[FLINK-30518] [flink-operator] Kubernetes HA Service not working with standalone mode

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: kubernetes-operator-1.3.0
• Fix Version/s: kubernetes-operator-1.4.0
• Component/s: Kubernetes Operator
• Labels: pull-request-available
• Assignee: Usamah Jassat
• Reporter: Binh-Nguyen Tran
• Created: 28/Dec/22 08:43
• Updated: 18/Jun/23 15:29
• Resolved: 03/Feb/23 13:39

Description

Since flink-conf.yaml is mounted as read-only configmap, the /docker-entrypoint.sh script is not able to inject correct Pod IP to `jobmanager.rpc.address`. This leads to same address (e.g flink.ns-ext) being set for all Job Manager pods. This causes:
Setting up FlinkDeployment in Standalone mode with Kubernetes HA Service. Problems:

(1) flink-cluster-config-map always contains wrong address for all 3 component leaders (see screenshot, should be pod IP instead of clusterIP service name)

(2) Accessing Web UI when jobmanager.replicas > 1 is not possible with error

{"errors":["Service temporarily unavailable due to an ongoing leader election. Please refresh."]} 

 

~ flinkdeployment.yaml ~

spec:
  mode: standalone
  flinkConfiguration:
    high-availability: kubernetes
    high-availability.storageDir: "file:///opt/flink/storage"
    ...
  jobManager:
    replicas: 3
  ... 

Issue Links

• relates to: FLINK-31187 Standalone HA mode does not work if dynamic properties are supplied
• links to: GitHub Pull Request #518