[GitHub] [geode] DonalEvans commented on issue #4726: Revert "GEODE-7727: modify sender thread to detect release of connect…

["DonalEvans (GitHub)" <gi...@apache.org>]

> Hi,
> 
> can you share some info regarding this internal tests, so we can speed up detection of problem in this PR?
> 
> Thanks

Hi Mario,

The purpose of the internal test is to force async message queueing and confirm that no deadlocks occur. The test in question starts 6 servers with 5 regions in the cluster, each using a custom CacheListener that sleeps the thread for 2 ms when onCreate() is called with a particular event key in order to encourage queueing. Random events are delivered to the regions, including ones using the key that will trigger a 2ms sleep.

Each server is configured to run with 5 threads, conserve sockets is set to true, async distribution timeout is 5ms, async queue timeout is 60000ms and async max queue size is 10.

We consistently see the test hang when two members log that they are sleeping due to the onCreate() listener call. If only one member has logged that call then the test continues regardless of how many times its called, but as soon as a second member logs it, the test encounters problems and a hang is declared 5 minutes later.

We will continue to investigate this issue after the weekend, but I hope that the information here is helpful to you in the meantime.

[ Full content available at: https://github.com/apache/geode/pull/4726 ]
This message was relayed via gitbox.apache.org for notifications@geode.apache.org