status: expose Go heap object count metrics #160607
Conversation
Add two new metrics sourced from Go's runtime/metrics package: - sys.go.heap.objects: number of objects (live + unswept) on the heap, from /gc/heap/objects:objects - sys.go.heap.livebytes: bytes of live heap objects marked by the previous GC, from /gc/heap/live:bytes These complement the existing heap byte metrics and provide visibility into object allocation patterns, which can be useful for diagnosing GC pressure from high object churn. Release note: None
Add sys.gc.total.ns metric sourced from Go's /cpu/classes/gc/total:cpu-seconds runtime metric. This provides visibility into total CPU time spent on GC tasks, complementing the existing sys.gc.assist.ns which only captures mutator assist time. Release note: None
|
It looks like your PR touches production code but doesn't add or edit any test code. Did you consider adding tests to your PR? 🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf. |
|
This change is |
This comment was marked as resolved.
This comment was marked as resolved.
tbg
requested review from
golgeek and
shailendra-patel
and removed request for
a team
Metrics change detectedThis PR adds or updates one or more CRDB metrics. If you want these metrics to be exported by CRDB Cloud clusters to Internal CRL Datadog and/or included in the customer metric export integration (Essential metrics for standard deployment, and Essential metrics for advanced deployment), refer to this Installation and Usage guide of a CLI tool that syncs the metric mappings in managed-service. Run this CLI tool after your CRDB PR is merged.
Note: Your metric will appear in Internal CRL Datadog only after the managed-service PR merges and the new OTel configuration rolls out to at least one cluster running a CRDB build that includes this metric. Docs: cockroach-metric-sync Questions: reach out to @obs-india-prs |
angeladietz
left a comment
angeladietz
left a comment
There was a problem hiding this comment.
@angeladietz made 1 comment.
Reviewable status:complete! 1 of 0 LGTMs obtained (waiting on @dhartunian, @golgeek, and @shailendra-patel).
|
Test failure is unrelated infra flake TFTR! bors r+ |
4 participants
status: expose additional Go runtime metrics
This PR adds three new metrics from Go's runtime/metrics package:
These provide better visibility into GC behavior and object allocation
patterns, useful for diagnosing memory pressure and GC overhead.
Inspired by poking at #160195
Epic: none