Refactor Counter metrics to be homeserver-scoped

changelog.d/18656.misc

@@ -0,0 +1 @@
+Refactor `Counter` metrics to be homeserver-scoped.

scripts-dev/mypy_synapse_plugin.py

@@ -28,8 +28,13 @@
 import mypy.types
 from mypy.erasetype import remove_instance_last_known_values
 from mypy.errorcodes import ErrorCode
-from mypy.nodes import ARG_NAMED_OPT, TempNode, Var
-from mypy.plugin import FunctionSigContext, MethodSigContext, Plugin
+from mypy.nodes import ARG_NAMED_OPT, ListExpr, NameExpr, TempNode, Var
+from mypy.plugin import (
+    FunctionLike,
+    FunctionSigContext,
+    MethodSigContext,
+    Plugin,
+)
 from mypy.typeops import bind_self
 from mypy.types import (
     AnyType,
@@ -43,8 +48,26 @@
     UnionType,
 )
 
+PROMETHEUS_METRIC_MISSING_SERVER_NAME_LABEL = ErrorCode(
+    "missing-server-name-label",
+    "`SERVER_NAME_LABEL` required in metric",
+    category="per-homeserver-tenant-metrics",
+)
+
 
 class SynapsePlugin(Plugin):
+    def get_function_signature_hook(
+        self, fullname: str
+    ) -> Optional[Callable[[FunctionSigContext], FunctionLike]]:
+        if fullname in (
+            "prometheus_client.metrics.Counter",
+            # TODO: Add other prometheus_client metrics that need checking as we
+            # refactor, see https://github.com/element-hq/synapse/issues/18592
+        ):
+            return check_prometheus_metric_instantiation
+
+        return None
+
     def get_method_signature_hook(
         self, fullname: str
     ) -> Optional[Callable[[MethodSigContext], CallableType]]:
@@ -65,6 +88,85 @@ def get_method_signature_hook(
         return None
 
 
+def check_prometheus_metric_instantiation(ctx: FunctionSigContext) -> CallableType:
+    """
+    Ensure that the `prometheus_client` metrics include the `SERVER_NAME_LABEL` label
+    when instantiated.
+
+    This is important because we support multiple Synapse instances running in the same
+    process, where all metrics share a single global `REGISTRY`. The `server_name` label
+    ensures metrics are correctly separated by homeserver.
+
+    There are also some metrics that apply at the process level, such as CPU usage,
+    Python garbage collection, Twisted reactor tick time which shouldn't have the
+    `SERVER_NAME_LABEL`. In those cases, use use a type ignore comment to disable the
+    check, e.g. `# type: ignore[missing-server-name-label]`.
+    """
+    # The true signature, this isn't being modified so this is what will be returned.
+    signature: CallableType = ctx.default_signature
+
+    # Sanity check the arguments are still as expected in this version of
+    # `prometheus_client`. ex. `Counter(name, documentation, labelnames, ...)`
+    #
+    # `signature.arg_names` should be: ["name", "documentation", "labelnames", ...]
+    if len(signature.arg_names) < 3 or signature.arg_names[2] != "labelnames":
+        ctx.api.fail(
+            f"Expected the 3rd argument of {signature.name} to be 'labelnames', but got "
+            f"{signature.arg_names[2]}",
+            ctx.context,
+        )
+        return signature
+
+    # Ensure mypy is passing the correct number of arguments because we are doing some
+    # dirty indexing into `ctx.args` later on.
+    assert len(ctx.args) == len(signature.arg_names), (
+        f"Expected the list of arguments in the {signature.name} signature ({len(signature.arg_names)})"
+        f"to match the number of arguments from the function signature context ({len(ctx.args)})"
+    )
+
+    # Check if the `labelnames` argument includes `SERVER_NAME_LABEL`
+    #
+    # `ctx.args` should look like this:
+    # ```
+    # [
+    #     [StrExpr("name")],
+    #     [StrExpr("documentation")],
+    #     [ListExpr([StrExpr("label1"), StrExpr("label2")])]
+    #     ...
+    # ]
+    # ```
+    labelnames_arg_expression = ctx.args[2][0] if len(ctx.args[2]) > 0 else None
+    if isinstance(labelnames_arg_expression, ListExpr):
+        # Check if the `labelnames` argument includes the `server_name` label (`SERVER_NAME_LABEL`).
+        for labelname_expression in labelnames_arg_expression.items:
+            if (
+                isinstance(labelname_expression, NameExpr)
+                and labelname_expression.fullname == "synapse.metrics.SERVER_NAME_LABEL"
+            ):
+                # Found the `SERVER_NAME_LABEL`, all good!
+                break
+        else:
+            ctx.api.fail(
+                f"Expected {signature.name} to include `SERVER_NAME_LABEL` in the list of labels. "
+                "If this is a process-level metric (vs homeserver-level), use a type ignore comment "
+                "to disable this check.",
+                ctx.context,
+                code=PROMETHEUS_METRIC_MISSING_SERVER_NAME_LABEL,
+            )
+    else:
+        ctx.api.fail(
+            f"Expected the `labelnames` argument of {signature.name} to be a list of label names "
+            f"(including `SERVER_NAME_LABEL`), but got {labelnames_arg_expression}. "
+            "If this is a process-level metric (vs homeserver-level), use a type ignore comment "
+            "to disable this check.",
+            ctx.context,
+            code=PROMETHEUS_METRIC_MISSING_SERVER_NAME_LABEL,
+        )
+        return signature
+
+    return signature
+
+
 def _get_true_return_type(signature: CallableType) -> mypy.types.Type:
     """
     Get the "final" return type of a callable which might return an Awaitable/Deferred.

synapse/appservice/api.py

@@ -48,6 +48,7 @@
 from synapse.events.utils import SerializeEventConfig, serialize_event
 from synapse.http.client import SimpleHttpClient, is_unknown_endpoint
 from synapse.logging import opentracing
+from synapse.metrics import SERVER_NAME_LABEL
 from synapse.types import DeviceListUpdates, JsonDict, JsonMapping, ThirdPartyInstanceID
 from synapse.util.caches.response_cache import ResponseCache
 
@@ -59,29 +60,31 @@
 sent_transactions_counter = Counter(
     "synapse_appservice_api_sent_transactions",
     "Number of /transactions/ requests sent",
-    ["service"],
+    labelnames=["service", SERVER_NAME_LABEL],
 )
 
 failed_transactions_counter = Counter(
     "synapse_appservice_api_failed_transactions",
     "Number of /transactions/ requests that failed to send",
-    ["service"],
+    labelnames=["service", SERVER_NAME_LABEL],
 )
 
 sent_events_counter = Counter(
-    "synapse_appservice_api_sent_events", "Number of events sent to the AS", ["service"]
+    "synapse_appservice_api_sent_events",
+    "Number of events sent to the AS",
+    labelnames=["service", SERVER_NAME_LABEL],
 )
 
 sent_ephemeral_counter = Counter(
     "synapse_appservice_api_sent_ephemeral",
     "Number of ephemeral events sent to the AS",
-    ["service"],
+    labelnames=["service", SERVER_NAME_LABEL],
 )
 
 sent_todevice_counter = Counter(
     "synapse_appservice_api_sent_todevice",
     "Number of todevice messages sent to the AS",
-    ["service"],
+    labelnames=["service", SERVER_NAME_LABEL],
 )
 
 HOUR_IN_MS = 60 * 60 * 1000
@@ -382,6 +385,7 @@ async def push_bulk(
                     "left": list(device_list_summary.left),
                 }
 
+        labels = {"service": service.id, SERVER_NAME_LABEL: self.server_name}
         try:
             args = None
             if self.config.use_appservice_legacy_authorization:
@@ -399,10 +403,10 @@ async def push_bulk(
                     service.url,
                     [event.get("event_id") for event in events],
                 )
-            sent_transactions_counter.labels(service.id).inc()
-            sent_events_counter.labels(service.id).inc(len(serialized_events))
-            sent_ephemeral_counter.labels(service.id).inc(len(ephemeral))
-            sent_todevice_counter.labels(service.id).inc(len(to_device_messages))
+            sent_transactions_counter.labels(**labels).inc()
+            sent_events_counter.labels(**labels).inc(len(serialized_events))
+            sent_ephemeral_counter.labels(**labels).inc(len(ephemeral))
+            sent_todevice_counter.labels(**labels).inc(len(to_device_messages))
             return True
         except CodeMessageException as e:
             logger.warning(
@@ -421,7 +425,7 @@ async def push_bulk(
                 ex.args,
                 exc_info=logger.isEnabledFor(logging.DEBUG),
             )
-        failed_transactions_counter.labels(service.id).inc()
+        failed_transactions_counter.labels(**labels).inc()
         return False
 
     async def claim_client_keys(

synapse/federation/federation_client.py

@@ -74,6 +74,7 @@
 from synapse.http.client import is_unknown_endpoint
 from synapse.http.types import QueryParams
 from synapse.logging.opentracing import SynapseTags, log_kv, set_tag, tag_args, trace
+from synapse.metrics import SERVER_NAME_LABEL
 from synapse.types import JsonDict, StrCollection, UserID, get_domain_from_id
 from synapse.types.handlers.policy_server import RECOMMENDATION_OK, RECOMMENDATION_SPAM
 from synapse.util.async_helpers import concurrently_execute
@@ -85,7 +86,9 @@
 
 logger = logging.getLogger(__name__)
 
-sent_queries_counter = Counter("synapse_federation_client_sent_queries", "", ["type"])
+sent_queries_counter = Counter(
+    "synapse_federation_client_sent_queries", "", labelnames=["type", SERVER_NAME_LABEL]
+)
 
 
 PDU_RETRY_TIME_MS = 1 * 60 * 1000
@@ -209,7 +212,10 @@ async def make_query(
         Returns:
             The JSON object from the response
         """
-        sent_queries_counter.labels(query_type).inc()
+        sent_queries_counter.labels(
+            type=query_type,
+            **{SERVER_NAME_LABEL: self.server_name},
+        ).inc()
 
         return await self.transport_layer.make_query(
             destination,
@@ -231,7 +237,10 @@ async def query_client_keys(
         Returns:
             The JSON object from the response
         """
-        sent_queries_counter.labels("client_device_keys").inc()
+        sent_queries_counter.labels(
+            type="client_device_keys",
+            **{SERVER_NAME_LABEL: self.server_name},
+        ).inc()
         return await self.transport_layer.query_client_keys(
             destination, content, timeout
         )
@@ -242,7 +251,10 @@ async def query_user_devices(
         """Query the device keys for a list of user ids hosted on a remote
         server.
         """
-        sent_queries_counter.labels("user_devices").inc()
+        sent_queries_counter.labels(
+            type="user_devices",
+            **{SERVER_NAME_LABEL: self.server_name},
+        ).inc()
         return await self.transport_layer.query_user_devices(
             destination, user_id, timeout
         )
@@ -264,7 +276,10 @@ async def claim_client_keys(
         Returns:
             The JSON object from the response
         """
-        sent_queries_counter.labels("client_one_time_keys").inc()
+        sent_queries_counter.labels(
+            type="client_one_time_keys",
+            **{SERVER_NAME_LABEL: self.server_name},
+        ).inc()
 
         # Convert the query with counts into a stable and unstable query and check
         # if attempting to claim more than 1 OTK.

synapse/federation/federation_server.py

@@ -82,6 +82,7 @@
     tag_args,
     trace,
 )
+from synapse.metrics import SERVER_NAME_LABEL
 from synapse.metrics.background_process_metrics import wrap_as_background_process
 from synapse.replication.http.federation import (
     ReplicationFederationSendEduRestServlet,
@@ -104,12 +105,18 @@
 
 logger = logging.getLogger(__name__)
 
-received_pdus_counter = Counter("synapse_federation_server_received_pdus", "")
+received_pdus_counter = Counter(
+    "synapse_federation_server_received_pdus", "", labelnames=[SERVER_NAME_LABEL]
+)
 
-received_edus_counter = Counter("synapse_federation_server_received_edus", "")
+received_edus_counter = Counter(
+    "synapse_federation_server_received_edus", "", labelnames=[SERVER_NAME_LABEL]
+)
 
 received_queries_counter = Counter(
-    "synapse_federation_server_received_queries", "", ["type"]
+    "synapse_federation_server_received_queries",
+    "",
+    labelnames=["type", SERVER_NAME_LABEL],
 )
 
 pdu_process_time = Histogram(
@@ -434,7 +441,9 @@ async def _handle_pdus_in_txn(
             report back to the sending server.
         """
 
-        received_pdus_counter.inc(len(transaction.pdus))
+        received_pdus_counter.labels(**{SERVER_NAME_LABEL: self.server_name}).inc(
+            len(transaction.pdus)
+        )
 
         origin_host, _ = parse_server_name(origin)
 
@@ -553,7 +562,7 @@ async def _handle_edus_in_txn(self, origin: str, transaction: Transaction) -> No
         """Process the EDUs in a received transaction."""
 
         async def _process_edu(edu_dict: JsonDict) -> None:
-            received_edus_counter.inc()
+            received_edus_counter.labels(**{SERVER_NAME_LABEL: self.server_name}).inc()
 
             edu = Edu(
                 origin=origin,
@@ -668,7 +677,10 @@ async def on_pdu_request(
     async def on_query_request(
         self, query_type: str, args: Dict[str, str]
     ) -> Tuple[int, Dict[str, Any]]:
-        received_queries_counter.labels(query_type).inc()
+        received_queries_counter.labels(
+            type=query_type,
+            **{SERVER_NAME_LABEL: self.server_name},
+        ).inc()
         resp = await self.registry.on_query(query_type, args)
         return 200, resp
 

synapse/federation/sender/__init__.py

@@ -160,6 +160,7 @@
 from synapse.federation.units import Edu
 from synapse.logging.context import make_deferred_yieldable, run_in_background
 from synapse.metrics import (
+    SERVER_NAME_LABEL,
     LaterGauge,
     event_processing_loop_counter,
     event_processing_loop_room_count,
@@ -189,11 +190,13 @@
 sent_pdus_destination_dist_count = Counter(
     "synapse_federation_client_sent_pdu_destinations_count",
     "Number of PDUs queued for sending to one or more destinations",
+    labelnames=[SERVER_NAME_LABEL],
 )
 
 sent_pdus_destination_dist_total = Counter(
     "synapse_federation_client_sent_pdu_destinations",
     "Total number of PDUs queued for sending across all destinations",
+    labelnames=[SERVER_NAME_LABEL],
 )
 
 # Time (in s) to wait before trying to wake up destinations that have
@@ -708,13 +711,19 @@ async def handle_room_events(events: List[EventBase]) -> None:
                         "federation_sender"
                     ).set(ts)
 
-                    events_processed_counter.inc(len(event_entries))
+                    events_processed_counter.labels(
+                        **{SERVER_NAME_LABEL: self.server_name}
+                    ).inc(len(event_entries))
 
-                    event_processing_loop_room_count.labels("federation_sender").inc(
-                        len(events_by_room)
-                    )
+                    event_processing_loop_room_count.labels(
+                        name="federation_sender",
+                        **{SERVER_NAME_LABEL: self.server_name},
+                    ).inc(len(events_by_room))
 
-                event_processing_loop_counter.labels("federation_sender").inc()
+                event_processing_loop_counter.labels(
+                    name="federation_sender",
+                    **{SERVER_NAME_LABEL: self.server_name},
+                ).inc()
 
                 synapse.metrics.event_processing_positions.labels(
                     "federation_sender"
@@ -735,8 +744,12 @@ async def _send_pdu(self, pdu: EventBase, destinations: Iterable[str]) -> None:
         if not destinations:
             return
 
-        sent_pdus_destination_dist_total.inc(len(destinations))
-        sent_pdus_destination_dist_count.inc()
+        sent_pdus_destination_dist_total.labels(
+            **{SERVER_NAME_LABEL: self.server_name}
+        ).inc(len(destinations))
+        sent_pdus_destination_dist_count.labels(
+            **{SERVER_NAME_LABEL: self.server_name}
+        ).inc()
 
         assert pdu.internal_metadata.stream_ordering