`job_def.submit(cluster)` fails with `503 Server Error`

[kpouget]

As part of my automated Codeflare testing, I'm hitting this exception:

ERROR:root:Caught exception HTTPError: 503 Server Error: Service Unavailable for url: http://ray-dashboard-mnisttest-user0-codeflare-sdk-user-test-user-0.apps.kpouget-sutest-20230726-07h01.psap.aws.rhperfscale.org/api/version
Traceback (most recent call last):
  File "/opt/ci-artifacts/src/testing/codeflare/test.py", line 180, in <module>
    sys.exit(main())
  File "/opt/ci-artifacts/src/testing/codeflare/test.py", line 175, in main
    fire.Fire(Entrypoint())
  File "/opt/venv/lib/python3.9/site-packages/fire/core.py", line 141, in Fire
    component_trace = _Fire(component, args, parsed_flag_args, context, name)
  File "/opt/venv/lib/python3.9/site-packages/fire/core.py", line 475, in _Fire
    component, remaining_args = _CallAndUpdateTrace(
  File "/opt/venv/lib/python3.9/site-packages/fire/core.py", line 691, in _CallAndUpdateTrace
    component = fn(*varargs, **kwargs)
  File "/opt/ci-artifacts/src/testing/codeflare/test.py", line 49, in wrapper
    fct(*args, **kwargs)
  File "/opt/ci-artifacts/src/testing/codeflare/test.py", line 148, in sdk_user_run_one
    test_sdk_user.run_one()
  File "/opt/ci-artifacts/src/testing/codeflare/test_sdk_user.py", line 165, in run_one
    timeout(entrypoint.main,
  File "/opt/ci-artifacts/src/testing/codeflare/test_sdk_user.py", line 148, in timeout
    return func(*args, **kwargs)
  File "/mnt/logs/002__run_one/sample.py", line 34, in main
    job = job_def.submit(cluster)
  File "/opt/venv/lib/python3.9/site-packages/codeflare_sdk/job/jobs.py", line 166, in submit
    return DDPJob(self, cluster)
  File "/opt/venv/lib/python3.9/site-packages/codeflare_sdk/job/jobs.py", line 174, in __init__
    self._app_handle = torchx_runner.schedule(job_definition._dry_run(cluster))
  File "/opt/venv/lib/python3.9/site-packages/torchx/runner/api.py", line 278, in schedule
    app_id = sched.schedule(dryrun_info)
  File "/opt/venv/lib/python3.9/site-packages/torchx/schedulers/ray_scheduler.py", line 199, in schedule
    client: JobSubmissionClient = JobSubmissionClient(
  File "/opt/venv/lib/python3.9/site-packages/ray/dashboard/modules/job/sdk.py", line 100, in __init__
    self._check_connection_and_version(
  File "/opt/venv/lib/python3.9/site-packages/ray/dashboard/modules/dashboard_sdk.py", line 228, in _check_connection_and_version
    self._check_connection_and_version_with_url(min_version, version_error_message)
  File "/opt/venv/lib/python3.9/site-packages/ray/dashboard/modules/dashboard_sdk.py", line 245, in _check_connection_and_version_with_url
    r.raise_for_status()
  File "/opt/venv/lib/python3.9/site-packages/requests/models.py", line 1021, in raise_for_status
    raise HTTPError(http_error_msg, response=self)
requests.exceptions.HTTPError: 503 Server Error: Service Unavailable for url: http://ray-dashboard-mnisttest-user0-codeflare-sdk-user-test-user-0.apps.kpouget-sutest-20230726-07h01.psap.aws.rhperfscale.org/api/version

This python file is being executed:

    # Create our cluster and submit appwrapper
    cluster = Cluster(ClusterConfiguration(
        namespace=namespace, name=f"mnisttest-user{user_idx}",
        min_worker=2, max_worker=2,
        min_cpus=2, max_cpus=2,
        min_memory=4, max_memory=4,
        gpu=0,
        instascale=False))
    # Bring up the cluster
    cluster.up()
    cluster.wait_ready()
    cluster.status()
    cluster.details()

    job_def = DDPJobDefinition(name="mnisttest", script="mnist.py", workspace=".", scheduler_args={"requirements": "./requirements.txt"})
    job = job_def.submit(cluster)

and the last line raises the exception.

---

• Codeflare SDK is installed from pip (latest version)
  • I'll remove the --quiet flag to capture the exact version being installed
• Codeflare stack is installed from ODH + OpenShift Codeflare operator

[kpouget]

hum, I noticed that I had this line missing from the example took the code from :

    image="quay.io/project-codeflare/ray:2.5.0-py38-cu116",

and thus, my RayCluster was created with the following (default?) image:

"image": "ghcr.io/foundation-model-stack/base:ray2.1.0-py38-gpu-pytorch1.12.0cu116-20221213-193103",

@Maxusmusti, is the default image supposed to work? 🤔

[Maxusmusti]

Yeah, that image should still work, though it's worth noting that in the upcoming release the default will be switching to the new 2.5.0 image linked there