Skip to main content

Flink Upgrade (Cross-Version)

  1. Delete ConfigMap and clean up flink Job list

    Execute script to delete ConfigMap in batches:

    If words like configmap "md-flink" deleted are not output, it means the namespace is wrong or ConfigMap does not exist, please check again.

    #Template: Please replace <namespace> with the actual namespace of your flink deployment
    for i in $(kubectl -n <namespace> get cm | awk '$1~"md-flink"{print $1}'); do kubectl -n <namespace> delete cm $i; done

    #Example (namespace is flink):
    for i in $(kubectl -n flink get cm | awk '$1~"md-flink"{print $1}'); do kubectl -n flink delete cm $i; done

  2. Import the image of the corresponding architecture into all Flink nodes (Offline Package Download)

    crictl pull nocoly/flink:version

  3. Modify configuration file

    Modify the image version used by flink-jobmanager and flink-taskmanager services in flink.yaml

    - name: jobmanager
      image: nocoly/flink:version
    - name: taskmanager
      image: nocoly/flink:version
    When upgrading to v1.19.710, click to view more adjustment details of flink.yaml

    1. Delete all original configuration items starting with metrics, and add the following Kafka indicator reporting configuration (please replace the Kafka address according to the actual environment):

       metrics.job.status.enable: STATE
       metrics.reporters: kafka_reporter,kafka_reporter_running,kafka_reporter2,kafka_reporter_running2
       metrics.reporter.kafka_reporter.factory.class: org.apache.flink.metrics.kafka.KafkaReporterFactory
       metrics.reporter.kafka_reporter.bootstrap.servers: 192.168.10.7:9092,192.168.10.8:9092,192.168.10.9:9092 #kafka address
       metrics.reporter.kafka_reporter.chunk.size: 20000
       metrics.reporter.kafka_reporter.interval: 60s
       metrics.reporter.kafka_reporter.filter.metrics: numRecordsIn,numRecordsOut,runningTime
       metrics.reporter.kafka_reporter.topic: flink_metrics_counter
       metrics.reporter.kafka_reporter.taskNamePrefix: HDP0x5c2_
       metrics.reporter.kafka_reporter_running.factory.class: org.apache.flink.metrics.kafka.KafkaReporterFactory
       metrics.reporter.kafka_reporter_running.bootstrap.servers: 192.168.10.7:9092,192.168.10.8:9092,192.168.10.9:9092 #kafka address
       metrics.reporter.kafka_reporter_running.chunk.size: 20000
       metrics.reporter.kafka_reporter_running.interval: 60s
       metrics.reporter.kafka_reporter_running.filter.metrics: RUNNINGState
       metrics.reporter.kafka_reporter_running.topic: flink_metrics_gauge
       metrics.reporter.kafka_reporter_running.taskNamePrefix: HDP0x5c2_
       metrics.reporter.kafka_reporter2.factory.class: org.apache.flink.metrics.kafka.KafkaReporterFactory
       metrics.reporter.kafka_reporter2.bootstrap.servers: 192.168.10.7:9092,192.168.10.8:9092,192.168.10.9:9092 #kafka address
       metrics.reporter.kafka_reporter2.chunk.size: 20000
       metrics.reporter.kafka_reporter2.interval: 60s
       metrics.reporter.kafka_reporter2.filter.metrics: numRecordsIn,numRecordsOut,runningTime
       metrics.reporter.kafka_reporter2.topic: flink_metrics_counter-hdp
       metrics.reporter.kafka_reporter2.taskNamePrefix: HDP0x5c2_
       metrics.reporter.kafka_reporter_running2.factory.class: org.apache.flink.metrics.kafka.KafkaReporterFactory
       metrics.reporter.kafka_reporter_running2.bootstrap.servers: 192.168.10.7:9092,192.168.10.8:9092,192.168.10.9:9092 #kafka address
       metrics.reporter.kafka_reporter_running2.chunk.size: 20000
       metrics.reporter.kafka_reporter_running2.interval: 60s
       metrics.reporter.kafka_reporter_running2.filter.metrics: RUNNINGState
       metrics.reporter.kafka_reporter_running2.topic: flink_metrics_gauge-hdp
       metrics.reporter.kafka_reporter_running2.taskNamePrefix: HDP0x5c2_

    2. Find the kind: Role configuration section and add patch permissions to the configmaps resource under the rules.verbs field.

      kind: Role
      apiVersion: rbac.authorization.k8s.io/v1
      metadata:
        name: configmap-access
        namespace: default
      rules:
      - apiGroups: [""]
        resources: ["configmaps"]
        verbs: ["update", "get", "watch", "list", "create", "edit", "delete", "patch"] #Add patch permission here

  4. Restart the service

    kubectl apply -f flink.yaml

  5. After the upgrade is completed, please manually reopen/publish the relevant tasks in the synchronization task list, and the synchronization tasks will resume normal operation.