Barry Oglesby created GEODE-3815:
------------------------------------
Summary: Entries are incorrectly removed from the index map during
parallel gateway queue conflation
Key: GEODE-3815
URL: https://issues.apache.org/jira/browse/GEODE-3815
Project: Geode
Issue Type: Bug
Components: wan
Reporter: Barry Oglesby
When an event is sent to a {{Gateway}} queue, it does a put into a map of
indexes tracking realKey to queueKey if conflation is enabled. This put returns
the previous queueKey which sent to a separate {{WAN Queue Conflation Thread}}
thread to be removed from the queue. The {{WAN Queue Conflation Thread}} also
removes the key from the map of indexes. This is not correct behavior. Its
always going to remove a current index.
Here is some logging that shows the behavior. In this case, the
{{ServerConnection}} thread is updating key=0. The previousTailKey=726 (thats
the key in the queue), which the {{WAN Queue Conflation Thread}} removes from
the queue. It also removes index=839 from the index map. This causes the next
update for key=0 into the queue to not find anything in the index map
(previousTailKey=null), so the previous event is not removed from the queue.
{noformat}
ServerConnection on port 60268 Thread 1: BucketRegionQueue.conflateOldEntry
putting keyToConflate=0; tailKey=839
ServerConnection on port 60268 Thread 1: BucketRegionQueue.conflateOldEntry
previousTailKey=726
WAN Queue Conflation Thread: ConflationHandler.run previousTailKey=726
WAN Queue Conflation Thread: ParallelGatewaySenderQueue.destroyEventFromQueue
about to destroyKey key=726
WAN Queue Conflation Thread: BucketRegionQueue.removeIndex index=839
WAN Queue Conflation Thread: ParallelGatewaySenderQueue.destroyEventFromQueue
done destroyKey key=726
ServerConnection on port 60268 Thread 1: BucketRegionQueue.conflateOldEntry
putting keyToConflate=0; tailKey=952
ServerConnection on port 60268 Thread 1: BucketRegionQueue.conflateOldEntry
previousTailKey=null
{noformat}
If I remove this code from {{BucketRegionQueue.basicDestroy}} which is called
by the {{WAN Queue Conflation Thread}}, conflation works fine:
{noformat}
if (getPartitionedRegion().isConflationEnabled()) {
removeIndex((Long)event.getKey());
}
{noformat}
The {{WAN Queue Conflation Thread}} handles {{EntryNotFoundException}}, so if
the entry doesn't exist (meaning it already was sent), its ok. Since this
method is also called during normal queue removal, I'm not sure if there are
consequences to removing this code, though. Also, I haven't looked to see if
the serial queue has this same issue.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
