[FLINK-22961] Incorrect calculation of alignment timeout for LocalInputChannel

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.14.0, 1.13.1
• Fix Version/s: 1.14.0, 1.13.2
• Component/s: Runtime / Checkpointing
• Labels: pull-request-available
• Assignee: Anton Kalashnikov
• Reporter: Anton Kalashnikov
• Created: 10/Jun/21 10:37
• Updated: 23/Sep/21 17:27
• Resolved: 23/Jun/21 06:49

Description

Right now, the calculation of alignment timeout happens inside of SingleCheckpointBarrierHandler(org.apache.flink.streaming.runtime.io.checkpointing.SingleCheckpointBarrierHandler.ControllerImpl#isTimedOut) and it based on  firstBarrierArrivalTime.  the  firstBarrierArrivalTime recalculated only when barrier announcement was received but if we receive the first checkpoint barrier from the LocalInputChannel which doesn't support announcement, the calculation of alignment timeout will be based on the firstBarrierArrivalTime from the previous checkpoint which is wrong.

Issue Links

• links to: GitHub Pull Request #16135