exprs_is_perfect = (num_import_ok == numexprs); |
1675 | |
|
1676 | 0 | if (astate != NULL) |
1677 | 0 | result = makeArrayResult(astate, CurrentMemoryContext); |
1678 | |
|
1679 | 0 | return result; |
1680 | | |
1681 | 0 | exprs_error: |
1682 | 0 | if (astate != NULL) |
1683 | 0 | pfree(astate); |
1684 | 0 | return (Datum) 0; |
1685 | 0 | }; |
1686 | | |
1687 | | /* |
1688 | | * Remove an existing pg_statistic_ext_data row for a given pg_statistic_ext |
1689 | | * row and "inherited" pair. |
1690 | | */ |
1691 | | static bool |
1692 | | delete_pg_statistic_ext_data(Oid stxoid, bool inherited) |
1693 | 0 | { |
1694 | 0 | Relation sed = table_open(StatisticExtDataRelationId, RowExclusiveLock); |
1695 | 0 | HeapTuple oldtup; |
1696 | 0 | bool result = false; |
1697 | | |
1698 | | /* Is there already a pg_statistic_ext_data tuple for this attribute? */ |
1699 | 0 | oldtup = SearchSysCache2(STATEXTDATASTXOID, |
1700 | 0 | ObjectIdGetDatum(stxoid), |
1701 | 0 | BoolGetDatum(inherited)); |
1702 | |
|
1703 | 0 | if (HeapTupleIsValid(oldtup)) |
1704 | 0 | { |
1705 | 0 | CatalogTupleDelete(sed, &oldtup->t_self); |
1706 | 0 | ReleaseSysCache(oldtup); |
1707 | 0 | result = true; |
1708 | 0 | } |
1709 | |
|
1710 | 0 | table_close(sed, RowExclusiveLock); |
1711 | |
|
1712 | 0 | CommandCounterIncrement(); |
1713 | |
|
1714 | 0 | return result; |
1715 | 0 | } |
1716 | | |
1717 | | /* |
1718 | | * Restore (insert or replace) statistics for the given statistics object. |
1719 | | * |
1720 | | * This function accepts variadic arguments in key-value pairs, which are |
1721 | | * given to stats_fill_fcinfo_from_arg_pairs to be mapped into positional |
1722 | | * arguments. |
1723 | | */ |
1724 | | Datum |
1725 | | pg_restore_extended_stats(PG_FUNCTION_ARGS) |
1726 | 0 | { |
1727 | 0 | LOCAL_FCINFO(positional_fcinfo, NUM_EXTENDED_STATS_ARGS); |
1728 | 0 | bool result = true; |
1729 | |
|
1730 | 0 | InitFunctionCallInfoData(*positional_fcinfo, NULL, NUM_EXTENDED_STATS_ARGS, |
1731 | 0 | InvalidOid, NULL, NULL); |
1732 | |
|
1733 | 0 | if (!stats_fill_fcinfo_from_arg_pairs(fcinfo, positional_fcinfo, extarginfo)) |
1734 | 0 | result = false; |
1735 | |
|
1736 | 0 | if (!extended_statistics_update(positional_fcinfo)) |
1737 | 0 | result = false; |
1738 | |
|
1739 | 0 | PG_RETURN_BOOL(result); |
1740 | 0 | } |
1741 | | |
1742 | | /* |
1743 | | * Delete statistics for the given statistics object. |
1744 | | */ |
1745 | | Datum |
1746 | | pg_clear_extended_stats(PG_FUNCTION_ARGS) |
1747 | | { |
1748 | | char *relnspname; |
1749 | | char *relname; |
1750 | | char *nspname; |
1751 | | Oid nspoid; |
1752 | | Oid relid; |
1753 | | char *stxname; |
1754 | | bool inherited; |
1755 | | Relation pg_stext; |
1756 | | HeapTuple tup; |
1757 | | Form_pg_statistic_ext stxform; |
1758 | | Oid locked_table = InvalidOid; |
1759 | | |
1760 | | /* relation arguments */ |
1761 | | stats_check_required_arg(fcinfo, extarginfo, RELSCHEMA_ARG); |
1762 | | relnspname = TextDatumGetCString(PG_GETARG_DATUM(RELSCHEMA_ARG)); |
1763 | | stats_check_required_arg(fcinfo, extarginfo, RELNAME_ARG); |
1764 | | relname = TextDatumGetCString(PG_GETARG_DATUM(RELNAME_ARG)); |
1765 | | |
1766 | | /* extended statistics arguments */ |
1767 | | stats_check_required_arg(fcinfo, extarginfo, STATSCHEMA_ARG); |
1768 | | nspname = TextDatumGetCString(PG_GETARG_DATUM(STATSCHEMA_ARG)); |
1769 | | stats_check_required_arg(fcinfo, extarginfo, STATNAME_ARG); |
1770 | | stxname = TextDatumGetCString(PG_GETARG_DATUM(STATNAME_ARG)); |
1771 | | stats_check_required_arg(fcinfo, extarginfo, INHERITED_ARG); |
1772 | | inherited = PG_GETARG_BOOL(INHERITED_ARG); |
1773 | | |
1774 | | if (RecoveryInProgress()) |
1775 | | { |
1776 | | ereport(WARNING, |
1777 | | errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
1778 | | errmsg("recovery is in progress"), |
1779 | | errhint("Statistics cannot be modified during recovery.")); |
1780 | | PG_RETURN_VOID(); |
1781 | | } |
1782 | | |
1783 | | /* |
1784 | | * First open the relation where we expect to find the statistics. This |
1785 | | * is similar to relation and attribute statistics, so as ACL checks are |
1786 | | * done before any locks are taken, even before any attempts related to |
1787 | | * the extended stats object. |
1788 | | */ |
1789 | | relid = RangeVarGetRelidExtended(makeRangeVar(relnspname, relname, -1), |
1790 | | ShareUpdateExclusiveLock, 0, |
1791 | | RangeVarCallbackForStats, &locked_table); |
1792 | | |
1793 | | /* Now check if the namespace of the stats object exists. */ |
1794 | | nspoid = get_namespace_oid(nspname, true); |
1795 | | if (nspoid == InvalidOid) |
1796 | | { |
1797 | | ereport(WARNING, |
1798 | | errcode(ERRCODE_UNDEFINED_OBJECT), |
1799 | | errmsg("could not find schema \"%s\"", nspname)); |
1800 | | PG_RETURN_VOID(); |
1801 | | } |
1802 | | |
1803 | | pg_stext = table_open(StatisticExtRelationId, RowExclusiveLock); |
1804 | | tup = get_pg_statistic_ext(pg_stext, nspoid, stxname); |
1805 | | |
1806 | | if (!HeapTupleIsValid(tup)) |
1807 | | { |
1808 | | table_close(pg_stext, RowExclusiveLock); |
1809 | | ereport(WARNING, |
1810 | | errcode(ERRCODE_UNDEFINED_OBJECT), |
1811 | | errmsg("could not find extended statistics object \"%s.%s\"", |
1812 | | nspname, stxname)); |
1813 | | PG_RETURN_VOID(); |
1814 | | } |
1815 | | |
1816 | | stxform = (Form_pg_statistic_ext) GETSTRUCT(tup); |
1817 | | |
1818 | | /* |
1819 | | * This should be consistent, based on the lock taken on the table when we |
1820 | | * started. |
1821 | | */ |
1822 | | if (stxform->stxrelid != relid) |
1823 | | { |
1824 | | heap_freetuple(tup); |
1825 | | table_close(pg_stext, RowExclusiveLock); |
1826 | | ereport(WARNING, |
1827 | | errcode(ERRCODE_INVALID_PARAMETER_VALUE), |
1828 | | errmsg("could not clear extended statistics object \"%s.%s\": incorrect relation \"%s.%s\" specified", |
1829 | | get_namespace_name(nspoid), stxname, |
1830 | | relnspname, relname)); |
1831 | | PG_RETURN_VOID(); |
1832 | | } |
1833 | | |
1834 | | delete_pg_statistic_ext_data(stxform->oid, inherited); |
1835 | | heap_freetuple(tup); |
1836 | | |
1837 | | table_close(pg_stext, RowExclusiveLock); |
1838 | | |
1839 | | PG_RETURN_VOID(); |
1840 | | } |