huskies: merge 498_bug_stale_merge_job_lock_prevents_new_merges_after_agent_dies

2026-04-10 18:51:07 +00:00
parent 6f7a0c7708
commit bc2b1e244c
2 changed files with 74 additions and 0 deletions
@@ -212,6 +212,70 @@ mod tests {
            .unwrap();
    }

+    // ── bug 498: stale Running job blocks retry ───────────────────────────────
+
+    /// Regression test for bug 498: a Running merge job left behind by a killed
+    /// mergemaster must not block the next call to start_merge_agent_work.
+    ///
+    /// Before the fix: start_merge_agent_work would return "Merge already in
+    /// progress" when a Running entry existed, even after the mergemaster died.
+    /// After the fix: the entry is cleared when the mergemaster exits, so a new
+    /// call succeeds.
+    #[tokio::test]
+    async fn stale_running_merge_job_is_cleared_and_retry_succeeds() {
+        use tempfile::tempdir;
+
+        let tmp = tempdir().unwrap();
+        let repo = tmp.path();
+        init_git_repo(repo);
+
+        let pool = Arc::new(AgentPool::new_test(3001));
+
+        // Inject a stale Running entry, simulating a mergemaster that died
+        // before the merge pipeline completed.
+        {
+            let mut jobs = pool.merge_jobs.lock().unwrap();
+            jobs.insert(
+                "77_story_stale".to_string(),
+                MergeJob {
+                    story_id: "77_story_stale".to_string(),
+                    status: MergeJobStatus::Running,
+                },
+            );
+        }
+
+        // With a stale Running entry, start_merge_agent_work must be blocked.
+        let blocked = pool.start_merge_agent_work(repo, "77_story_stale");
+        assert!(
+            blocked.is_err(),
+            "start_merge_agent_work must be blocked while Running job exists"
+        );
+        let err_msg = blocked.unwrap_err();
+        assert!(
+            err_msg.contains("already in progress"),
+            "unexpected error: {err_msg}"
+        );
+
+        // Simulate the mergemaster exit path: clear the stale Running entry.
+        {
+            let mut jobs = pool.merge_jobs.lock().unwrap();
+            if let Some(job) = jobs.get("77_story_stale")
+                && matches!(job.status, MergeJobStatus::Running)
+            {
+                jobs.remove("77_story_stale");
+            }
+        }
+
+        // After clearing, start_merge_agent_work must succeed (it will fail
+        // the pipeline because there's no feature branch, but it must not be
+        // blocked by "Merge already in progress").
+        let result = pool.start_merge_agent_work(repo, "77_story_stale");
+        assert!(
+            result.is_ok(),
+            "start_merge_agent_work must succeed after stale Running job is cleared; got: {result:?}"
+        );
+    }
+
    // ── merge_agent_work tests ────────────────────────────────────────────────

    /// Helper: start a merge and poll until terminal state.