kvserver: expose a per-store replica cpu histogram metric #138672
Assignees
Labels
C-enhancement
Solution expected to add code/behavior + preserve backward-compat (pg compat issues are exception)
T-observability
Comments
angles-n-daemons
added
the
C-enhancement
angles-n-daemons
changed the title
angles-n-daemons
added a commit
to angles-n-daemons/cockroach
that referenced
this issue
Jan 16, 2025
One of our goals with adding more hotspot telemetry is to better understand what's happening to a cluster when it has a hotspot. Today this is possible real time, but information is limited when trying to understand hotspots from the past. We currently have a log for the hot ranges in a cluster, which can be enabled to periodically report the hot ranges, but to limit the output it runs infrequently, and therefore is likely to miss temporal, or short lived hotspots. In the replica deciders, there already exists some functionality for determining when a specific replica is the target of an unbalanced portion of the system's load. What this change does is allow for other parts of the system (namely the hot range logger) to subscribe to when that tipping point is reached. The following change will link the hot range logger to this new notification system, so that temporal hotspots can be better examined. Fixes: cockroachdb#138672 Epic: CRDB-43150 Release note: none
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
In the interest of better troubleshooting hotspots, we're aiming to expose more observability around range usage and load.
To be considered complete, for each store, when
updateReplicationGaugesis called, update this new metric to a histogram of seen cpu values.Epic: CRDB-43150
Jira issue: CRDB-46300
The text was updated successfully, but these errors were encountered: