That log line usually means the worker’s polling loop stopped making progress for long enough that Prefect marked it unhealthy. A few important details: - The message is tied to the worker healthcheck logic - For workers, Prefect tracks the last successful poll for scheduled work - If it hasn’t polled in roughly

PREFECT_WORKER_QUERY_SECONDS * 30

, it logs:

Worker has not polled in the last ... seconds and should be restarted

- With the default poll interval of

10s

, the threshold is about 300 seconds, so your

307 seconds

is right at that cutoff So this is usually a symptom, not the root cause. Common causes - Temporary or repeated loss of connectivity to your Prefect API / Cloud - API timeouts / transport errors / 5xx responses - The worker process getting stuck or starved for CPU/memory - Very slow environment / host pauses - Less commonly, auth/config issues with

PREFECT_API_URL

or

PREFECT_API_KEY

What I’d check first 1. Turn up worker logging Restart the worker with debug logging:

PREFECT_LOGGING_LEVEL=DEBUG PREFECT_WORKER_DEBUG_MODE=true prefect worker start --pool "your-pool"

Then look before the

has not polled

line for things like: -

TransportError

-

HTTPStatusError

- retries / backoff messages - connection reset / DNS / timeout errors 2. Verify API connectivity from the worker host From the same machine/container running the worker, confirm it can consistently reach your API. 3. Check host/container health If the worker host is under memory pressure, CPU throttling, or being paused/restarted by your platform, the polling loop can miss its deadline. 4. If you are using healthchecks/liveness probes A failing

/health

endpoint may cause Kubernetes/Docker/systemd to restart the worker, which is expected behavior once the worker is considered unhealthy. Mitigations - If your environment is just slow/latent, try increasing the worker poll interval:

export PREFECT_WORKER_QUERY_SECONDS=30

This also increases the unhealthy threshold to about

900s

because the threshold is based on that value. - Make sure the worker has stable outbound network access to Prefect - If running in Kubernetes, check for: - pod evictions - CPU throttling - OOM kills - node instability Healthcheck flag If you’re explicitly using worker healthchecks, the CLI option is:

prefect worker start --pool "your-pool" --with-healthcheck

I verified the command exists: - Prefect docs - Worker healthcheck behavior was introduced here: PR #9687 I also didn’t find a clear existing GitHub issue specifically diagnosing this exact message beyond the healthcheck implementation itself. If you want, I can help you narrow this down further if you share: - where the worker is running (Docker, Kubernetes, VM, ECS, etc.) - whether you’re using Prefect Cloud or self-hosted server - the 20–50 log lines before this error