Start fewer opentracing spans (#8640)

#8567 started a span for every background process. This is good as it means all Synapse code that gets run should be in a span (unless in the sentinel logging context), but it means we generate about 15x the number of spans as we did previously.

This PR attempts to reduce that number by a) not starting one for send commands to Redis, and b) deferring starting background processes until after we're sure they're necessary.

I don't really know how much this will help.

1 files changed, 16 insertions, 2 deletions

diff --git a/synapse/push/pusherpool.py b/synapse/push/pusherpool.py
index 0080c68ce2..f325964983 100644
--- a/synapse/push/pusherpool.py
+++ b/synapse/push/pusherpool.py
@@ -19,7 +19,10 @@ from typing import TYPE_CHECKING, Dict, Union
 
 from prometheus_client import Gauge
 
-from synapse.metrics.background_process_metrics import run_as_background_process
+from synapse.metrics.background_process_metrics import (
+    run_as_background_process,
+    wrap_as_background_process,
+)
 from synapse.push import PusherConfigException
 from synapse.push.emailpusher import EmailPusher
 from synapse.push.httppusher import HttpPusher
@@ -187,7 +190,7 @@ class PusherPool:
                 )
                 await self.remove_pusher(p["app_id"], p["pushkey"], p["user_name"])
 
-    async def on_new_notifications(self, max_token: RoomStreamToken):
+    def on_new_notifications(self, max_token: RoomStreamToken):
         if not self.pushers:
             # nothing to do here.
             return
@@ -201,6 +204,17 @@ class PusherPool:
             # Nothing to do
             return
 
+        # We only start a new background process if necessary rather than
+        # optimistically (to cut down on overhead).
+        self._on_new_notifications(max_token)
+
+    @wrap_as_background_process("on_new_notifications")
+    async def _on_new_notifications(self, max_token: RoomStreamToken):
+        # We just use the minimum stream ordering and ignore the vector clock
+        # component. This is safe to do as long as we *always* ignore the vector
+        # clock components.
+        max_stream_id = max_token.stream
+
         prev_stream_id = self._last_room_stream_id_seen
         self._last_room_stream_id_seen = max_stream_id