[rmcdaniel]

Part of #325.

Why

In the multi-node runs (#549), a paused SDK worker resumed holding workflow tasks the server had already given to the other worker. Native refused its late lease renewals and completion with only a reason (task_not_leased, conflicting_completion). The worker could not tell that its claim was lost, so it exited; this happened 6 times. Published PHP refuses the same requests through its ownership guard. Its body carries the task status and, once the run has closed, can_continue: false, and current workers read those fields to drop the claim and keep running.

What changes

• Workflow task completion, failure, lease renewal and task history now share one guard that follows PHP's WorkflowTaskOwnership::guard order:
  • Task no longer leased. Refused with run_closed (stop reason, run status, closure, task status, owner) when its run has closed. Otherwise refused with task_not_leased and the task status.
  • Task still leased. Checked for expiry (lease_expired), then owner (lease_owner_mismatch), then attempt (workflow_task_attempt_mismatch with current_attempt).
• Each refusal has PHP's body: error, the reason's fields, protocol_version and server_capabilities.
• A completion that differs from the recorded one now gets the guard's refusal instead of the native-only conflicting_completion. An identical replay still returns the recorded already_completed receipt.
• A completed workflow task now clears its lease expiry and keeps its owner, as PHP's does.

Evidence

• New fixture. late-workflow-completion re-registers a worker that holds a claim. It then sends the replaced claim's completion, renewal and failure three times: while the new claim holds the lease, after the new claim completes, and after the run closes. The contract checks PHP's exact bodies apart from timestamps.
• Local parity. In a local copy of the SQLite parity step, published PHP 2.5.14, embedded PHP and native all passed, and the comparison passed. The pre-change native binary failed the same fixture.
• Native tests. a_replaced_workflow_claim_is_refused_with_the_published_ownership_fields is new. Two existing tests asserted conflicting_completion after a run-closing completion. They now expect PHP's run_closed; after a failed run, the task status is failed, as PHP stores it.
• Local runs. The execution suite passed on SQLite, PostgreSQL and MySQL (283 tests each).
• Local full parity. On the same SQLite copy, native passed every default fixture except visibility-current-runs, which needs the published CLI artifact that CI supplies.
• Contract unit tests. tests/Unit/ServerParityLateWorkflowCompletionContractTest.mjs (19 tests) is added to the CI list.

Not covered

PHP also refuses an otherwise valid claim whose worker registration is no longer fresh (stale_worker_registration). The local activity and cancellation scope checkpoints still use the older fence. Both are recorded as known gaps and need their own fixtures.

🤖 Generated with Claude Code

[rmcdaniel]

Merged main (9c10bd0eb, with #499 and #510) as 74ac2675f; the merged tree passed the SQLite execution suite (287 tests). I cancelled the superseded Shared Server fixtures run 39471824937 for f43e6fae2. The run for 74ac2675f qualifies this PR.

[rmcdaniel]

Sonar's quality gate failed on new-code duplication (3.0%, limit 3): workflow_task_guard.rs repeated the impl Store where-clause that every store module carries. I'm moving the guard into store.rs beside the fence it replaces; the module keeps only the refusal-body builder. I cancelled the in-flight Shared Server fixtures run 39473539144, which the fix supersedes. The push follows local verification.

[rmcdaniel]

Merged main (49a56f5c0: #551 namespace update, #545 database TLS, #556 certifi) into this branch as 24285ccc1. The only conflict was in the port-log checkpoint list, and I kept both entries.

Parity. I cancelled the Shared Server fixtures run for the merged head (39507398656) as a duplicate:

• the branch's own change is unchanged since 39483535410, which passed all 13 jobs on 2fcefc4ee;
• main's additions run their own parity after each merge.

Local check of the merged head. On SQLite, the native lib suite (120) and the execution suite (288, including late_workflow_claims) pass.

Auto-merge is on for the required checks.

[rmcdaniel]

772cc5849 merges main after #543 (terminate and worker listing). Both branches add fixtures to the same registries, so I resolved additive conflicts in:

• contract.mjs and probe.php (fixture wiring);
• server-parity.yml (contract unit test lists);
• the port log (known gaps and checkpoints).

Local checks of the merged head:

• Native suites on SQLite. The lib suite (122) and the execution suite (292, including late_workflow_claims and the terminate tests) pass.
• Contract unit tests. The late completion, worker listing and worker replacement tests pass (50).
• SQLite parity replica. PHP 2.5.14, embedded PHP and native pass these fixtures, and the comparison passes:
  • late-workflow-completion;
  • worker-listing;
  • worker-replacement-activity-lease;
  • terminate-before-claim, terminate-pending-timer and terminate-leased-activity.

The Shared Server fixtures run for this head (39511283057) is left running. Auto-merge is back on for the required checks.

[rmcdaniel]

36e013e1e merges main again (987b07e10, #552 Workflow 748 core consumer), which changes no native code, parity script or parity workflow. I cancelled its duplicate parity run 39516312432; the merged-head run 39511283057 stays.

[rmcdaniel]

b306127ce merges main after #550 (search attribute registry), resolving the same additive registry conflicts. The resolution also restored a closing brace between the late-completion and search-attribute embedded blocks in probe.php. Locally on the merged head:

• the native lib suite (124) and SQLite execution suite (295) pass;
• the SQLite parity replica passes late-workflow-completion, search-attributes and worker-listing on all three targets, and the comparison passes.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud