René's URL Explorer Experiment

Title: perf(ci): remove container from go unit test jobs by davdhacs · Pull Request #20484 · stackrox/stackrox · GitHub

Open Graph Title: perf(ci): remove container from go unit test jobs by davdhacs · Pull Request #20484 · stackrox/stackrox

X Title: perf(ci): remove container from go unit test jobs by davdhacs · Pull Request #20484 · stackrox/stackrox

Description: Description Three changes to the Go unit test CI: 1. Remove apollo-ci container from go, go-postgres, and go-bench jobs Run directly on GHA runner instead of inside the apollo-ci container. Eliminates ~55s container init overhead per job. 2. Add toolchain go1.25.10 directive to go.mod Go 1.25.10 includes the backported fix for a covdata bug (golang/go#75031) that causes go test -cover to fail when Go is auto-downloaded to the read-only module cache. Without the container, the system Go (1.22) auto-downloads the toolchain version via GOTOOLCHAIN=auto. The toolchain directive (not the go directive) is used because Konflux builders have Go 1.25.9 with GOTOOLCHAIN=local, which ignores the toolchain directive -- Konflux builds don't use go test -cover so the bug doesn't apply there. 3. Add pre-installed postgres as an alternative go-postgres test target Adds pg: ['default', '15'] matrix to go-postgres. The default variant uses the runner's pre-installed postgres (currently v16) via systemctl. The 15 variant runs the sclorg postgres 15 container (same base image as production). The image digest is pinned in .github/test-images/postgres.dockerfile for Dependabot auto-bumping. default is excluded on PRs (saves ~33m from the total per-commit compute). Master/nightlies run both for coverage (I think after some weeks, this will prove we can switch to default for PRs also). Override with ci-postgres-tests label to run both on PRs. Note: Share GOCACHE across pg versions Removed matrix.pg from the GOCACHE key suffix since Go compilation is identical regardless of postgres version. Both default and 15 jobs now share the same cache entry. Verified: seeding run saved 955 MB cache, re-run restored it for both default (1m50s) and 15 (2m11s) -- down from 33-36m cold. Testing and quality CI results are inspected How I validated my change All GHA Go jobs pass: go, go-postgres (default + 15), go-bench All Konflux Go builds pass with go 1.25.7 + toolchain go1.25.10 go-version-ceiling passes Covdata fix verified: reproduced locally on ubuntu:24.04, confirmed toolchain go1.25.10 resolves it Shared GOCACHE verified: both default and 15 restore from same 955MB cache entry (re-run attempt 2 of eefd4df) Root cause investigation documented in experiment PRs #20456, #20468, #20469, #20511

Open Graph Description: Description Three changes to the Go unit test CI: 1. Remove apollo-ci container from go, go-postgres, and go-bench jobs Run directly on GHA runner instead of inside the apollo-ci container. Elimina...

X Description: Description Three changes to the Go unit test CI: 1. Remove apollo-ci container from go, go-postgres, and go-bench jobs Run directly on GHA runner instead of inside the apollo-ci container. Elimina...

Opengraph URL: https://github.com/stackrox/stackrox/pull/20484

X: @github

direct link

Domain: github.com

Links:

Viewport: width=device-width