[
https://issues.apache.org/jira/browse/BOOKKEEPER-620?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13693350#comment-13693350
]
Sijie Guo commented on BOOKKEEPER-620:
--------------------------------------
{code}
You're right. I assumed wrongly that keySet() constructed a new set. I've
updated the patch to use guava's ImmutableSet to store the keys to be errored
out.
{code}
where the QPS is high, there will be lots of pending requests, copying keys
would introduce more memory. as my comments before, there is no issue in
original sequence (error out entries first and set DISCONNECTED later), why
change it?
> PerChannelBookieClient race during channel disconnect
> -----------------------------------------------------
>
> Key: BOOKKEEPER-620
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-620
> Project: Bookkeeper
> Issue Type: Bug
> Components: bookkeeper-client
> Reporter: Sijie Guo
> Assignee: Ivan Kelly
> Fix For: 4.2.2, 4.3.0
>
> Attachments:
> 0001-BOOKKEEPER-620-PerChannelBookieClient-race-during-ch.patch,
> 0001-BOOKKEEPER-620-PerChannelBookieClient-race-during-ch.patch,
> 0001-BOOKKEEPER-620-PerChannelBookieClient-race-during-ch.patch,
> 0001-BOOKKEEPER-620-PerChannelBookieClient-race-during-ch.patch,
> 0002-BOOKKEEPER-620-PerChannelBookieClient-race-during-ch.patch,
> BOOKKEEPER-620.diff
>
>
> channel & state are not synchronized in PerChannelBookieClient#closeInternal.
> so it might cause state is set to CONNECTED but the netty channel is closed
> by mistake in closeInternal.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
