git.net

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[jira] [Created] (FLINK-9524) NPE from ProcTimeBoundedRangeOver.scala


yan zhou created FLINK-9524:
-------------------------------

             Summary: NPE from ProcTimeBoundedRangeOver.scala
                 Key: FLINK-9524
                 URL: https://issues.apache.org/jira/browse/FLINK-9524
             Project: Flink
          Issue Type: Bug
          Components: Table API & SQL
    Affects Versions: 1.5.0
            Reporter: yan zhou
         Attachments: npe_from_ProcTimeBoundedRangeOver.txt

The class _ProcTimeBoundedRangeOver_ would throws NPE _minRetentionTime_ and _maxRetentionTime_ are set to greater then 1. 

Please see [^npe_from_ProcTimeBoundedRangeOver.txt] for the detail of  exception. Below is a short description the cause:
 * When the first event for a key arrives,  the cleanup time is registered with _timerservice_ and recorded in _cleanupTimeState_. If the second event with same key arrives before the cleanup time, the value in _cleanupTimeState_ is updated and a new timer is registered to _timerService_. So now we have two registered timers for cleanup. One is registered because of the first event, the other for the second event.
 * However, when _onTimer_ method is fired for the first cleanup timer, the _cleanupTimeStates_ value has already been updated to second cleanup time. So it will bypass the _needToCleanupState_ check, and yet run through the remained code of _onTimer_ (which is intended to update the accumulator and emit output) and cause NPE.

_RowTimeBoundedRangeOver_ has very similar logic with _ProcTimeBoundedRangeOver. But_ It won't cause NPE by the same reason. To avoid the exception, it simply add a null check before running the logic for updating accumulator.

 

 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)