Policies

Policies control how long a job runs and when it stops automatically.

{
  "policies": {
    "maxDurationSec": 7200,
    "idleTimeoutSec": 60
  }
}

Field	Type	Default	Description
maxDurationSec	int	0 (unlimited)	Maximum job duration in seconds. 0 disables the cap. When reached, the job completes. Clamped to max 86400 (24h).
idleTimeoutSec	int	60	Seconds without media activity before the job ends. Activity comes from components_metrics (videoMillis / audioMillis / videoFrames / audioFrames). Set 0 to disable. Clamped to max 3600.

Idle detection

Idle is evaluated every ~10s with this priority:

1. Source idle — if the job has sources and none have reported activity within idleTimeoutSec, the job ends with event job.idle_timeout and reason source (Prometheus label source_idle).
2. Sink idle — otherwise, if the job has sinks and none have reported activity within the threshold, it ends with the same event and reason sink (Prometheus label sink_idle).
3. Jobs with no sinks only check sources.

If a role never saw activity, the grace baseline is the job’s updatedAt / createdAt (not wall-clock zero). Job status is completed, same as other policy stops.

• RTC sources (livekit, jitsi, daily, agora): idle when room media matching your select filters stops producing metrics.
• Pull sources (rtmp_pull, hls_pull, etc.): idle when the stream stalls or disconnects (no further metrics activity).

Idle periods are not billed for media processing. See Billing overview.

Stopping jobs

Jobs also stop when you call DELETE /v1/jobs/:name or when the platform container reaches its maximum lifetime.