perf(profiling): reduce CPU and off-CPU sampling overhead #87

Merged

binjovi-bot merged 1 commit from codex/pipeline-profile-rate-20260909 into trunk

2026-09-09 14:15:15 +00:00

sean commented

2026-09-09 14:13:59 +00:00

Owner

Add a configuration checksum so rate changes restart the collector. Keep the single-node limit, exact context binding and all existing resource and security limits.

Validation: repository checks, Kubernetes server dry-run and validation with the pinned Alloy binary passed. Repeat the fixed-revision warm Build measurement after deployment.

The collector-on/off/on warm Build check measured 8.91% added median time and 7.09% at p95 with 99 Hz CPU and 10% off-CPU sampling. Lower these to 49 Hz and 1% for the next check. Keep all observed syscall returns in the summary counters. Add a configuration checksum so rate changes restart the collector. Keep the single-node limit, exact context binding and all existing resource and security limits. Validation: repository checks, Kubernetes server dry-run and validation with the pinned Alloy binary passed. Repeat the fixed-revision warm Build measurement after deployment.

sean added 1 commit

2026-09-09 14:13:59 +00:00