Re-introduce: Fix LaterGauge metrics to collect from all servers

changelog.d/18791.misc

@@ -0,0 +1 @@
+Fix `LaterGauge` metrics to collect from all servers.

synapse/_scripts/generate_workers_map.py

@@ -153,9 +153,13 @@ def get_registered_paths_for_default(
     """
 
     hs = MockHomeserver(base_config, worker_app)
+
     # TODO We only do this to avoid an error, but don't need the database etc
     hs.setup()
-    return get_registered_paths_for_hs(hs)
+    registered_paths = get_registered_paths_for_hs(hs)
+    hs.cleanup()
+
+    return registered_paths
 
 
 def elide_http_methods_if_unconflicting(

synapse/_scripts/synapse_port_db.py

@@ -99,6 +99,7 @@
 from synapse.storage.prepare_database import prepare_database
 from synapse.types import ISynapseReactor
 from synapse.util import SYNAPSE_VERSION, Clock
+from synapse.util.stringutils import random_string
 
 # Cast safety: Twisted does some naughty magic which replaces the
 # twisted.internet.reactor module with a Reactor instance at runtime.
@@ -323,13 +324,17 @@ def __init__(self, config: HomeServerConfig):
         self.config = config
         self.hostname = config.server.server_name
         self.version_string = SYNAPSE_VERSION
+        self.instance_id = random_string(5)
 
     def get_clock(self) -> Clock:
         return self.clock
 
     def get_reactor(self) -> ISynapseReactor:
         return reactor
 
+    def get_instance_id(self) -> str:
+        return self.instance_id
+
     def get_instance_name(self) -> str:
         return "master"
 
@@ -685,7 +690,15 @@ def build_db_store(
             )
             prepare_database(db_conn, engine, config=self.hs_config)
             # Type safety: ignore that we're using Mock homeservers here.
-            store = Store(DatabasePool(hs, db_config, engine), db_conn, hs)  # type: ignore[arg-type]
+            store = Store(
+                DatabasePool(
+                    hs,  # type: ignore[arg-type]
+                    db_config,
+                    engine,
+                ),
+                db_conn,
+                hs,  # type: ignore[arg-type]
+            )
             db_conn.commit()
 
         return store

synapse/federation/send_queue.py

@@ -37,6 +37,7 @@
 """
 
 import logging
+from enum import Enum
 from typing import (
     TYPE_CHECKING,
     Dict,
@@ -67,6 +68,25 @@
 logger = logging.getLogger(__name__)
 
 
+class QueueNames(str, Enum):
+    PRESENCE_MAP = "presence_map"
+    KEYED_EDU = "keyed_edu"
+    KEYED_EDU_CHANGED = "keyed_edu_changed"
+    EDUS = "edus"
+    POS_TIME = "pos_time"
+    PRESENCE_DESTINATIONS = "presence_destinations"
+
+
+queue_name_to_gauge_map: Dict[QueueNames, LaterGauge] = {}
+
+for queue_name in QueueNames:
+    queue_name_to_gauge_map[queue_name] = LaterGauge(
+        name=f"synapse_federation_send_queue_{queue_name.value}_size",
+        desc="",
+        labelnames=[SERVER_NAME_LABEL],
+    )
+
+
 class FederationRemoteSendQueue(AbstractFederationSender):
     """A drop in replacement for FederationSender"""
 
@@ -111,23 +131,16 @@ def __init__(self, hs: "HomeServer"):
         # we make a new function, so we need to make a new function so the inner
         # lambda binds to the queue rather than to the name of the queue which
         # changes. ARGH.
-        def register(name: str, queue: Sized) -> None:
-            LaterGauge(
-                name="synapse_federation_send_queue_%s_size" % (queue_name,),
-                desc="",
-                labelnames=[SERVER_NAME_LABEL],
-                caller=lambda: {(self.server_name,): len(queue)},
+        def register(queue_name: QueueNames, queue: Sized) -> None:
+            queue_name_to_gauge_map[queue_name].register_hook(
+                homeserver_instance_id=hs.get_instance_id(),
+                hook=lambda: {(self.server_name,): len(queue)},
             )
 
-        for queue_name in [
-            "presence_map",
-            "keyed_edu",
-            "keyed_edu_changed",
-            "edus",
-            "pos_time",
-            "presence_destinations",
-        ]:
-            register(queue_name, getattr(self, queue_name))
+        for queue_name in QueueNames:
+            queue = getattr(self, queue_name.value)
+            assert isinstance(queue, Sized)
+            register(queue_name, queue=queue)
 
         self.clock.looping_call(self._clear_queue, 30 * 1000)
 

synapse/federation/sender/__init__.py

@@ -199,6 +199,24 @@
     labelnames=[SERVER_NAME_LABEL],
 )
 
+transaction_queue_pending_destinations_gauge = LaterGauge(
+    name="synapse_federation_transaction_queue_pending_destinations",
+    desc="",
+    labelnames=[SERVER_NAME_LABEL],
+)
+
+transaction_queue_pending_pdus_gauge = LaterGauge(
+    name="synapse_federation_transaction_queue_pending_pdus",
+    desc="",
+    labelnames=[SERVER_NAME_LABEL],
+)
+
+transaction_queue_pending_edus_gauge = LaterGauge(
+    name="synapse_federation_transaction_queue_pending_edus",
+    desc="",
+    labelnames=[SERVER_NAME_LABEL],
+)
+
 # Time (in s) to wait before trying to wake up destinations that have
 # catch-up outstanding.
 # Please note that rate limiting still applies, so while the loop is
@@ -398,34 +416,27 @@ def __init__(self, hs: "HomeServer"):
         # map from destination to PerDestinationQueue
         self._per_destination_queues: Dict[str, PerDestinationQueue] = {}
 
-        LaterGauge(
-            name="synapse_federation_transaction_queue_pending_destinations",
-            desc="",
-            labelnames=[SERVER_NAME_LABEL],
-            caller=lambda: {
+        transaction_queue_pending_destinations_gauge.register_hook(
+            homeserver_instance_id=hs.get_instance_id(),
+            hook=lambda: {
                 (self.server_name,): sum(
                     1
                     for d in self._per_destination_queues.values()
                     if d.transmission_loop_running
                 )
             },
         )
-
-        LaterGauge(
-            name="synapse_federation_transaction_queue_pending_pdus",
-            desc="",
-            labelnames=[SERVER_NAME_LABEL],
-            caller=lambda: {
+        transaction_queue_pending_pdus_gauge.register_hook(
+            homeserver_instance_id=hs.get_instance_id(),
+            hook=lambda: {
                 (self.server_name,): sum(
                     d.pending_pdu_count() for d in self._per_destination_queues.values()
                 )
             },
         )
-        LaterGauge(
-            name="synapse_federation_transaction_queue_pending_edus",
-            desc="",
-            labelnames=[SERVER_NAME_LABEL],
-            caller=lambda: {
+        transaction_queue_pending_edus_gauge.register_hook(
+            homeserver_instance_id=hs.get_instance_id(),
+            hook=lambda: {
                 (self.server_name,): sum(
                     d.pending_edu_count() for d in self._per_destination_queues.values()
                 )

synapse/handlers/presence.py

@@ -173,6 +173,18 @@
     labelnames=["locality", "from", "to", SERVER_NAME_LABEL],
 )
 
+presence_user_to_current_state_size_gauge = LaterGauge(
+    name="synapse_handlers_presence_user_to_current_state_size",
+    desc="",
+    labelnames=[SERVER_NAME_LABEL],
+)
+
+presence_wheel_timer_size_gauge = LaterGauge(
+    name="synapse_handlers_presence_wheel_timer_size",
+    desc="",
+    labelnames=[SERVER_NAME_LABEL],
+)
+
 # If a user was last active in the last LAST_ACTIVE_GRANULARITY, consider them
 # "currently_active"
 LAST_ACTIVE_GRANULARITY = 60 * 1000
@@ -779,11 +791,9 @@ def __init__(self, hs: "HomeServer"):
             EduTypes.PRESENCE, self.incoming_presence
         )
 
-        LaterGauge(
-            name="synapse_handlers_presence_user_to_current_state_size",
-            desc="",
-            labelnames=[SERVER_NAME_LABEL],
-            caller=lambda: {(self.server_name,): len(self.user_to_current_state)},
+        presence_user_to_current_state_size_gauge.register_hook(
+            homeserver_instance_id=hs.get_instance_id(),
+            hook=lambda: {(self.server_name,): len(self.user_to_current_state)},
         )
 
         # The per-device presence state, maps user to devices to per-device presence state.
@@ -882,11 +892,9 @@ def __init__(self, hs: "HomeServer"):
                 60 * 1000,
             )
 
-        LaterGauge(
-            name="synapse_handlers_presence_wheel_timer_size",
-            desc="",
-            labelnames=[SERVER_NAME_LABEL],
-            caller=lambda: {(self.server_name,): len(self.wheel_timer)},
+        presence_wheel_timer_size_gauge.register_hook(
+            homeserver_instance_id=hs.get_instance_id(),
+            hook=lambda: {(self.server_name,): len(self.wheel_timer)},
         )
 
         # Used to handle sending of presence to newly joined users/servers

synapse/http/request_metrics.py

@@ -164,11 +164,13 @@ def _get_in_flight_counts() -> Mapping[Tuple[str, ...], int]:
     return counts
 
 
-LaterGauge(
+in_flight_requests = LaterGauge(
     name="synapse_http_server_in_flight_requests_count",
     desc="",
     labelnames=["method", "servlet", SERVER_NAME_LABEL],
-    caller=_get_in_flight_counts,
+)
+in_flight_requests.register_hook(
+    homeserver_instance_id=None, hook=_get_in_flight_counts
 )