ClearML Agent fails scheduling tasks in Kubernetes after updating to v1.9.3

[dberkerdem]

After updating ClearML Agent that run in Kubernetes to v1.9.3, we start encountering following error on new scheduled tasks.

ERROR: Could not push back task [2667c614e10a46df90882ae3aa3ca7c8] to k8s pending queue k8s_scheduler [45bfd703d59245acbe3f3982fdb5f2d2], error: Validation error (Cannot skip setting execution queue for a task that is not enqueued or does not have execution queue set)

[mb-ii]

Seeing the same issue using Helm deployment (charts "clearml-7.14.1" and "clearml-agent-5.3.1")

[jkhenning]

Hi @dberkerdem and @@mb-ii

Can you share the exact server version/build from the UI's profile page?

[dberkerdem]

WebApp: 2.0.0-613 • Server: 2.0.0-613 • API: 2.31

[mb-ii]

WebApp: 2.0.0-613 • Server: 2.0.0-613 • API: 2.31

[mb-ii]

@jkhenning here's full Helm configuration for easier debugging:

values.yaml

clearml:
  elasticsearch:
    enabled: false
  mongodb:
    enabled: false
  redis:
    enabled: false

  apiserver:
    replicaCount: 1
    service:
      nodePort: 30040
    resources:
      requests:
        memory: 500Mi
        cpu: "500m"
      limits:
        memory: 500Mi
    ingress:
      enabled: true
      hostName: "***"

  fileserver:
    replicaCount: 1
    service:
      nodePort: 30041
    resources:
      requests:
        memory: 500Mi
        cpu: "500m"
      limits:
        memory: 500Mi
    storage:
      data:
        class: clearml-fileserver
        size: 10Gi
      enabled: true
    ingress:
      enabled: true
      hostName: "***"

  webserver:
    replicaCount: 1
    service:
      nodePort: 30042
    resources:
      requests:
        memory: 500Mi
        cpu: "500m"
      limits:
        memory: 500Mi
    ingress:
      enabled: true
      hostName: "***"

  externalServices:
    elasticsearchConnectionString: "***"
    redisHost: "***"
    redisPort: 6379
    mongodbConnectionStringAuth: '***'
    mongodbConnectionStringBackend: '***'

clearml-agent:
  agentk8sglue:
    webServerUrlReference: "***"
    apiServerUrlReference: "***"
    fileServerUrlReference: "***"
  clearml:
    agentk8sglueKey: "GGS9F4M6XB2DXJ5AFT9F"
    agentk8sglueSecret: "2oGujVFhPfaozhpuz2GzQfA5OyxmMsR3WVJpsCR5hrgHFs20PO"

Chart.yaml

apiVersion: v2
name: clearml-dev
version: 0.0.0
dependencies:
  - name: clearml
    version: 7.14.2
    repository: https://clearml.github.io/clearml-helm-charts
  - name: clearml-agent
    version: 5.3.1
    repository: https://clearml.github.io/clearml-helm-charts

[mb-ii]

May I ask what's the last Helm chart version this was working for you @dberkerdem ?

[dberkerdem]

I use 5.3.1 but it is not related with Chart version. To solve this problem you need to expose following environment variable agents environment. The problem is caused by dynamic installation of latest ClearML version on ClearML Agent startup.See behavior

      agentk8sglue:
          extraEnvs:
            - name: CLEARML_AGENT_UPDATE_VERSION
              value: "==1.9.2"

[mb-ii]

Thanks, it helped! @dberkerdem

[illuser-maker]

I also saw this issue in clearml-agent 1.9.3. Downgrade to 1.9.2 helped. It seems a bug in 1.9.3 version :/