[ https://issues.apache.org/jira/browse/BEAM-7112?focusedWorklogId=233917&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-233917 ]
ASF GitHub Bot logged work on BEAM-7112: ---------------------------------------- Author: ASF GitHub Bot Created on: 27/Apr/19 09:04 Start Date: 27/Apr/19 09:04 Worklog Time Spent: 10m Work Description: mxm commented on pull request #8399: [BEAM-7112] Timer race with state cleanup - take two URL: https://github.com/apache/beam/pull/8399#discussion_r279148991 ########## File path: runners/flink/src/main/java/org/apache/beam/runners/flink/translation/wrappers/streaming/ExecutableStageDoFnOperator.java ########## @@ -812,21 +796,35 @@ public boolean isForWindow( private final List<String> userStateNames; private final Coder windowCoder; - private final StateInternals stateInternals; + private final ArrayDeque<KV<ByteBuffer, BoundedWindow>> cleanupQueue; + private final Supplier<ByteBuffer> keyedStateBackend; - StateCleaner(List<String> userStateNames, Coder windowCoder, StateInternals stateInternals) { + StateCleaner( + List<String> userStateNames, Coder windowCoder, Supplier<ByteBuffer> keyedStateBackend) { this.userStateNames = userStateNames; this.windowCoder = windowCoder; - this.stateInternals = stateInternals; + this.keyedStateBackend = keyedStateBackend; + this.cleanupQueue = new ArrayDeque<>(); Review comment: Right, if after `invokeFinishBundle()` has been called the cleanup queue is always cleared as part of however many additional bundles, then we are fine. If we delay any cleanup for after the checkpoint has completed, then we would have a problem. ---------------------------------------------------------------- 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 Issue Time Tracking ------------------- Worklog Id: (was: 233917) Time Spent: 7h 50m (was: 7h 40m) > State cleanup interferes with user timer callback > ------------------------------------------------- > > Key: BEAM-7112 > URL: https://issues.apache.org/jira/browse/BEAM-7112 > Project: Beam > Issue Type: Bug > Components: runner-flink > Affects Versions: 2.12.0 > Reporter: Thomas Weise > Assignee: Thomas Weise > Priority: Major > Labels: portability-flink > Fix For: 2.13.0 > > Time Spent: 7h 50m > Remaining Estimate: 0h > > Cleanup timers and user timers are fired at the watermark. Processing of > timers in the SDK worker is asynchronous, so it is possible that the state is > already removed when the user timer callback executes. -- This message was sent by Atlassian JIRA (v7.6.3#76005)