Skip to content

Latest commit

 

History

History
323 lines (249 loc) · 14.7 KB

troubleshooting-common-problems-with-distributed-workloads-for-administrators.adoc

File metadata and controls

323 lines (249 loc) · 14.7 KB
Raw

Troubleshooting common problems with distributed workloads for administrators

If your users are experiencing errors in {productname-long} relating to distributed workloads, read this section to understand what could be causing the problem, and how to resolve the problem.

If the problem is not documented here or in the release notes, contact {org-name} Support.

A user’s Ray cluster is in a suspended state

Problem

The resource quota specified in the cluster queue configuration might be insufficient, or the resource flavor might not yet be created.

Diagnosis

The user’s Ray cluster head pod or worker pods remain in a suspended state. Check the status of the Workloads resource that is created with the RayCluster resource. The status.conditions.message field provides the reason for the suspended state, as shown in the following example:

status:
 conditions:
   - lastTransitionTime: '2024-05-29T13:05:09Z'
     message: 'couldn''t assign flavors to pod set small-group-jobtest12: insufficient quota for nvidia.com/gpu in flavor default-flavor in ClusterQueue'
Resolution

  1. Check whether the resource flavor is created, as follows:

    1. Click Home → Search, and from the Resources list, select ResourceFlavor.

    2. If necessary, create the resource flavor.

  2. Check the cluster queue configuration in the user’s code, to ensure that the resources that they requested are within the limits defined for the project.

  3. If necessary, increase the resource quota.

For information about configuring resource flavors and quotas, see Configuring quota management for distributed workloads.

A user’s Ray cluster is in a failed state

Problem

The user might have insufficient resources.

Diagnosis

The user’s Ray cluster head pod or worker pods are not running. When a Ray cluster is created, it initially enters a failed state. This failed state usually resolves after the reconciliation process completes and the Ray cluster pods are running.

Resolution

If the failed state persists, complete the following steps:

  1. Click Workloads → Pods.

  2. Click the user’s pod name to open the pod details page.

  3. Click the Events tab, and review the pod events to identify the cause of the problem.

  4. Check the status of the Workloads resource that is created with the RayCluster resource. The status.conditions.message field provides the reason for the failed state.

A user receives a failed to call webhook error message for the CodeFlare Operator

Problem

After the user runs the cluster.up() command, the following error is shown:

ApiException: (500)
Reason: Internal Server Error
HTTP response body: {"kind":"Status","apiVersion":"v1","metadata":{},"status":"Failure","message":"Internal error occurred: failed calling webhook \"mraycluster.ray.openshift.ai\": failed to call webhook: Post \"https://codeflare-operator-webhook-service.redhat-ods-applications.svc:443/mutate-ray-io-v1-raycluster?timeout=10s\": no endpoints available for service \"codeflare-operator-webhook-service\"","reason":"InternalError","details":{"causes":[{"message":"failed calling webhook \"mraycluster.ray.openshift.ai\": failed to call webhook: Post \"https://codeflare-operator-webhook-service.redhat-ods-applications.svc:443/mutate-ray-io-v1-raycluster?timeout=10s\": no endpoints available for service \"codeflare-operator-webhook-service\""}]},"code":500}
Diagnosis

The CodeFlare Operator pod might not be running.

Resolution

  1. Click Workloads → Pods.

  2. Verify that the CodeFlare Operator pod is running. If necessary, restart the CodeFlare Operator pod.

  3. Review the logs for the CodeFlare Operator pod to verify that the webhook server is serving, as shown in the following example:

    INFO	controller-runtime.webhook	  Serving webhook server	{"host": "", "port": 9443}

A user receives a failed to call webhook error message for Kueue

Problem

After the user runs the cluster.up() command, the following error is shown:

ApiException: (500)
Reason: Internal Server Error
HTTP response body: {"kind":"Status","apiVersion":"v1","metadata":{},"status":"Failure","message":"Internal error occurred: failed calling webhook \"mraycluster.kb.io\": failed to call webhook: Post \"https://kueue-webhook-service.redhat-ods-applications.svc:443/mutate-ray-io-v1-raycluster?timeout=10s\": no endpoints available for service \"kueue-webhook-service\"","reason":"InternalError","details":{"causes":[{"message":"failed calling webhook \"mraycluster.kb.io\": failed to call webhook: Post \"https://kueue-webhook-service.redhat-ods-applications.svc:443/mutate-ray-io-v1-raycluster?timeout=10s\": no endpoints available for service \"kueue-webhook-service\""}]},"code":500}
Diagnosis

The Kueue pod might not be running.

Resolution

  1. Click Workloads → Pods.

  2. Verify that the Kueue pod is running. If necessary, restart the Kueue pod.

  3. Review the logs for the Kueue pod to verify that the webhook server is serving, as shown in the following example:

    {"level":"info","ts":"2024-06-24T14:36:24.255137871Z","logger":"controller-runtime.webhook","caller":"webhook/server.go:242","msg":"Serving webhook server","host":"","port":9443}

A user’s Ray cluster does not start

Problem

After the user runs the cluster.up() command, when they run either the cluster.details() command or the cluster.status() command, the Ray cluster status remains as Starting instead of changing to Ready. No pods are created.

Diagnosis

Check the status of the Workloads resource that is created with the RayCluster resource. The status.conditions.message field provides the reason for remaining in the Starting state. Similarly, check the status.conditions.message field for the RayCluster resource.

Resolution

  1. Click Workloads → Pods.

  2. Verify that the KubeRay pod is running. If necessary, restart the KubeRay pod.

  3. Review the logs for the KubeRay pod to identify errors.

A user receives a Default Local Queue …​ not found error message

Problem

After the user runs the cluster.up() command, the following error is shown:

Default Local Queue with kueue.x-k8s.io/default-queue: true annotation not found please create a default Local Queue or provide the local_queue name in Cluster Configuration.
Diagnosis

No default local queue is defined, and a local queue is not specified in the cluster configuration.

Resolution

  1. Check whether a local queue exists in the user’s project, as follows:

    1. Click Home → Search, and from the Resources list, select LocalQueue.

    2. If no local queues are found, create a local queue.

    3. Provide the user with the details of the local queues in their project, and advise them to add a local queue to their cluster configuration.

  2. Define a default local queue.

    For information about creating a local queue and defining a default local queue, see Configuring quota management for distributed workloads.

A user receives a local_queue provided does not exist error message

Problem

After the user runs the cluster.up() command, the following error is shown:

local_queue provided does not exist or is not in this namespace. Please provide the correct local_queue name in Cluster Configuration.
Diagnosis

An incorrect value is specified for the local queue in the cluster configuration, or an incorrect default local queue is defined. The specified local queue either does not exist, or exists in a different namespace.

Resolution

  1. Click Search, and from the Resources list, select LocalQueue.

  2. Resolve the problem in one of the following ways:

    • If no local queues are found, create a local queue.

    • If one or more local queues are found, provide the user with the details of the local queues in their project. Advise the user to ensure that they spelled the local queue name correctly in their cluster configuration, and that the namespace value in the cluster configuration matches their project name. If the user does not specify a namespace value in the cluster configuration, the Ray cluster is created in the current project.

  3. Define a default local queue.

    For information about creating a local queue and defining a default local queue, see Configuring quota management for distributed workloads.

A user cannot create a Ray cluster or submit jobs

Problem

After the user runs the cluster.up() command, an error similar to the following text is shown:

RuntimeError: Failed to get RayCluster CustomResourceDefinition: (403)
Reason: Forbidden
HTTP response body: {"kind":"Status","apiVersion":"v1","metadata":{},"status":"Failure","message":"rayclusters.ray.io is forbidden: User \"system:serviceaccount:regularuser-project:regularuser-workbench\" cannot list resource \"rayclusters\" in API group \"ray.io\" in the namespace \"regularuser-project\"","reason":"Forbidden","details":{"group":"ray.io","kind":"rayclusters"},"code":403}
Diagnosis

The correct OpenShift login credentials are not specified in the TokenAuthentication section of the user’s notebook code.

Resolution

  1. Advise the user to identify and specify the correct OpenShift login credentials as follows:

    1. In the new tab that opens, log in as the user whose credentials you want to use.

    2. Click Display Token.

    3. From the Log in with this token section, copy the token and server values.

    4. Specify the copied token and server values in your notebook code as follows:

      auth = TokenAuthentication(
    token = "<token>",
    server = "<server>",
    skip_tls=False
)
auth.login()

  2. Verify that the user has the correct permissions and is part of the rhoai-users group.

The user’s pod provisioned by Kueue is terminated before the user’s image is pulled

Problem

Kueue waits for a period of time before marking a workload as ready, to enable all of the workload pods to become provisioned and running. By default, Kueue waits for 5 minutes. If the pod image is very large and is still being pulled after the 5-minute waiting period elapses, Kueue fails the workload and terminates the related pods.

Diagnosis

  1. Click Workloads → Pods.

  2. Click the user’s pod name to open the pod details page.

  3. Click the Events tab, and review the pod events to check whether the image pull completed successfully.

Resolution

If the pod takes more than 5 minutes to pull the image, resolve the problem in one of the following ways:

  • Add an OnFailure restart policy for resources that are managed by Kueue.

  • In the redhat-ods-applications namespace, edit the kueue-manager-config ConfigMap to set a custom timeout for the waitForPodsReady property. For more information about this configuration option, see Enabling waitForPodsReady in the Kueue documentation.