yunfengzhou-hub commented on code in PR #22931: URL: https://github.com/apache/flink/pull/22931#discussion_r1267857378
########## flink-runtime/src/main/java/org/apache/flink/runtime/operators/coordination/OperatorCoordinatorHolder.java: ########## @@ -164,6 +167,15 @@ public void lazyInitialize( context.lazyInitialize( globalFailureHandler, mainThreadExecutor, operatorCoordinatorMetricGroup); + OperatorCoordinator rootCoordinator = coordinator; + if (coordinator instanceof RecreateOnResetOperatorCoordinator) { + rootCoordinator = + ((RecreateOnResetOperatorCoordinator) rootCoordinator).getInternalCoordinator(); + } + if (rootCoordinator instanceof SourceCoordinator) { + ((SourceCoordinator<?, ?>) rootCoordinator).lazyInitialize(checkpointCoordinator); Review Comment: Below is part of the invocation stack of SourceCoordinatorContext's constructor. ``` <init>:142, SourceCoordinatorContext (org.apache.flink.runtime.source.coordinator) <init>:119, SourceCoordinatorContext (org.apache.flink.runtime.source.coordinator) getCoordinator:89, SourceCoordinatorProvider (org.apache.flink.runtime.source.coordinator) createNewInternalCoordinator:337, RecreateOnResetOperatorCoordinator$DeferrableCoordinator (org.apache.flink.runtime.operators.coordination) <init>:60, RecreateOnResetOperatorCoordinator (org.apache.flink.runtime.operators.coordination) <init>:43, RecreateOnResetOperatorCoordinator (org.apache.flink.runtime.operators.coordination) create:200, RecreateOnResetOperatorCoordinator$Provider (org.apache.flink.runtime.operators.coordination) create:194, RecreateOnResetOperatorCoordinator$Provider (org.apache.flink.runtime.operators.coordination) create:546, OperatorCoordinatorHolder (org.apache.flink.runtime.operators.coordination) create:509, OperatorCoordinatorHolder (org.apache.flink.runtime.operators.coordination) createOperatorCoordinatorHolder:286, ExecutionJobVertex (org.apache.flink.runtime.executiongraph) initialize:223, ExecutionJobVertex (org.apache.flink.runtime.executiongraph) initializeJobVertex:888, DefaultExecutionGraph (org.apache.flink.runtime.executiongraph) initializeJobVertex:218, ExecutionGraph (org.apache.flink.runtime.executiongraph) initializeJobVertices:870, DefaultExecutionGraph (org.apache.flink.runtime.executiongraph) attachJobGraph:826, DefaultExecutionGraph (org.apache.flink.runtime.executiongraph) buildGraph:219, DefaultExecutionGraphBuilder (org.apache.flink.runtime.executiongraph) createAndRestoreExecutionGraph:163, DefaultExecutionGraphFactory (org.apache.flink.runtime.scheduler) createAndRestoreExecutionGraph:368, SchedulerBase (org.apache.flink.runtime.scheduler) <init>:210, SchedulerBase (org.apache.flink.runtime.scheduler) <init>:140, DefaultScheduler (org.apache.flink.runtime.scheduler) ``` In order to pass `checkpointCoordinator` through this stack to SourceCoordinatorContext's constructor, we may need to add a `getCheckpointCoordinator` method to `OperatorCoordinator.Context` class. This is a public API change that has not been agreed in the FLIP discussion, and I also personally think this API will yield too much freedom to OperatorCoordinator's developers so should not be introduced. Thus I did not choose this path. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@flink.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org