You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Merging one release PR left the other two CONFLICTING. All three edit the
shared .release-please-manifest.json, and by default Release Please
rewrites a release branch only when the generated release notes change —
so a straggler kept a stale base, and neither the post-merge run nor an
explicit workflow_dispatch repaired it. always-update is the supported
option for exactly this ("can be useful if pull requests must not be
out-of-date with the base branch"); it costs extra API calls per run,
which is nothing at three charts. Needs release-please >= 16.15.0 and the
action pins ^17.6.1.
Stagger the integration matrix. Five jobs starting together pull the same
large image from five clusters at once, out of the shared IP range
GitHub-hosted runners use, which registries rate limit against. This is
the suspected cause of the 2026-08-13 stalls; still unproven, and the
diagnostics added alongside it will say either way.
Let Release Please author its PRs as a GitHub App. A GITHUB_TOKEN PR
cannot start workflow runs, so release PR checks wait on manual approval
and Integration Tests — push-triggered — never covers a release PR at
all. The step is skipped when the variable is unset and the action falls
back to GITHUB_TOKEN, so this changes nothing until an App is configured.
0 commit comments