From a201e0898f4cc46c827aec1e41c015a960b87a4b Mon Sep 17 00:00:00 2001
From: SomSamantray <>
Date: Mon, 3 Aug 2026 14:38:33 +0530
Subject: [PATCH] fix: catch transient getRun failures in postgres waitFor poll

waitFor's setInterval poll called getRun(runId).then(...) with no
.catch(). A transient query failure (e.g. brief Postgres unavailability)
became an unhandled rejection outside the returned promise instead of
letting waitFor's own timeout report it normally. finish() also only
cleared the poll interval, leaving the timeout timer pending.

Fixes #57
---
 src/runs/postgres-run-store.ts | 11 ++++++++---
 test/postgres-store.test.ts    | 25 +++++++++++++++++++++++++
 2 files changed, 33 insertions(+), 3 deletions(-)

diff --git a/src/runs/postgres-run-store.ts b/src/runs/postgres-run-store.ts
index 4ab32edc..ae12c416 100644
--- a/src/runs/postgres-run-store.ts
+++ b/src/runs/postgres-run-store.ts
@@ -290,6 +290,7 @@ export function createPostgresRunStore(connectionString: string, opts?: { maxCla
           if (done) return;
           done = true;
           clearInterval(poll);
+          clearTimeout(timer);
           events.off(runId, onSettle);
           resolve(r);
         };
@@ -298,9 +299,13 @@ export function createPostgresRunStore(connectionString: string, opts?: { maxCla
         }
         events.once(runId, onSettle);
         const poll = setInterval(() => {
-          void getRun(runId).then((r) => {
-            if (r && isTerminal(r.status)) finish(r);
-          });
+          void getRun(runId)
+            .then((r) => {
+              if (r && isTerminal(r.status)) finish(r);
+            })
+            .catch((err: unknown) => {
+              console.error(`[postgres-run-store] waitFor poll for run ${runId} failed transiently:`, err);
+            });
         }, 250);
         poll.unref?.();
         const timer = setTimeout(() => {
diff --git a/test/postgres-store.test.ts b/test/postgres-store.test.ts
index d9e0b091..c591cc9f 100644
--- a/test/postgres-store.test.ts
+++ b/test/postgres-store.test.ts
@@ -893,6 +893,31 @@ test("pg run store: enqueue dedup, atomic one-per-session claim, fencing, ledger
   }
 });
 
+test(
+  "pg run store: waitFor survives a transient poll failure without an unhandled rejection",
+  { skip },
+  async () => {
+    const { runs, close } = createPostgresRunStore(URL!);
+    let unhandled: unknown;
+    const onUnhandledRejection = (err: unknown): void => {
+      unhandled = err;
+    };
+    process.once("unhandledRejection", onUnhandledRejection);
+    try {
+      const r = (await runs.enqueue({ sessionId: "sWaitFor", request: turn("x") })).run;
+      const pending = runs.waitFor(r.id, 800);
+      // Kill the pool mid-poll: the next getRun() tick will reject, exercising the
+      // catch path instead of leaking an unhandled rejection out of setInterval.
+      await new Promise((res) => setTimeout(res, 50));
+      await close();
+      await assert.rejects(pending, /did not finish within 800ms/, "still settles via its own timeout, not a crash");
+    } finally {
+      process.removeListener("unhandledRejection", onUnhandledRejection);
+    }
+    assert.equal(unhandled, undefined, "a transient poll failure must not escape as an unhandled rejection");
+  },
+);
+
 test(
   "pg run store: releaseLease (deploy drain) hands the run back as a retry without spending budget; stale token is a no-op",
   { skip },