Properly update retry_last_ts when hitting the maximum retry interval

changelog.d/16156.bugfix

@@ -0,0 +1 @@
+Fix a bug introduced in 1.87 where synapse would send an excessive amount of federation requests to servers which have been offline for a long time. Contributed by Nico.

synapse/storage/databases/main/transactions.py

@@ -242,6 +242,8 @@ def _set_destination_retry_timings_txn(
     ) -> None:
         # Upsert retry time interval if retry_interval is zero (i.e. we're
         # resetting it) or greater than the existing retry interval.
+        # We also upsert when the new retry interval is the same as the existing one,
+        # since it will be the case when `destination_max_retry_interval` is reached.
         #
         # WARNING: This is executed in autocommit, so we shouldn't add any more
         # SQL calls in here (without being very careful).
@@ -257,7 +259,7 @@ def _set_destination_retry_timings_txn(
                 WHERE
                     EXCLUDED.retry_interval = 0
                     OR destinations.retry_interval IS NULL
-                    OR destinations.retry_interval < EXCLUDED.retry_interval
+                    OR destinations.retry_interval <= EXCLUDED.retry_interval
         """
 
         txn.execute(sql, (destination, failure_ts, retry_last_ts, retry_interval))

tests/util/test_retryutils.py

@@ -108,3 +108,54 @@ def test_limiter(self) -> None:
 
         new_timings = self.get_success(store.get_destination_retry_timings("test_dest"))
         self.assertIsNone(new_timings)
+
+    def test_max_retry_interval(self) -> None:
+        """Test that `destination_max_retry_interval` setting works as expected"""
+        store = self.hs.get_datastores().main
+
+        destination_max_retry_interval_ms = (
+            self.hs.config.federation.destination_max_retry_interval_ms
+        )
+
+        self.get_success(get_retry_limiter("test_dest", self.clock, store))
+        self.pump(1)
+
+        failure_ts = self.clock.time_msec()
+
+        # Simulate reaching destination_max_retry_interval
+        self.get_success(
+            store.set_destination_retry_timings(
+                "test_dest",
+                failure_ts=failure_ts,
+                retry_last_ts=failure_ts,
+                retry_interval=destination_max_retry_interval_ms,
+            )
+        )
+
+        # Check it fails
+        self.get_failure(
+            get_retry_limiter("test_dest", self.clock, store), NotRetryingDestination
+        )
+
+        # Get past retry_interval and we can try again, and still throw an error to continue the backoff
+        self.reactor.advance(destination_max_retry_interval_ms / 1000 + 1)
+        limiter = self.get_success(get_retry_limiter("test_dest", self.clock, store))
+        self.pump(1)
+        try:
+            with limiter:
+                self.pump(1)
+                raise AssertionError("argh")
+        except AssertionError:
+            pass
+
+        self.pump()
+
+        # retry_interval does not increase and stays at destination_max_retry_interval_ms
+        new_timings = self.get_success(store.get_destination_retry_timings("test_dest"))
+        assert new_timings is not None
+        self.assertEqual(new_timings.retry_interval, destination_max_retry_interval_ms)
+
+        # Check it fails
+        self.get_failure(
+            get_retry_limiter("test_dest", self.clock, store), NotRetryingDestination
+        )