perf-sentinel chart v0.9.21

What's new in chart-v0.9.21

A single addition, and it exists because two defaults disagree with each other. The daemon ships with [daemon.ack] enabled = true, so runtime acknowledgments are on out of the box, while the chart's default topology cannot serve them: without a mounted volume the daemon resolves its ack store under a home directory the FROM scratch image does not provide, logs a warning at startup and answers 503 on the three ack routes. Until now nothing surfaced that at install time, and operators met it hours later as an unexplained 503. A post-install note now says it plainly, names the two values that fix it, and states the cluster prerequisite. appVersion is intentionally held at the current daemon release 0.9.17 because the binary did not change. The chart version advances to 0.9.21 on its own.

Added

• A post-install note whenever the install cannot persist runtime acknowledgments. It fires on the absence of a mounted volume rather than on workload.kind, so a StatefulSet left with persistence.enabled: false gets it too, which is correct since that combination mounts nothing. The note separates what breaks from what does not: POST and DELETE /api/findings/{sig}/ack plus GET /api/acks answer 503, while ingestion, detection, /metrics and the rest of the query API are untouched, and the committed CI acknowledgment baseline is still honoured because the daemon reads it independently of the JSONL store. It then points at workload.kind=StatefulSet with workload.statefulset.persistence.enabled=true, and warns that this mode needs a usable StorageClass.

Behavior

• No rendered manifest changes. Only the message printed after helm install or helm upgrade differs, so upgrading from 0.9.20 rolls nothing and requires no migration.
• Deployment remains the default, deliberately. Making the persistent mode the default would mean two costs paid by everyone: the install would depend on a usable StorageClass, leaving the PVC and therefore the pod Pending on clusters without a default one, and workload.kind cannot be changed by helm upgrade, since Helm does not convert a Deployment into a StatefulSet, so every existing release would have to be deleted and recreated. The chart keeps an install path that works anywhere and tells you, at install time, what that default costs you.
• Nothing changes for installs that already run StatefulSet with persistence. They persisted acknowledgments before this release and still do, and they see no note.

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 0.9.21

Upgrade an existing release:

helm upgrade perf-sentinel oci://ghcr.io/robintra/charts/perf-sentinel --version 0.9.21

Full Changelog: chart-v0.9.20...chart-v0.9.21