Bump greenfield ingestor baseline v0.3.2 → v0.3.6 (customer stuck on 0.3.2)

[saadqbal]

Context

A customer (Henrik / Opentarget) reports the data ingestor still behaving as v0.3.2 — specifically the "empty cells flagged as non-numeric" error, which is tracebloc/data-ingestors#150, fixed in v0.3.3. He keeps landing back on 0.3.2 after fresh namespace installs and helm upgrade.

Root cause

The chart's greenfield ingestor baseline is frozen at v0.3.2:

• client/values.yaml → images.ingestor.digest: sha256:d361fa77… — confirmed on ghcr to be exactly ingestor:0.3.2.
• Pinned 2026-06-03 (Pin greenfield ingestor baseline to v0.3.2 in client chart #184) and never bumped through 0.3.3 / 0.3.4 / 0.3.5 / 0.3.6 (RELEASING.md step 7 made optional in data-ingestors#144).

The floating 0.3 tag (now → 0.3.6) auto-refreshes running deployments, but the baseline wins in the customer's exact situations:

• Fresh installs land on 0.3.2 and run it for 1–2 image-refresh ticks (~15–30 min) before converging (CronJob "don't churn on install" first-tick contract).
• helm upgrade re-asserts the 0.3.2 baseline onto the live env.
• If image-refresh is also silently failing on his cluster (open: image-refresh: surface persistent ingestor (ghcr) digest-resolve failures instead of silently skipping #190), he is permanently stuck on 0.3.2.

Fix (this ticket)

Bump the greenfield baseline to v0.3.6 so fresh installs and helm upgrade land on the fixed image:

• client/values.yaml: images.ingestor.digest → sha256:dfdaa7e633a8df0f46b403e9422eba5c16bdb7d9c39047e6d3738f9e38fbdba8 (ghcr ingestor:0.3.6, multi-arch amd64+arm64 — passes the ingestor-multiarch guard).
• client/Chart.yaml: version/appVersion 1.5.0 → 1.5.1.
• Precedent: chore: pin client chart's ingestor digest to v0.3.1 #161 (v0.3.0→v0.3.1), chore(#184): pin greenfield ingestor baseline to v0.3.2 #185 (v0.3.1→v0.3.2).

Acceptance

• images.ingestor.digest bumped to the v0.3.6 multi-arch digest; comment updated with the cumulative 0.3.3→0.3.6 fixes.
• Chart.yaml version + appVersion bumped in lockstep.
• PR to develop, reviewed + merged (second approver per release gates).
• ingestor-multiarch CI guard green.
• Chart promoted/published so customers can helm upgrade onto it.

Out of scope / follow-up

• image-refresh: surface persistent ingestor (ghcr) digest-resolve failures instead of silently skipping #190 — image-refresh silently skipping when the ghcr digest can't resolve. Needs Henrik's cluster diagnostics (chart version, live INGESTOR_IMAGE_DIGEST, tracebloc.io/last-refreshed-ingestor-digest annotation, CronJob logs) to confirm whether on-prem ghcr reachability is also blocking auto-refresh.
• Immediate customer unblock (separate from this chart release): helm upgrade <release> tracebloc/client -n <ns> --reuse-values --set images.ingestor.digest=sha256:dfdaa7e6….

[saadqbal]

✅ Delivered — chart v1.5.1 with ingestor baseline v0.3.6

Full chain complete:

Step	Result
Baseline bump	#228 (digest 0.3.2→0.3.6, Chart 1.5.0→1.5.1) merged to develop
FR verification on dev	Ran the #150 regression in pods built from each digest on tb-client-dev-templates. v0.3.2 (d361fa77) flags INT/FLOAT missing cells as invalid (reproduces the customer's "non-numeric" error); v0.3.6 (dfdaa7e6) treats them as NULL = valid, and still rejects genuine non-numeric input.
FR gate	#228 → "Ready for prod"; gate / gate passed on #230
Promotion	#230 (develop→main, v1.5.1) approved + merged (eb4589c)
Publish	GitHub Release v1.5.1 → release-helm-chart succeeded; client-1.5.1.tgz + index.yaml live on gh-pages (public URL propagating via Pages CDN)

Customer upgrade

helm repo update
helm upgrade <release> tracebloc/client -n <namespace>

Fresh installs and helm upgrade now land on v0.3.6 instead of v0.3.2.

Still open (separate)

• image-refresh: surface persistent ingestor (ghcr) digest-resolve failures instead of silently skipping #190 — if the customer's on-prem cluster can't resolve the ghcr digest, image-refresh silently skips and the cluster won't auto-roll even with a current baseline. Needs his cluster diagnostics to confirm. The v1.5.1 baseline bump fixes the fresh-install / helm upgrade path regardless.