1# Copyright 2023 The gRPC authors.
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7# http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14
15from __future__ import annotations
16
17import abc
18import contextlib
19import logging
20import threading
21from typing import Any, Generator, Generic, List, Optional, TypeVar
22
23from grpc._cython import cygrpc as _cygrpc
24from grpc._typing import ChannelArgumentType
25
26_LOGGER = logging.getLogger(__name__)
27
28_channel = Any # _channel.py imports this module.
29ClientCallTracerCapsule = TypeVar("ClientCallTracerCapsule")
30ServerCallTracerFactoryCapsule = TypeVar("ServerCallTracerFactoryCapsule")
31
32_plugin_lock: threading.RLock = threading.RLock()
33_OBSERVABILITY_PLUGIN: Optional["ObservabilityPlugin"] = None
34_SERVICES_TO_EXCLUDE: List[bytes] = [
35 b"google.monitoring.v3.MetricService",
36 b"google.devtools.cloudtrace.v2.TraceService",
37]
38
39
40class ServerCallTracerFactory:
41 """An encapsulation of a ServerCallTracerFactory.
42
43 Instances of this class can be passed to a Channel as values for the
44 grpc.experimental.server_call_tracer_factory option
45 """
46
47 def __init__(self, address):
48 self._address = address
49
50 def __int__(self):
51 return self._address
52
53
54class ObservabilityPlugin(
55 Generic[ClientCallTracerCapsule, ServerCallTracerFactoryCapsule],
56 metaclass=abc.ABCMeta,
57):
58 """Abstract base class for observability plugin.
59
60 *This is a semi-private class that was intended for the exclusive use of
61 the gRPC team.*
62
63 The ClientCallTracerCapsule and ClientCallTracerCapsule created by this
64 plugin should be inject to gRPC core using observability_init at the
65 start of a program, before any channels/servers are built.
66
67 Any future methods added to this interface cannot have the
68 @abc.abstractmethod annotation.
69
70 Attributes:
71 _stats_enabled: A bool indicates whether tracing is enabled.
72 _tracing_enabled: A bool indicates whether stats(metrics) is enabled.
73 _registered_methods: A set which stores the registered method names in
74 bytes.
75 """
76
77 _tracing_enabled: bool = False
78 _stats_enabled: bool = False
79
80 @abc.abstractmethod
81 def create_client_call_tracer(
82 self, method_name: bytes, target: bytes
83 ) -> ClientCallTracerCapsule:
84 """Creates a ClientCallTracerCapsule.
85
86 After register the plugin, if tracing or stats is enabled, this method
87 will be called after a call was created, the ClientCallTracer created
88 by this method will be saved to call context.
89
90 The ClientCallTracer is an object which implements `grpc_core::ClientCallTracer`
91 interface and wrapped in a PyCapsule using `client_call_tracer` as name.
92
93 Args:
94 method_name: The method name of the call in byte format.
95 target: The channel target of the call in byte format.
96 registered_method: Wether this method is pre-registered.
97
98 Returns:
99 A PyCapsule which stores a ClientCallTracer object.
100 """
101 raise NotImplementedError()
102
103 @abc.abstractmethod
104 def delete_client_call_tracer(
105 self, client_call_tracer: ClientCallTracerCapsule
106 ) -> None:
107 """Deletes the ClientCallTracer stored in ClientCallTracerCapsule.
108
109 After register the plugin, if tracing or stats is enabled, this method
110 will be called at the end of the call to destroy the ClientCallTracer.
111
112 The ClientCallTracer is an object which implements `grpc_core::ClientCallTracer`
113 interface and wrapped in a PyCapsule using `client_call_tracer` as name.
114
115 Args:
116 client_call_tracer: A PyCapsule which stores a ClientCallTracer object.
117 """
118 raise NotImplementedError()
119
120 @abc.abstractmethod
121 def save_trace_context(
122 self, trace_id: str, span_id: str, is_sampled: bool
123 ) -> None:
124 """Saves the trace_id and span_id related to the current span.
125
126 After register the plugin, if tracing is enabled, this method will be
127 called after the server finished sending response.
128
129 This method can be used to propagate census context.
130
131 Args:
132 trace_id: The identifier for the trace associated with the span as a
133 32-character hexadecimal encoded string,
134 e.g. 26ed0036f2eff2b7317bccce3e28d01f
135 span_id: The identifier for the span as a 16-character hexadecimal encoded
136 string. e.g. 113ec879e62583bc
137 is_sampled: A bool indicates whether the span is sampled.
138 """
139 raise NotImplementedError()
140
141 @abc.abstractmethod
142 def create_server_call_tracer_factory(
143 self,
144 *,
145 xds: bool = False,
146 ) -> Optional[ServerCallTracerFactoryCapsule]:
147 """Creates a ServerCallTracerFactoryCapsule.
148
149 This method will be called at server initialization time to create a
150 ServerCallTracerFactory, which will be registered to gRPC core.
151
152 The ServerCallTracerFactory is an object which implements
153 `grpc_core::ServerCallTracerFactory` interface and wrapped in a PyCapsule
154 using `server_call_tracer_factory` as name.
155
156 Args:
157 xds: Whether the server is xds server.
158 Returns:
159 A PyCapsule which stores a ServerCallTracerFactory object. Or None if
160 plugin decides not to create ServerCallTracerFactory.
161 """
162 raise NotImplementedError()
163
164 @abc.abstractmethod
165 def record_rpc_latency(
166 self, method: str, target: str, rpc_latency: float, status_code: Any
167 ) -> None:
168 """Record the latency of the RPC.
169
170 After register the plugin, if stats is enabled, this method will be
171 called at the end of each RPC.
172
173 Args:
174 method: The fully-qualified name of the RPC method being invoked.
175 target: The target name of the RPC method being invoked.
176 rpc_latency: The latency for the RPC in seconds, equals to the time between
177 when the client invokes the RPC and when the client receives the status.
178 status_code: An element of grpc.StatusCode in string format representing the
179 final status for the RPC.
180 """
181 raise NotImplementedError()
182
183 def set_tracing(self, enable: bool) -> None:
184 """Enable or disable tracing.
185
186 Args:
187 enable: A bool indicates whether tracing should be enabled.
188 """
189 self._tracing_enabled = enable
190
191 def set_stats(self, enable: bool) -> None:
192 """Enable or disable stats(metrics).
193
194 Args:
195 enable: A bool indicates whether stats should be enabled.
196 """
197 self._stats_enabled = enable
198
199 def save_registered_method(self, method_name: bytes) -> None:
200 """Saves the method name to registered_method list.
201
202 When exporting metrics, method name for unregistered methods will be replaced
203 with 'other' by default.
204
205 Args:
206 method_name: The method name in bytes.
207 """
208 raise NotImplementedError()
209
210 @property
211 def tracing_enabled(self) -> bool:
212 return self._tracing_enabled
213
214 @property
215 def stats_enabled(self) -> bool:
216 return self._stats_enabled
217
218 @property
219 def observability_enabled(self) -> bool:
220 return self.tracing_enabled or self.stats_enabled
221
222
223@contextlib.contextmanager
224def get_plugin() -> Generator[Optional[ObservabilityPlugin], None, None]:
225 """Get the ObservabilityPlugin in _observability module.
226
227 Returns:
228 The ObservabilityPlugin currently registered with the _observability
229 module. Or None if no plugin exists at the time of calling this method.
230 """
231 with _plugin_lock:
232 yield _OBSERVABILITY_PLUGIN
233
234
235def set_plugin(observability_plugin: Optional[ObservabilityPlugin]) -> None:
236 """Save ObservabilityPlugin to _observability module.
237
238 Args:
239 observability_plugin: The ObservabilityPlugin to save.
240
241 Raises:
242 ValueError: If an ObservabilityPlugin was already registered at the
243 time of calling this method.
244 """
245 global _OBSERVABILITY_PLUGIN # pylint: disable=global-statement
246 with _plugin_lock:
247 if observability_plugin and _OBSERVABILITY_PLUGIN:
248 raise ValueError("observability_plugin was already set!")
249 _OBSERVABILITY_PLUGIN = observability_plugin
250
251
252def observability_init(observability_plugin: ObservabilityPlugin) -> None:
253 """Initialize observability with provided ObservabilityPlugin.
254
255 This method have to be called at the start of a program, before any
256 channels/servers are built.
257
258 Args:
259 observability_plugin: The ObservabilityPlugin to use.
260
261 Raises:
262 ValueError: If an ObservabilityPlugin was already registered at the
263 time of calling this method.
264 """
265 set_plugin(observability_plugin)
266
267
268def observability_deinit() -> None:
269 """Clear the observability context, including ObservabilityPlugin and
270 ServerCallTracerFactory
271
272 This method have to be called after exit observability context so that
273 it's possible to re-initialize again.
274 """
275 set_plugin(None)
276 _cygrpc.clear_server_call_tracer_factory()
277
278
279def delete_call_tracer(client_call_tracer_capsule: Any) -> None:
280 """Deletes the ClientCallTracer stored in ClientCallTracerCapsule.
281
282 This method will be called at the end of the call to destroy the ClientCallTracer.
283
284 The ClientCallTracer is an object which implements `grpc_core::ClientCallTracer`
285 interface and wrapped in a PyCapsule using `client_call_tracer` as the name.
286
287 Args:
288 client_call_tracer_capsule: A PyCapsule which stores a ClientCallTracer object.
289 """
290 with get_plugin() as plugin:
291 if plugin and plugin.observability_enabled:
292 plugin.delete_client_call_tracer(client_call_tracer_capsule)
293
294
295def maybe_record_rpc_latency(state: "_channel._RPCState") -> None:
296 """Record the latency of the RPC, if the plugin is registered and stats is enabled.
297
298 This method will be called at the end of each RPC.
299
300 Args:
301 state: a grpc._channel._RPCState object which contains the stats related to the
302 RPC.
303 """
304 # TODO(xuanwn): use channel args to exclude those metrics.
305 for exclude_prefix in _SERVICES_TO_EXCLUDE:
306 if exclude_prefix in state.method.encode("utf8"):
307 return
308 with get_plugin() as plugin:
309 if plugin and plugin.stats_enabled:
310 rpc_latency_s = state.rpc_end_time - state.rpc_start_time
311 rpc_latency_ms = rpc_latency_s * 1000
312 plugin.record_rpc_latency(
313 state.method, state.target, rpc_latency_ms, state.code
314 )
315
316
317def create_server_call_tracer_factory_option(xds: bool) -> ChannelArgumentType:
318 with get_plugin() as plugin:
319 if plugin and plugin.stats_enabled:
320 server_call_tracer_factory_address = (
321 _cygrpc.get_server_call_tracer_factory_address(plugin, xds)
322 )
323 if server_call_tracer_factory_address:
324 return (
325 (
326 "grpc.experimental.server_call_tracer_factory",
327 ServerCallTracerFactory(
328 server_call_tracer_factory_address
329 ),
330 ),
331 )
332 return ()