qinjunjerry commented on a change in pull request #13309: URL: https://github.com/apache/flink/pull/13309#discussion_r483211118
########## File path: flink-libraries/flink-state-processing-api/src/main/java/org/apache/flink/state/api/WritableSavepoint.java ########## @@ -90,38 +93,37 @@ public final void write(String path) { List<OperatorState> existingOperators = metadata.getExistingOperators(); - DataSet<OperatorState> finalOperatorStates = unionOperatorStates(newOperatorStates, existingOperators); - - finalOperatorStates - .reduceGroup(new MergeOperatorStates(metadata.getMasterStates())) - .name("reduce(OperatorState)") - .output(new SavepointOutputFormat(savepointPath)) - .name(path); - } - - private DataSet<OperatorState> unionOperatorStates(DataSet<OperatorState> newOperatorStates, List<OperatorState> existingOperators) { DataSet<OperatorState> finalOperatorStates; if (existingOperators.isEmpty()) { finalOperatorStates = newOperatorStates; } else { - DataSet<OperatorState> wrappedCollection = newOperatorStates - .getExecutionEnvironment() + DataSet<OperatorState> existingOperatorStates = newOperatorStates.getExecutionEnvironment() .fromCollection(existingOperators); - finalOperatorStates = newOperatorStates.union(wrappedCollection); + existingOperatorStates + .flatMap(new StatePathExtractor()) Review comment: If we have a huge savepoint (e.g., in TB range) to process, do we also do with parallelism=1? meaning only one thread does the file copy, potentially many of them. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org