Skip to content

chore(deploy): Type=exec on grob.container + close Tier 2 CI cluster (#15 #16 #17 #18 #19 #20)#256

Merged
Destynova2 merged 1 commit intomainfrom
chore/ci-ops-polish
Apr 22, 2026
Merged

chore(deploy): Type=exec on grob.container + close Tier 2 CI cluster (#15 #16 #17 #18 #19 #20)#256
Destynova2 merged 1 commit intomainfrom
chore/ci-ops-polish

Conversation

@Destynova2
Copy link
Copy Markdown
Contributor

Summary

Tier 2 CI/infra audit cluster close-out.

Only #20 needed a code change — items #15 through #19 had already landed in #225 (ci/phoenix-tier2-pipeline) before this branch was cut. This PR is scoped to the last remaining item plus a status note so the tracker can close.

#20 — deploy/grob.container Type=simple → Type=exec

Type=simple considers the unit started the instant fork() returns, so an immediate execve() failure (missing image, bad arg) presents as a restart loop rather than a clean start failure. Type=exec waits for execve() to succeed, surfacing exec-time failures as real unit failures and letting Restart=on-failure do the right thing.

The audit suggested Type=notify, but that is wrong for grob: the binary does not call sd_notify, so systemd would block forever on a readiness signal that never arrives. Readiness stays with the Quadlet HealthCmd. Rationale is inlined in the unit file as a # NOTE: so future readers do not retry notify.

Already-landed items (status only, no change in this PR)

# Target Status
#15 ci.yml retry-masking on hurl/k6 Landed in #225 — retry removed, # NOTE: no retry — flaky E2E must be fixed at the source
#16 ci-docs-shim.yml orphan Already deleted — file is absent from .github/workflows/
#17 Daily rustsec/audit-check cron Landed as audit-cron.yml with cron: '17 4 * * *'
#18 release-plz.yml timeouts Both release-pr and release-tag jobs already carry timeout-minutes: 10
#19 nightly.yml multi-OS + alert Matrix covers ubuntu/macos/windows × beta/nightly, dedicated alert job opens/updates a tracking issue on beta failure

Test plan

🤖 Generated with Claude Code

Type=simple treats the service as started the instant fork() returns, so
an immediate execve() failure (missing image, bad arg) surfaces as a
restart loop rather than a clean start failure. Type=exec waits for
execve() to succeed before considering the unit started.

Type=notify was suggested by the audit but is wrong for grob: the binary
does not call sd_notify, so systemd would block on a readiness signal
that never arrives. Readiness stays with the Quadlet HealthCmd.

Closes Tier 2 item #20. Items #15-#19 were already landed in
#225 (ci/phoenix-tier2-pipeline).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@Destynova2 Destynova2 enabled auto-merge (squash) April 22, 2026 07:31
@Destynova2 Destynova2 merged commit c6d796e into main Apr 22, 2026
28 checks passed
@Destynova2 Destynova2 deleted the chore/ci-ops-polish branch April 22, 2026 07:32
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