Skip to content

feat(LinuxTentacleE2E): Phase 12.L.E.11 — E11.u2-Linux stale lockfile recovery

Placeholder ppxd requested to merge phase12.L.E.11-e11u2-linux-stale-lockfile into main

Summary

Pins the kernel-flock-auto-release contract on Linux: a leftover lockfile (from a tentacle process killed mid-upgrade) MUST NOT block the next dispatch, because the kernel automatically releases flocks when the holding process dies.

Why this is a regression test, not a feature test

Linux already does the right thing here — kernel BSD flock has built-in stale-lock recovery via process-death auto-release. The test isn't validating new functionality; it's pinning the contract so a future polish that "improved" lock detection (e.g. by mirroring Windows' PID-based file-content check) doesn't silently regress kernel-flock semantics.

Without this pin: a regression where someone added file-existence-based blocking would leave production agents stuck with stale lockfiles requiring manual cleanup.

Platform Mechanism Stale-detection
Windows .ps1 reads PID from lockfile, checks process liveness Explicit
Linux Kernel BSD flock Automatic via kernel auto-release

Test mechanism

Pre-stage a lockfile with sentinel "99999\n" content (no live holder), then run a normal upgrade. The kernel sees no active holder → grants the lock immediately → upgrade proceeds end-to-end exactly like E1.h-Linux.

Assertions

  • exitCode == 0 (SUCCESS path, NOT no-op path)
  • stdout does NOT contain "upgrade is already in progress on this host" (regression marker — would surface if .sh's idempotency guard misfired)
  • last-upgrade.json.Status == "SUCCESS"
  • Service marker swapped to "2.0.0-test" (Phase B fully completed)

Pair with E11.u1-Linux (PR #220)

Test Lockfile Holder Expected
E11.u1-Linux exists active flock holder exit 0 no-op (real concurrent dispatch)
E11.u2-Linux exists none exit 0 SUCCESS (stale lockfile recovery)

Together these pin both halves of the lock contract: held flock blocks, unattended file does not.

Fidelity tier

🟢 High (Rule 12.4): real production .sh + real systemd + real bash + real LocalReleaseMirror + real Phase A+B end-to-end.

Test plan

  • Linux E2E workflow runs Squid.LinuxTentacleE2ETests (manual workflow_dispatch after merge)
  • E11u2_StaleLockfile_NoLiveHolder_DispatchProceedsNormally_Linux passes within ~5s
  • No regression on existing 12 Linux E2E tests

🤖 Generated with Claude Code

Merge request reports