Bridge and connector troubleshooting

Français · Troubleshooting · Published, reviewed by a person · Applies from: 4.1.20

What a Reality Bridge's error codes mean, and what to do about them. The Bridge is the server that carries signals from the real world (an email, a badge, a terminal) to a player's game; its connectors send those signals. This page covers the Bridge's own answers, not the hosted Studio's, which has codes of the same numbers.

The Bridge answers 409

A Bridge's 409 means "this was already done", never "your copy is stale":

• {"error": "pairing", "message": "already confirmed"} on POST /v1/pairings/<code>/confirm: the pairing code was confirmed already, by this connector or another one. Two connectors confirming one code at once: one wins, the other gets this 409. The player is paired; nothing to redo.
• {"error": "duplicate", …"this run was submitted already (the first submitter keeps it)"} on POST /v1/runs: the same speedrun run was sent twice, and the second is refused as it would join the queue. The first submission keeps its place in the queue; it is ranked only once the speedrun worker finds it valid.
• {"error": "trust", …"only a run the worker found valid can be moderated"}: a moderator tried to verify a run the speedrun worker did not find valid.

None of them loses data: the first action stands. To check it, read the Bridge's log (JSON lines on its standard output): "event":"pairing.confirmed" with the playerId for the pairing, "event":"run.submitted" (then run.verified) for the run. The Bridge has no operator route that reads a pairing's state. Never call GET /v1/pairings/<code> to check one: that is the player's one-time claim, and calling it hands out, and uses up, the player's capability.

The Bridge answers 503

• {"error": "busy", "message": "the store is busy: try again in a second"}, with Retry-After: 1: another process held the store's write lock longer than the request could wait (SQLite). Nothing was written; send it again.
• GET /readyz answers 503 {"status": "unavailable"}: the store does not answer. Look at the database first.
• From 4.1.20, a Bridge whose maintenance lease is lost (not renewed, or the database connection that holds its lock dropped) holds its writes, which answer 503, and fails /readyz until it takes a lease again (its log line maintenance.lease.lost says why).

The same event arrived twice

A connector may send the same signal twice (a retry after a timeout, a restart). The Bridge accepts it once: POST /v1/signals answers 202 the first time, and 200 with duplicate for a dedupeKey it already accepted. The player's save keeps the sequence of signals it applied, so a signal is never applied twice in the game.

A connector is down

• The game plays on: a player offline, or without a signal, keeps playing; what arrives later is applied when it arrives.
• A connector started with a health port answers GET /health with 200 when it is well, 503 otherwise, with a detail; GET /metrics on the same port gives its counters.
• An email connector leaves a message unseen when the Bridge refused it for a reason that may pass (the Bridge away or busy, a quota): it is read again at the next poll. A message refused for good is marked seen and flagged, for you to look at. Over a quota the Bridge answers 429: the connector sees it, nothing is dropped in silence.

Edit the source of this page · Report an error