Daemon liveness checks can restart a process that has stopped progressing
Vercel Daemons now support HTTP liveness probes that restart a container after consecutive failed checks, even if its process has not exited. The endpoint should cheaply report whether the work loop can progress. A liveness-triggered restart immediately kills the container rather than using graceful shutdown, so workers must tolerate interruption and safely retry unfinished work.
Vercel · Oct 30, 2026
Why it matters
Keeping a process alive and keeping a worker useful are separate operational checks.
More context
The October 30 update configures the endpoint through livenessProbe in the daemon definition. Daemons remain a public beta on all plans and consume provisioned memory while waiting. Test a deliberately stuck worker and its recovery behavior before deploying a probe.
Keep in mind
An endpoint disconnected from worker state can report success while work is stuck. Making health depend on downstream services can cause needless restarts; probe settings and billing still require workload-specific review.
Original read