Total memory allocation is much higher than mem_tracker_Read_Buffer_Inbound_RPC_Sending in case it is flooded with queued RPC messages #3041
Assignees
Labels
kind/enhancement
This is an enhancement of an existing feature
Comments
ttyusupov
added
the
kind/enhancement
ttyusupov
added a commit
that referenced
this issue
Dec 16, 2019
Summary: We only accounted serialized message size when tracking `TcpStream::sending_` queue memory consumption, but for example, for 4-bytes heartbeat message, overall memory consumption was roughly about 120 bytes. And in a scenario when we had sending queue flooded with heartbeats we were tracking only 4*N instead of ~120*N. This also would give us unreliable memory tracking results when we have a lot of other small enough messages sitting in `TcpStream::sending_` queue. Added `TestRpc.SendingQueueMemoryUsage`. Test Plan: Jenkins Reviewers: mikhail, bogdan, sergei Reviewed By: sergei Subscribers: ybase Differential Revision: https://phabricator.dev.yugabyte.com/D7629
This was referenced Oct 21, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
In the case of RPC sending queue is flooded with small messages (like in #2964) total memory allocated is much higher than
mem_tracker_Read_Buffer_Inbound_RPC_Sendingand there are no other noticeable tracked memory allocations.For example, in one scenario
mem_tracker_Read_Buffer_Inbound_RPC_Sendingwas taking up to ~250MB and everything else is less than ~10MB, but total memory allocation was more than 6GB.The text was updated successfully, but these errors were encountered: