Skip to content

Use OBI v0.10.0 and align component versions#135

Merged
paweljw merged 1 commit into
mainfrom
pawel/use-obi-0.10.0
Jul 10, 2026
Merged

Use OBI v0.10.0 and align component versions#135
paweljw merged 1 commit into
mainfrom
pawel/use-obi-0.10.0

Conversation

@paweljw

@paweljw paweljw commented Jul 1, 2026

Copy link
Copy Markdown
Member

Bumps OBI eBPF instrumentation from v0.7.1 to v0.10.0, then fixes version metadata that had drifted from the Dockerfiles and adds CI to keep it honest.

OBI

  • ebpf/Dockerfile: OBI v0.7.1 → v0.10.0 (both arch tarballs verified to resolve)
  • collector/Dockerfile: collector version → 1.1.28

Cluster-agent drift (real fix)

The collector pulled the stock coroot-cluster-agent:1.2.4 while the ebpf image builds v1.6.1 from source with the statement_timeout DSN patch. In cluster-collector mode the collector's agent does database monitoring, so the stock image could reintroduce the PgBouncer statement_timeout rejection the patch fixes.

  • collector/Dockerfile now builds the cluster-agent from source with the same patch as ebpf (v1.6.1); the patch stays a single source of truth in ebpf/patches/cluster-agent/.

Debian bases

All Debian bases are now pinned to 13.5 (trixie):

  • obi-source was leftover on bookworm 12.11-slim (only a tar-extraction stage, but inconsistent) → 13.5-slim
  • the builder stages floated on the trixie tag while the final was pinned 13.5-slim. node-agent is CGO_ENABLED=1 (glibc-linked), so a floating builder could out-run the pinned final's glibc. Builders (ebpf ×2 + the new collector one) → pinned debian:13.5.

Metadata true-up

  • collector/Dockerfile: ENV OBI_VERSION 0.4.1 → 0.10.0, CLUSTER_AGENT_VERSION 1.2.4 → 1.6.1
  • CLAUDE.md: Node Agent 1.27.0 → 1.30.0, Debian 12.11 → 13.5, Cluster Agent 1.2.4 → 1.6.1; dropped the stale Debian version from the ebpf comment; added a keep-in-sync note

CI

  • .github/workflows/version-sync.yml fails when the Dockerfile pins, the collector ENV *_VERSION block, and the CLAUDE.md summary disagree — Vector, OBI, Node Agent, Cluster Agent, and a single pinned Debian version across all stages. Verified green on the current tree and red on injected drift (including a reintroduced floating trixie).

@paweljw
paweljw force-pushed the pawel/use-obi-0.10.0 branch 3 times, most recently from cea4d10 to b27ab1b Compare July 1, 2026 08:49
@paweljw paweljw changed the title Use OBI v0.10.0 Use OBI v0.10.0 and align component versions Jul 1, 2026
- ebpf: bump OBI v0.7.1 -> v0.10.0; collector version -> 1.1.28
- collector: build the cluster-agent from source with the same statement_timeout patch as ebpf (v1.6.1) instead of pulling the stock 1.2.4 image. The stock image lacks the patch, so cluster-mode database monitoring could hit the PgBouncer statement_timeout rejection.
- pin all Debian bases to 13.5 (trixie): obi-source was leftover on bookworm 12.11, and the builders floated on the trixie tag while the final was pinned -- a CGO glibc-skew risk for node-agent.
- true up drifted version metadata (collector ENV OBI_VERSION, CLAUDE.md Node Agent/Debian/cluster-agent) against the Dockerfile pins.
- add a version-sync CI check that fails when the pins, the collector ENV block, and the CLAUDE.md summary disagree (component versions + Debian).
@paweljw
paweljw force-pushed the pawel/use-obi-0.10.0 branch from b27ab1b to 5a2d5c3 Compare July 1, 2026 09:14
@paweljw
paweljw merged commit ea81e24 into main Jul 10, 2026
9 checks passed
@paweljw
paweljw deleted the pawel/use-obi-0.10.0 branch July 10, 2026 11:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant