feat: move remaining bundled images off bitnamilegacy (postgresql, redis, kubectl sidecar) #4861

Is your feature request related to a problem? Please describe.

Chart 4.5.15 still pulls four images from the bitnamilegacy/* namespace. Broadcom froze that namespace when Bitnami moved to a paid catalogue — those tags no longer receive security patches, so the Debian base layers accumulate CVEs indefinitely with no upgrade path.

From values.yaml in chart 4.5.15:

The first four come from Bitnami subchart dependencies (postgresql 16.7.14, redis 21.2.6, minio 17.0.9 via the s3 alias). The fifth is a first-party chart value.

bitnamilegacy/kubectl:1.28 is worth calling out separately. kubectl supports roughly ±1 minor version of skew against the API server. Pinned at 1.28, the token-syncer sidecar is already outside that window on any cluster running 1.30+, and increasingly so over time. It is also the only one of the five that is not tied to a subchart, so it is a one-line values change.

This affects every self-hosted deployment that uses the bundled datastores, which is the chart's default (postgres.deploy, redis.deploy and s3.deploy are all true).

Describe the solution you'd like to see

The same treatment ClickHouse received in 4.5.15 — first-party templates using the upstream official image, with digest support for reproducible pinning:

clickhouse:
  image:
    registry: docker.io
    repository: clickhouse/clickhouse-server
    tag: "26.2"
    digest: ""

That change was handled well: persistence.existingClaim auto-detects the PVC the Bitnami subchart left behind, a data-paths.xml ConfigMap points the server at the directory the old layout used, and a chown init container relabels the volume for the new uid — including on storage classes where fsGroup cannot relabel. Migrating from the bundled Bitnami ClickHouse was genuinely seamless as a result.

Applying the same pattern to postgres and redis would remove the remaining exposure. Upstream official images exist for both (postgres, redis, or valkey), and for the sidecar registry.k8s.io/kubectl is the upstream-maintained option.

Order of value, in our view:

1. webapp.tokenSyncer.image — no subchart, no state, no migration. A values change.
2. redis — stateful but largely reconstructible.
3. postgresql — highest value and highest care required, since it holds the secret store and run state.

Describe alternate solutions

Override the subchart images in values. Not viable in practice: the Bitnami charts assume Bitnami's filesystem layout, entrypoints and uid conventions, so substituting an official image breaks the probes and init scripts. We looked at this and discounted it.

Point the chart at external services (deploy: false). This works and is what we have done for object storage — s3.deploy: false with s3.external.* removed both MinIO images cleanly. But it is not a general answer: it pushes every self-hoster toward managed Postgres and Redis, which defeats much of the point of a self-hosted chart, and the bundled path stays unpatched for everyone who does not.

Stay on bitnamilegacy and accept the risk. Defensible only while these services are cluster-internal with no ingress, which is the common case. It still shows up in image scanning and in security questionnaires, and it degrades over time as CVEs accumulate against frozen base layers.

Additional information

Related prior work:

• fix(hosting): deploy ClickHouse from the official image instead of Bitnami #4249 — deploy ClickHouse from the official image instead of Bitnami (the precedent this asks to extend)
• v4.5.0 ClickHouse writers require CH ≥25.8, but the Helm chart still bundles ClickHouse 25.7.5 → every insert fails with UNKNOWN_SETTING #4197 — ClickHouse writers require CH ≥25.8 while the chart bundled ClickHouse 25

Verified against chart 4.5.15 (oci://ghcr.io/triggerdotdev/charts/trigger) and against a running self-hosted deployment on k3s, where kubectl get pods -o jsonpath over the namespace confirms the five images above are what actually run.

Happy to test a pre-release chart against a live single-node self-hosted install if that is useful — we have just completed the 4.4.6 → 4.5.15 upgrade including the ClickHouse volume adoption, so the migration path is fresh.