Troubleshooting

Readiness timeout

First check the Mesos task states and callback-address reachability. A task can be TASK_RUNNING while its HTTP callback is still failing. Also verify that the container reports the correct port.

No matching offers

Compare the CPU, memory, and GPU requirements with the offer. With MESOS_ATTRIBUTES, the offer text must match the key:value constraint. Invalid values are rejected instead of being silently interpreted.

TLS or authentication errors

Make sure MESOS_SSL matches the listener, credentials are complete, and the CA certificate is available to the process. MESOS_VERIFY_SSL=false is not a substitute for a correct CA configuration.

TensorFlow does not start

Verify that TensorFlow can be imported inside the task image and that cluster_def contains every expected job with reachable host:port entries. The scheduler can be tested without TensorFlow; runtime validation requires a compatible image.