[staging-livepeer-api] 'api_token id=undefined not found' — lastSeen update fails hourly for token 2f25c979 (18 occurrences)

[livepeer-tessa]

Summary

staging-livepeer-api is logging an error every hour when trying to update the lastSeen timestamp for a specific API token. The error indicates that the token ID is being passed as undefined to the DB update query, causing the api_token record lookup to fail.

Error Details

Service: staging-livepeer-api
Time window: 2026-03-29 18:09 UTC – 2026-03-30 06:09 UTC
Total occurrences: 18 (roughly every 30-60 mins, clustered around the :31 minute mark of each hour)
Token ID: 2f25c979-904f-4fce-b329-4174b54fdcbf

Error message

error saving last seen: table=api_token id=2f25c979-904f-4fce-b329-4174b54fdcbf err= Ge [Error]: api_token id=undefined not found

Surrounding context (from Loki)

The same token 2f25c979-904f-4fce-b329-4174b54fdcbf is being successfully used for DB SELECTs and UPDATEs in other queries (e.g. UPDATE api_token SET data = data || ... WHERE id = '2f25c979...' rows=1), but some code path is calling saveLastSeen() with id=undefined instead of the resolved ID. The error message format id=<real-uuid> err=...: api_token id=undefined not found suggests the UUID is logged correctly at the call site, but the DB query receives undefined.

Occurrence timestamps (UTC)

2026-03-29 19:31:42 UTC
2026-03-29 20:31:39 UTC
2026-03-29 21:31:37 UTC
2026-03-29 21:31:48 UTC
2026-03-29 22:31:37 UTC
2026-03-29 22:31:42 UTC
2026-03-29 22:31:48 UTC
2026-03-29 22:31:52 UTC
2026-03-29 23:00:00 UTC (approx)
2026-03-30 00:31:37 UTC
2026-03-30 00:31:45 UTC
2026-03-30 00:31:52 UTC
2026-03-30 01:31:45 UTC
2026-03-30 02:31:48 UTC
2026-03-30 03:31:48 UTC
2026-03-30 04:31:45 UTC
2026-03-30 04:31:51 UTC
2026-03-30 05:31:39 UTC
2026-03-30 05:31:48 UTC

Likely Cause

The saveLastSeen function (or caller) in staging-livepeer-api is receiving the API token object but dereferencing the ID field incorrectly — e.g. token.id returns undefined when the field is named differently (e.g. token.userId, token.tokenId), or a race condition causes the token object to be undefined when the async callback fires.

The real UUID is logged in the error context (id=2f25c979-904f-4fce-b329-4174b54fdcbf), suggesting it's available at the log-writing level but not passed correctly into the DB query.

Impact

• Staging only (no indication in prod Loki)
• lastSeen timestamps for this API token are not being updated
• Token remains functional for auth (SELECT queries succeed)
• ~18 spurious error log entries per 12h window

Suggested Fix

Search for the saveLastSeen function in staging-livepeer-api and check:

1. The field name used to extract the token ID (might be id vs. tokenId vs. another field)
2. Whether the token object could be undefined or re-shaped before being passed to saveLastSeen

Loki Query

{app="staging-livepeer-api"} |~ "api_token id=undefined"

---

Detected automatically by scope-monitor cron at 2026-03-30 06:09 UTC

[livepeer-tessa]

Root cause identified. The bug is in livepeer/studio, not in this repo — filed a PR there.

What's happening

In packages/api/src/middleware/tracking.ts, the flushLastSeen() method calls table.update() with an array of SQL conditions (including a guard coalesce((data->'lastSeen')::bigint, 0) < ${lastSeen}). When that UPDATE returns 0 rows (because lastSeen is already up-to-date — benign race condition), table.ts throws a NotFoundError with the message constructed from doc.id. But doc is the partial update payload { lastSeen: <timestamp> } — it has no id field — so the error reads api_token id=undefined not found.

The real UUID is visible in the log prefix, but the thrown error always says id=undefined.

Fix

Primary: Pass { throwIfEmpty: false } for lastSeen updates — they're best-effort; a no-op UPDATE is not an error.
Secondary: Fix the error message in table.ts to not blindly use doc.id when query is an array.

PR

livepeer/studio#2349