Add prometheus metrics to track federation delays (#8430)

Add a pair of federation metrics to track the delays in sending PDUs to/from 
particular servers.

2 files changed, 45 insertions, 1 deletions

diff --git a/synapse/federation/federation_server.py b/synapse/federation/federation_server.py
index 2dcd081cbc..24329dd0e3 100644
--- a/synapse/federation/federation_server.py
+++ b/synapse/federation/federation_server.py
@@ -28,7 +28,7 @@ from typing import (
     Union,
 )
 
-from prometheus_client import Counter, Histogram
+from prometheus_client import Counter, Gauge, Histogram
 
 from twisted.internet import defer
 from twisted.internet.abstract import isIPAddress
@@ -88,6 +88,13 @@ pdu_process_time = Histogram(
 )
 
 
+last_pdu_age_metric = Gauge(
+    "synapse_federation_last_received_pdu_age",
+    "The age (in seconds) of the last PDU successfully received from the given domain",
+    labelnames=("server_name",),
+)
+
+
 class FederationServer(FederationBase):
     def __init__(self, hs):
         super().__init__(hs)
@@ -118,6 +125,10 @@ class FederationServer(FederationBase):
             hs, "state_ids_resp", timeout_ms=30000
         )
 
+        self._federation_metrics_domains = (
+            hs.get_config().federation.federation_metrics_domains
+        )
+
     async def on_backfill_request(
         self, origin: str, room_id: str, versions: List[str], limit: int
     ) -> Tuple[int, Dict[str, Any]]:
@@ -262,7 +273,11 @@ class FederationServer(FederationBase):
 
         pdus_by_room = {}  # type: Dict[str, List[EventBase]]
 
+        newest_pdu_ts = 0
+
         for p in transaction.pdus:  # type: ignore
+            # FIXME (richardv): I don't think this works:
+            #  https://github.com/matrix-org/synapse/issues/8429
             if "unsigned" in p:
                 unsigned = p["unsigned"]
                 if "age" in unsigned:
@@ -300,6 +315,9 @@ class FederationServer(FederationBase):
             event = event_from_pdu_json(p, room_version)
             pdus_by_room.setdefault(room_id, []).append(event)
 
+            if event.origin_server_ts > newest_pdu_ts:
+                newest_pdu_ts = event.origin_server_ts
+
         pdu_results = {}
 
         # we can process different rooms in parallel (which is useful if they
@@ -340,6 +358,10 @@ class FederationServer(FederationBase):
             process_pdus_for_room, pdus_by_room.keys(), TRANSACTION_CONCURRENCY_LIMIT
         )
 
+        if newest_pdu_ts and origin in self._federation_metrics_domains:
+            newest_pdu_age = self._clock.time_msec() - newest_pdu_ts
+            last_pdu_age_metric.labels(server_name=origin).set(newest_pdu_age / 1000)
+
         return pdu_results
 
     async def _handle_edus_in_txn(self, origin: str, transaction: Transaction):
diff --git a/synapse/federation/sender/transaction_manager.py b/synapse/federation/sender/transaction_manager.py
index c84072ab73..3e07f925e0 100644
--- a/synapse/federation/sender/transaction_manager.py
+++ b/synapse/federation/sender/transaction_manager.py
@@ -15,6 +15,8 @@
 import logging
 from typing import TYPE_CHECKING, List
 
+from prometheus_client import Gauge
+
 from synapse.api.errors import HttpResponseException
 from synapse.events import EventBase
 from synapse.federation.persistence import TransactionActions
@@ -34,6 +36,12 @@ if TYPE_CHECKING:
 
 logger = logging.getLogger(__name__)
 
+last_pdu_age_metric = Gauge(
+    "synapse_federation_last_sent_pdu_age",
+    "The age (in seconds) of the last PDU successfully sent to the given domain",
+    labelnames=("server_name",),
+)
+
 
 class TransactionManager:
     """Helper class which handles building and sending transactions
@@ -48,6 +56,10 @@ class TransactionManager:
         self._transaction_actions = TransactionActions(self._store)
         self._transport_layer = hs.get_federation_transport_client()
 
+        self._federation_metrics_domains = (
+            hs.get_config().federation.federation_metrics_domains
+        )
+
         # HACK to get unique tx id
         self._next_txn_id = int(self.clock.time_msec())
 
@@ -119,6 +131,9 @@ class TransactionManager:
 
             # FIXME (erikj): This is a bit of a hack to make the Pdu age
             # keys work
+            # FIXME (richardv): I also believe it no longer works. We (now?) store
+            #  "age_ts" in "unsigned" rather than at the top level. See
+            #  https://github.com/matrix-org/synapse/issues/8429.
             def json_data_cb():
                 data = transaction.get_dict()
                 now = int(self.clock.time_msec())
@@ -167,5 +182,12 @@ class TransactionManager:
                     )
                 success = False
 
+            if success and pdus and destination in self._federation_metrics_domains:
+                last_pdu = pdus[-1]
+                last_pdu_age = self.clock.time_msec() - last_pdu.origin_server_ts
+                last_pdu_age_metric.labels(server_name=destination).set(
+                    last_pdu_age / 1000
+                )
+
             set_tag(tags.ERROR, not success)
             return success