Reported by the founder 2026-09-01 ~23:20Z on https://dashboard.vexa.ai/meetings/27493 right after the exact-row transcript fix went live (#1386): "now it renders zoom, but no audio playback".
Measured on prod (meeting-api DB, read-only):
- 27493 (zoom
92446951537, user 23, calendar auto-join, 17:58–18:43Z, 45 min, 271 segments, completion_reason: stopped, service_provenance.bot_outcome: served, recording_enabled: true) has no data.recordings key and no chunks. The by-id and native transcript routes both return recordings: [] — the dashboard is showing the truth, not a fetch bug.
- 26491 (same zoom id, same user, calendar auto-join, 17:57Z, 1 min, 0 segments) — also no recording.
- Last 7 days: 285 completed meetings, 254 with a recording; of the 31 without, 26 had recording disabled by the caller. 27493 is the only recording-enabled, served, >5-min meeting in 7 days with no recording.
- The same user's non-calendar Zooms on 2026-08-31 (27334, 27306) recorded; calendar-joined Meet (27301, 27341) recorded. The miss is specific to calendar-joined Zoom on this link (
zoom.us/j/…?pwd=…, the LFX weekly).
- Runtime: workload
mtg-27493-db7a3285 spawned 17:58:33Z, stoppedAt 2026-09-01T23:21:10Z, exitCode 0, stopReason completed — recorded as stopped ~4.6 h after the meeting ended. Whether the pod lingered or the runtime observed late is not established.
- No bot log survives (pod gone); meeting-api access logs carry no upload lines at INFO, so the chunk-upload path (
POST /internal/recordings/upload via recordingUploadUrl) could not be traced. recording.ts logs recording: no recordingUploadUrl — chunk … NOT uploaded when the invocation lacks the URL — worth checking first on the calendar-spawn path for Zoom.
Not caused by the dashboard change in #1386/#1387 (client fetch only). Possibly related: #1204 (recording tap misses late tracks), #934 (unbounded teardown).
Reproduction opportunity: the next occurrence of this recurring meeting (row 27499, scheduled) — capture the bot log before the pod is reaped.
Reported by the founder 2026-09-01 ~23:20Z on https://dashboard.vexa.ai/meetings/27493 right after the exact-row transcript fix went live (#1386): "now it renders zoom, but no audio playback".
Measured on prod (meeting-api DB, read-only):
92446951537, user 23, calendar auto-join, 17:58–18:43Z, 45 min, 271 segments,completion_reason: stopped,service_provenance.bot_outcome: served,recording_enabled: true) has nodata.recordingskey and no chunks. The by-id and native transcript routes both returnrecordings: []— the dashboard is showing the truth, not a fetch bug.zoom.us/j/…?pwd=…, the LFX weekly).mtg-27493-db7a3285spawned 17:58:33Z,stoppedAt 2026-09-01T23:21:10Z, exitCode 0, stopReason completed— recorded as stopped ~4.6 h after the meeting ended. Whether the pod lingered or the runtime observed late is not established.POST /internal/recordings/uploadviarecordingUploadUrl) could not be traced.recording.tslogsrecording: no recordingUploadUrl — chunk … NOT uploadedwhen the invocation lacks the URL — worth checking first on the calendar-spawn path for Zoom.Not caused by the dashboard change in #1386/#1387 (client fetch only). Possibly related: #1204 (recording tap misses late tracks), #934 (unbounded teardown).
Reproduction opportunity: the next occurrence of this recurring meeting (row 27499, scheduled) — capture the bot log before the pod is reaped.