[
https://issues.apache.org/jira/browse/GEODE-10144?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17511339#comment-17511339
]
Barrett Oglesby commented on GEODE-10144:
-----------------------------------------
This issue comes down to a few factors:
- The default PoolFactory::DEFAULT_SUBSCRIPTION_ACK_INTERVAL of 100 seconds
means all the events stay on the queue for entire test. This means that every
time the client disconnects, the server starts over again processing the queue
from the beginning.
- The NC client is version GFE 9.0 (earlier) so no ClientReAuthenticateMessage
is sent to it when a AuthenticationExpiredException occurs. The server waits
anyway in case new credentials are sent through another operation.
- With the new changes, the server waits 5 seconds to be notified of a
re-auth. If no re-auth occurs, it waits the entire 5 seconds.
- With the old code, the server waits 200 ms before attempting to process the
event again (which includes asking for authorization again). The
SimulatedExpirationSecurityManager randomly decides whether to authorize the
event. 99% of the time, it returns true. So, the second request will almost
always return true.
- So without any external event (like new credentials):
-- With the the old code, the Message Dispatcher processes the event
successfully after 200 ms with no client disconnect
-- With the new code, the Message Dispatcher waits 5 seconds and then
disconnects the client
> Regression in geode-native test
> CqPlusAuthInitializeTest.reAuthenticateWithDurable
> ----------------------------------------------------------------------------------
>
> Key: GEODE-10144
> URL: https://issues.apache.org/jira/browse/GEODE-10144
> Project: Geode
> Issue Type: Bug
> Components: client/server
> Affects Versions: 1.15.0
> Reporter: Blake Bender
> Assignee: Jinmei Liao
> Priority: Major
> Labels: blocks-1.15.0, needsTriage
> Fix For: 1.15.0
>
>
> This test is failing across the board in the `geode-native` PR pipeline.
> Main develop pipeline is green only because nothing can get through the PR
> pipeline to clear checkin gates. We have green CI runs with 1.15. build 918,
> then it started failing when we picked up build 924.
>
> [~moleske] tracked this back to this commit:
> [https://github.com/apache/geode/commit/2554f42b925f2b9b8ca7eee14c7a887436b1d9db|https://github.com/apache/geode/commit/2554f42b925f2b9b8ca7eee14c7a887436b1d9db].
> See his notes in `geode-native` PR # 947
> ([https://github.com/apache/geode-native/pull/947])
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
