[ 
https://issues.apache.org/jira/browse/GEODE-8971?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17367898#comment-17367898
 ] 

ASF subversion and git services commented on GEODE-8971:
--------------------------------------------------------

Commit 86a8b1ea4c26b515ca96b1230c993c8ca1d7ed8a in geode's branch 
refs/heads/develop from Alberto Gomez
[ https://gitbox.apache.org/repos/asf?p=geode.git;h=86a8b1e ]

Revert "GEODE-8971: Add grace period when stopping gateway sender with group-… 
(#6052)" (#6634)

This reverts commit 841fa06c7b34916c09df920b7029974a78255cd0.

The reason is that this commit creates problems with ongoing
put operations during the grace period (very long response
times) and also, if operations rate is high, it could prevent
that the gateway sender is stopped.

Put operations can be very slow during the grace period
because they need to traverse the gateway sender queue
to find events with the same transactionId. If the queue
size is big and events are evicted to disk, the time to
process a request can be unacceptably long.
At the same time, this can provoke that the gateway sender
is never stopped because it is blocked trying to get
a write lock and the lock is held by ongoing operations
for a long time.

> Batches with incomplete transactions when stopping the gateway sender
> ---------------------------------------------------------------------
>
>                 Key: GEODE-8971
>                 URL: https://issues.apache.org/jira/browse/GEODE-8971
>             Project: Geode
>          Issue Type: Improvement
>          Components: wan
>    Affects Versions: 1.14.0
>            Reporter: Alberto Gomez
>            Assignee: Alberto Gomez
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 1.15.0
>
>
> When the gateway sender is stopped there is a high probability that batches 
> with incomplete transactions are sent even if group-transaction-events is 
> enabled.
> The reason is that once the stop command reaches the gateway sender, it 
> immediately stops queueing events, and this could happen in the middle of 
> receiving events for the same transaction. If this is the case, some events 
> for the transaction may have reached the queue right before the stop command 
> was received and the rest of events for that transaction would not make it to 
> the queue (they would be dropped) because they arrived right after the stop 
> command was received at the gateway sender.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to