Handle PostgreSQL poll failures in run waitFor

[Mr-Neutr0n]

createPostgresRunStore().waitFor() starts a polling interval whose getRun(runId) rejection is not handled:

void getRun(runId).then((r) => {
  if (r && isTerminal(r.status)) finish(r);
});

If Postgres is temporarily unavailable while a caller is waiting, the query rejects outside the returned waitFor promise. On current Node versions that is an unhandled rejection and can terminate the process instead of letting waitFor time out or report the database error normally.

The interval also leaves its timeout scheduled after finish, although it is unref'd.

Could the poll catch transient query failures (or reject waitFor deliberately) and clear both timers on settlement?

[Rowerguy508]

Fixed in e59ac30.

Root cause: The polling in createPostgresRunStore.waitFor() used a fire-and-forget promise with no rejection handler:

void getRun(runId).then((r) => {
  if (r && isTerminal(r.status)) finish(r);
});

When Postgres was temporarily unavailable, getRun() rejected, and since there was no .catch(), this became an unhandled promise rejection. On Node >= 15, unhandled rejections in fire-and-forget promises terminate the process — so waitFor never got to time out gracefully.

Fix:

1. Added a rejection handler to .then() — transient DB failures just log silently and keep polling until the timeout fires (as suggested in the issue)
2. Added clearTimeout(timer) alongside the existing clearInterval(poll) in both finish() and the timeout handler, so neither timer leaks after settlement

[Mr-Neutr0n]

Thanks for working on this. I cannot find e59ac30 in the upstream repository or an open PR yet. Could you share the commit link or open a PR for review?

[time-attack]

This landed on main in #494 (c58bbae). waitFor catches transient getRun failures and clears both timers. Test in test/postgres-store.test.ts. Please close this issue.

[Mr-Neutr0n]

closing