-
Type:
Improvement
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
-
Storage Engines - Server Integration
-
SESINickBeckenhome 2026-12-01
-
0
-
None
-
None
-
None
-
None
-
None
-
None
-
None
In BF-46704 we saw slowness in the appendLogResponseArrivalLatency which is when the log server sends us an acknowledgement but PALI is slow to receive and apply that acknowledgement causing slowness in our write path. This happens on a CPU cliff when we are almost at CPU capacity, we end up waiting on updateCommittedOpTimeFromLogServer to update and it gets held up because our threads are busy and this takes a global lock. Due to this slowness I suspect this is what is causing the slowness in our write path and attributing to a high appendLogResponseArrivalLatency value. See the BF for more details about the parca run and CPU distribution here.
Investigate this issue and see how we resolve this slowness, I think we can look taking this updateCommittedOpTimeFromLogServer off the response thread (receiving and acknowledging entries from LS) so it doesn't hold up our writes.