[
https://issues.apache.org/jira/browse/KAFKA-2010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14353958#comment-14353958
]
Guozhang Wang commented on KAFKA-2010:
--------------------------------------
Jun, how frequently will this happen on your Mac?
Aside from the TCP issue (would be interesting to learn what is the triggering
threshold), I'm wondering if we should
1) Add backoff for the replica fetcher threads "HandleError", there is already
a ticket open and in progress for this;
2) Use a different timeout for Socket.connect().
Today the controllerSocketTimeoutMs is actually used in two places,
Socket.setSoTime() and Socket.connect(). Arguably for setting the socket
timeout we can use a large value of 30 sec, but we may probably want much
smaller values for timeout in connect calls.
> unit tests sometimes are slow during shutdown
> ---------------------------------------------
>
> Key: KAFKA-2010
> URL: https://issues.apache.org/jira/browse/KAFKA-2010
> Project: Kafka
> Issue Type: Bug
> Reporter: Jun Rao
> Assignee: Jun Rao
> Fix For: 0.8.3
>
>
> Our unit tests in trunk seem to be slower than before. The slowness seems to
> be due to a handful of tests.
> For example, if you run the following test, sometimes it can take more than
> 40
> secs, while normally it takes less than 10 secs.
> ./gradlew -i cleanTest core:test --tests
> kafka.admin.AddPartitionsTest.testIncrementPartitions
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
