robintra/perf-sentinel chart-v0.9.5

perf-sentinel chart v0.9.5

on GitHub

What's new in chart-v0.9.5

This chart release ships a template change alongside advancing appVersion to the current daemon release 0.9.5. The opt-in PrometheusRule gains a dedicated memory-pressure alert to match the daemon's new ingest memory guard, and the existing ingest-rejection alert is re-scoped so the two do not overlap. The chart version and appVersion stay in lockstep at 0.9.5.

Added

• PerfSentinelMemoryPressureRejecting alert in the opt-in PrometheusRule (prometheusRule.enabled). It fires on perf_sentinel_ingest_memory_pressure == 1 held for 5m, the state gauge the 0.9.5 daemon exposes while its opt-in memory guard ([daemon] memory_high_water_pct) is rejecting OTLP ingest to keep RSS under the cgroup limit. Being state-based, it keeps firing while ingest is paused even after exporters stop retrying, and its remediation points at resources.limits.memory or adding replicas, not the ingest queue.

Changed

• PerfSentinelIngestRejecting re-scoped to reason!="memory_pressure". It still covers channel_full, parse_error, and unsupported_media_type, so the queue-saturation and bad-exporter cases keep alerting, while memory-pressure rejections are owned by the new dedicated alert above. Each rejection reason keeps pointing at the correct fix.
• appVersion advanced from 0.9.4 to 0.9.5, and the artifacthub.io/images tag follows to ghcr.io/robintra/perf-sentinel:0.9.5. The 0.9.5 daemon adds OTLP JSON auto-detection to batch analyze --input, a mysql-stat subcommand with a matching mysql_stat tab in the self-contained HTML dashboard, and the opt-in cgroup memory-pressure ingest guard. See the v0.9.5 release.

Behavior

• The PrometheusRule change only affects clusters that opt in with prometheusRule.enabled=true. With it off (the default), the rendered manifest set is unchanged apart from the image tag. With it on, a memory-guard rejection now raises a dedicated state-based alert instead of being folded into the generic ingest-rejection alert.
• The appVersion bump rolls the pod to the 0.9.5 image. An upgrade that keeps the default empty image.tag resolves to ghcr.io/robintra/perf-sentinel:0.9.5 and reschedules the pod. No /api/* route, OTLP wire shape, or configuration key is removed, and the new [daemon] memory_high_water_pct knob defaults to off, so the upgrade is behavior-preserving unless you opt in.

Install

The chart is published as an OCI artifact on GHCR, install it directly with no helm repo add step:

helm install perf-sentinel oci://ghcr.io/robintra/charts/perf-sentinel --version 0.9.5

Upgrade an existing release:

helm upgrade perf-sentinel oci://ghcr.io/robintra/charts/perf-sentinel --version 0.9.5

Full Changelog: chart-v0.9.4...chart-v0.9.5