[
https://issues.apache.org/jira/browse/GEODE-9704?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17512627#comment-17512627
]
Mark Hanson commented on GEODE-9704:
------------------------------------
PR 7442 is available. I have made changes to fix the behavior that was causing
the problems. The core of the problem was that registerinterst should be
called before readyforevents. It was reversed effectively, so that has been
corrected.
LocalRegionUpdateTest.java was created to house two unit tests for the new code.
AuthExpirationDUnitTest has a test by Jinmei that has been uncommented that
would typically be flaky, but with this fix, no longer fails.
I believe this bug is done with the exception of the review phase of the PR and
associated changes.
> When durable clients recovers, it sends "ready for event" signal before
> register for interest, this might cause problem for caching_proxy regions
> -------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: GEODE-9704
> URL: https://issues.apache.org/jira/browse/GEODE-9704
> Project: Geode
> Issue Type: Bug
> Components: regions
> Affects Versions: 1.15.0
> Reporter: Jinmei Liao
> Assignee: Mark Hanson
> Priority: Major
> Labels: GeodeOperationAPI, blocks-1.15.1, pull-request-available
>
> This is the old Geode behavior, but may or may not be the correct behavior.
> When durable clients recovers, there is a queueTimer thread that runs
> `QueueManagerImp.recoverPrimary` method, it
> * makes new connection to server
> - sends readyForEvents (which will cause the server to start sending the
> queued events)
> - recovers interest
> - clears the region of keys of interest
> - re-registers interest
> It sends readyForEvents before it clears region of keys of interest, if
> server sends some events of those keys in between, it will clear them, thus
> it seems to the user that the client region doesn't have those keys.
>
> Run geode-core distributedTest
> AuthExpirationDUnitTest.registeredInterest_slowReAuth_policyKeys_durableClient(),
> change the InterestResultPolicy to NONE, you would see the test would fail
> occasionally, Adding sleep code in QueueManagerImp.recoverPrimary between
> `createNewPrimary` and `recoverInterest` would make the test fail more
> consistently.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
