Configuration #

Specifying Operator Configuration #

The operator allows users to specify default configuration that will be shared by the Flink operator itself and the Flink deployments.

These configuration files are mounted externally via ConfigMaps. The Configuration files with default values are shipped in the Helm chart. It is recommended to review and adjust them if needed in the values.yaml file before deploying the Operator in production environments.

To append to the default configuration, define the flink-conf.yaml key in the defaultConfiguration section of the Helm values.yaml file:

  create: true
  # Set append to false to replace configuration files
  append: true
  flink-conf.yaml: |+
    # Flink Config Overrides
    kubernetes.operator.metrics.reporter.slf4j.factory.class: org.apache.flink.metrics.slf4j.Slf4jReporterFactory
    kubernetes.operator.metrics.reporter.slf4j.interval: 5 MINUTE

    kubernetes.operator.reconcile.interval: 15 s 5 s

To learn more about metrics and logging configuration please refer to the dedicated docs page.

The operator also supports default configuration overrides for selected Flink versions and namespaces. This can be important if some behaviour changed across Flink versions or we want to treat certain namespaces differently (such as reconcile it more or less frequently etc).

# Flink Version specific defaults 
kubernetes.operator.default-configuration.flink-version.v1_17.k1: v1
kubernetes.operator.default-configuration.flink-version.v1_17.k2: v2
kubernetes.operator.default-configuration.flink-version.v1_17.k3: v3

# Namespace specific defaults
kubernetes.operator.default-configuration.namespace.ns1.k1: v1
kubernetes.operator.default-configuration.namespace.ns1.k2: v2
kubernetes.operator.default-configuration.namespace.ns2.k1: v1

Flink version specific defaults will have a higher precedence so namespace defaults would be overridden by the same key.

Dynamic Operator Configuration #

The Kubernetes operator supports dynamic config changes through the operator ConfigMaps. Dynamic operator configuration is enabled by default, and can be disabled by setting kubernetes.operator.dynamic.config.enabled to false. Time interval for checking dynamic config changes is specified by kubernetes.operator.dynamic.config.check.interval of which default value is 5 minutes.

Verify whether dynamic operator configuration updates is enabled via the deploy/flink-kubernetes-operator log has:

2022-05-28 13:08:29,222 o.a.f.k.o.c.FlinkConfigManager [INFO ] Enabled dynamic config updates, checking config changes every PT5M

To change config values dynamically the ConfigMap can be directly edited via kubectl patch or kubectl edit command. For example to change the reschedule interval you can override kubernetes.operator.reconcile.interval.

Verify whether the config value of kubernetes.operator.reconcile.interval is updated to 30 seconds via the deploy/flink-kubernetes-operator log has:

2022-05-28 13:08:30,115 o.a.f.k.o.c.FlinkConfigManager [INFO ] Updating default configuration to {kubernetes.operator.reconcile.interval=PT30S}

Leader Election and High Availability #

The operator supports high availability through leader election and standby operator instances. To enable leader election you need to add the following two mandatory operator configuration parameters.

kubernetes.operator.leader-election.enabled: true flink-operator-lease

Lease name must be unique in the current lease namespace. For other more advanced config parameters please refer to the configuration reference.

Once you enabled leader election you can increase the replicas for the operator Deployment using the Helm chart to enable high availability.

If replicas value is greater than 1, you can define topologySpreadConstraints via operatorPod.topologySpreadConstraints.

Environment variables #

The operator exposes several environment variables which can be used for custom plugins.

Name Description FieldRef
HOST_IP The host which the pod is deployed on status.hostIP
POD_IP Pod IP status.podIP

Operator Configuration Reference #

System Configuration #

General operator system configuration. Cannot be overridden on a per-resource basis.

Key Default Type Description
false Boolean Enables dynamic change of watched/monitored namespaces.
2048 Integer Maximum length of each exception field including stack trace to be included in CR status error field.
false Boolean Enable exception stacktrace to be included in CR status error field.
2048 Integer Maximum length of stacktrace to be included in CR status error field.
2 Integer Maximum number of throwable to be included in CR status error field.
1 min Duration The timeout for the reconciler to wait for flink to cancel job.
10 s Duration The timeout for the observer to wait the flink rest client to return.
false Boolean Enable leader election for the operator to allow running standby instances.
15 s Duration Leader election lease duration.
(none) String Leader election lease name, must be unique for leases in the same namespace.
10 s Duration Leader election renew deadline.
2 s Duration Leader election retry period.
1 min Duration The interval for the controller to reschedule the reconcile process.
200 Integer The maximum number of threads running the reconciliation loop. Use -1 for infinite.
1 min Duration The timeout for the resource clean up to wait for flink to shutdown cluster.
5 s Duration Initial interval of automatic reconcile retries on recoverable errors.
2.0 Double Interval multiplier of automatic reconcile retries on recoverable errors.
10 Integer Max attempts of automatic reconcile retries on recoverable errors.
"/opt/flink/artifacts" String The base dir to put the session job artifacts.
"JOSDK_ALL_NAMESPACES" String Comma separated list of namespaces the operator monitors for custom resources.

Resource/User Configuration #

These options can be configured on both an operator and a per-resource level. When set under spec.flinkConfiguration for the Flink resources it will override the default value provided in the operator default configuration (flink-conf.yaml).

Key Default Type Description
false Boolean Whether to enable checkpoint progress health check for clusters.
5 min Duration If no checkpoints are completed within the defined time window, the job is considered unhealthy. This must be bigger than checkpointing interval.
false Boolean Whether to enable health check for clusters.
64 Integer The threshold which is checked against job restart count within a configured window. If the restart count is reaching the threshold then full cluster restart is initiated.
2 min Duration The duration of the time window where job restart count measured.
5 min Duration The timeout for deployments to become ready/stable before being rolled back if rollback is enabled.
false Boolean Whether to enable rolling back failed deployment upgrades.
Map Key-Value pair where key is the REGEX to filter through the exception messages and value is the string to be included in CR status error label field if the REGEX matches. Expected format: headerKey1:headerValue1,headerKey2:headerValue2.
true Boolean Whether to enable recovery of missing/deleted jobmanager deployments.
86400000 ms Duration Time after which jobmanager pods of terminal application deployments are shut down.
true Boolean Enable job manager startup probe to allow detecting when the jobmanager could not submit the job.
false Boolean Whether to restart failed jobs.
false Boolean Indicate whether a savepoint must be taken when deleting a FlinkDeployment or FlinkSessionJob.
false Boolean Whether to ignore pending savepoint during job upgrade.
true Boolean Whether to enable inplace scaling for Flink 1.18+ using the resource requirements API. On failure or earlier Flink versions it falls back to regular full redeployment.
true Boolean Enables last-state fallback for savepoint upgrade mode. When the job is not running thus savepoint cannot be triggered but HA metadata is available for last state restore the operator can initiate the upgrade process when the flag is enabled.
(none) Duration Max allowed checkpoint age for initiating last-state upgrades on running jobs. If a checkpoint is not available within the desired age (and nothing in progress) a savepoint will be triggered.
0 ms Duration Interval at which periodic savepoints will be triggered. The triggering schedule is not guaranteed, savepoints will be triggered as part of the regular reconcile loop.
false Boolean Configure the array merge behaviour during pod merging. Arrays can be either merged by position or name matching.
true Boolean Whether to enable clean up of savepoint history.


Type of the binary format in which a savepoint should be taken.

Possible values:
  • "CANONICAL": A canonical, common for all state backends format. It lets you switch state backends.
  • "NATIVE": A format specific for the chosen state backend, in its native binary format. Might be faster to take and restore from than the canonical one.
86400000 ms Duration Maximum age for savepoint history entries to retain. Due to lazy clean-up, the most recent savepoint may live longer than the max age.
10 Integer Maximum number of savepoint history entries to retain.
1 min Duration The interval before a savepoint trigger attempt is marked as unsuccessful.
(none) Map Custom HTTP header for HttpArtifactFetcher. The header will be applied when getting the session job artifacts. Expected format: headerKey1:headerValue1,headerKey2:headerValue2.

Autoscaler Configuration #

Like other resource options these can be configured on both an operator and a per-resource level. When set under spec.flinkConfiguration for the Flink resources it will override the default value provided in the operator default configuration (flink-conf.yaml).

Key Default Type Description
5 min Duration Lag threshold which will prevent unnecessary scalings while removing the pending messages responsible for the lag.
5 min Duration The target duration for fully processing any backlog after a scaling operation. Set to 0 to disable backlog based scaling.
false Boolean Enable job autoscaler module.
86400000 ms Duration Maximum age for past scaling decisions to retain.
3 Integer Maximum number of past scaling decisions to retain per vertex.


Metric aggregator to use for busyTime metrics. This affects how true processing/output rate will be computed. Using max allows us to handle jobs with data skew more robustly, while avg may provide better stability when we know that the load distribution is even.

Possible values:
  • "AVG"
  • "MAX"
  • "MIN"
10 min Duration Scaling metrics aggregation window size.
3 min Duration Expected restart time to be used until the operator can determine it reliably from history.
0.6 Double Max scale down factor. 1 means no limit on scale down, 0.6 means job can only be scaled down with 60% of the original parallelism.
1 h Duration Duration in which no scale down of a vertex is allowed after it has been scaled up.
100000.0 Double Max scale up factor. 2.0 means job can only be scaled up with 200% of the current parallelism.
false Boolean Whether to enable detection of ineffective scaling operations and allowing the autoscaler to block further scale ups.
0.1 Double Processing rate increase threshold for detecting ineffective scaling threshold. 0.1 means if we do not accomplish at least 10% of the desired capacity increase with scaling, the action is marked ineffective.
true Boolean Enable vertex scaling execution by the autoscaler. If disabled, the autoscaler will only collect metrics and evaluate the suggested parallelism for each vertex but will not upgrade the jobs.
5 min Duration Stabilization period in which no new scaling will be executed
0.7 Double Target vertex utilization
0.4 Double Target vertex utilization boundary. Scaling won't be performed if the current processing rate is within [target_rate / (target_utilization - boundary), (target_rate / (target_utilization + boundary)]
List<String> A (semicolon-separated) list of vertex ids in hexstring for which to disable scaling. Caution: For non-sink vertices this will still scale their downstream operators until is implemented.
200 Integer The maximum parallelism the autoscaler can use. Note that this limit will be ignored if it is higher than the max parallelism configured in the Flink config or directly on each operator.
1 Integer The minimum parallelism the autoscaler can use.

System Metrics Configuration #

Operator system metrics configuration. Cannot be overridden on a per-resource basis.

Key Default Type Description
true Boolean Enable forwarding of Java Operator SDK metrics to the Flink metric registry.
true Boolean Enable Kubernetes Operator JVM metrics.
true Boolean Enable KubernetesClient metrics for measuring the HTTP traffic to the Kubernetes API Server.
false Boolean Enable KubernetesClient metrics for measuring the HTTP traffic to the Kubernetes API Server by response code group, e.g. 1xx, 2xx.
1000 Integer Defines the number of measured samples when calculating statistics.
"<host>.k8soperator.<namespace>.<name>.resource.<resourcens>.<resourcename>.<resourcetype>" String Defines the scope format string that is applied to all metrics scoped to the kubernetes operator resource.
"<host>.k8soperator.<namespace>.<name>.namespace.<resourcens>.<resourcetype>" String Defines the scope format string that is applied to all metrics scoped to the kubernetes operator resource namespace.
"<host>.k8soperator.<namespace>.<name>.system" String Defines the scope format string that is applied to all metrics scoped to the kubernetes operator.
true Boolean Enable resource lifecycle state metrics. This enables both state and transition counts/histograms.
true Boolean In addition to the system level histograms, enable per namespace tracking of state and transition times.
true Boolean Enables metrics for FlinkDeployment and FlinkSessionJob custom resources.

Advanced System Configuration #

Advanced operator system configuration. Cannot be overridden on a per-resource basis.

Key Default Type Description
1000 Integer Max config cache size.
10 min Duration Expiration time for cached configs.
5 min Duration Time interval for checking config changes.
true Boolean Whether to enable on-the-fly config changes through the operator configmap.
1 min Duration Allowed max time between spec update and reconciliation for canary resources.
true Boolean Enables health probe for the kubernetes operator.
8085 Integer The port the health probe will use to expose the status.
(none) String Label selector of the custom resources to be watched. Please see for the format supported.
10 s Duration The interval for observing status for in-progress operations such as deployment and savepoints.
10 s Duration Final delay before deployment is marked ready after port becomes accessible.


JM/TM Deployment deletion propagation.

Possible values:
  • "Orphan"
  • "Background"
  • "Foreground"
(none) Duration Maximum age threshold for savepoint history entries to retain.
(none) Integer Maximum number threshold of savepoint history entries to retain.
true Boolean Whether informer errors should stop operator startup. If false, the startup will ignore recoverable errors, caused for example by RBAC issues and will retry periodically.
10 s Duration Operator shutdown timeout before reconciliation threads are killed.