Patrick Hunt updated ZOOKEEPER-473:

    Attachment: ZOOKEEPER-473.patch

this patch addresses the stated problems.

Additional note -- I've moved all of the log4j output to a logfile in the 
build/test/logs directory.

this was necessary for testng since it considers "stdout" special - and we were 
logging all
of the log4j to the console previously. I think this change is an improvement 

1) hudson console information will be much easier to view (currently huge)
2) there's a single file to open/view for all log output. in general this is 
 and specifically it's nice if you are trying to see if a test at time t is 
causing problems
with a subsequent t+n test (like for this jira)
3) the junit test output log files for each test now are much more 
readable/parsable, they
contain just the results of the test (incl excep) and any stdout/stderr.

Please review and commit asap.

> cleanup junit tests to eliminate false positives due to "socket reuse" and 
> failure to close client
> --------------------------------------------------------------------------------------------------
>                 Key: ZOOKEEPER-473
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-473
>             Project: Zookeeper
>          Issue Type: Improvement
>          Components: tests
>    Affects Versions: 3.2.0
>            Reporter: Patrick Hunt
>            Assignee: Patrick Hunt
>             Fix For: 3.3.0
>         Attachments: ZOOKEEPER-473.patch
> The junit tests will occasionally fail on hudson due to a few problems:
> 1) socket in use - sometimes the code/testcode will fail to close a socket, 
> either that or tcp wait
>  kicks in and the kernel holds onto the socket, either way subsequent tests 
> that use this same
> port will fail
> 2) sometimes the tests fail to close clients, this results in clients 
> attempting to re-connect to
> the server started as part of a subsequent test (the timeout is very long). 
> this causes a few
> problems, typically extra load on the server or more likely unusual changes 
> to the server that the
> next test(s) don't expect. also causes problems when trying to debug the test 
> as the old clients
> spew messages into the log.
> we see this on hudson, occasionally failures. We also saw this when working 
> on ZOOKEEPER-397, the testng
> port. testng does not support "fork mode" for running the tests. as a result 
> all the tests (all) run in a single
> vm instance, start to finish. so this problem is exacerbated significantly in 
> that case. this patch only
> fixes the problem in junit btw, it does not move to testng

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

Reply via email to