zookeeper-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Hadoop QA (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ZOOKEEPER-1856) zookeeper C-client can fail to switch from a dead server in a 3+ server ensemble if the client only has a 2 server list.
Date Thu, 10 May 2018 20:45:00 GMT

    [ https://issues.apache.org/jira/browse/ZOOKEEPER-1856?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16471080#comment-16471080

Hadoop QA commented on ZOOKEEPER-1856:

-1 overall.  Here are the results of testing the latest attachment 
  against trunk revision 2fa315b7d0ed65828479fcdcc9e76ca8552fba4a.

    +1 @author.  The patch does not contain any @author tags.

    -1 tests included.  The patch doesn't appear to include any new or modified tests.
                        Please justify why no new tests are needed for this patch.
                        Also please list what manual steps were performed to verify this patch.

    -1 patch.  The patch command could not apply the patch.

Console output: https://builds.apache.org/job/PreCommit-ZOOKEEPER-Build/3680//console

This message is automatically generated.

> zookeeper C-client can fail to switch from a dead server in a 3+ server ensemble if the
client only has a 2 server list.
> ------------------------------------------------------------------------------------------------------------------------
>                 Key: ZOOKEEPER-1856
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-1856
>             Project: ZooKeeper
>          Issue Type: Bug
>          Components: c client
>            Reporter: Dutch T. Meyer
>            Assignee: Michi Mutsuzaki
>            Priority: Major
>             Fix For: 3.6.0, 3.5.5
>         Attachments: ZOOKEEPER-1856.patch
> If a client has a 2 server list, and is currently connected to the last server in that
list, and that server then goes offline, the addrvec_next() call handle_error() will push
the client to the start of the list and terminate the connection.
> Then, the zoo_cycle_next_server() call in zookeeper_interest will be called in response
to the connection failure, and the client will cycle back to the failed server.
> In this way, a client who has a list of only 2 servers can get stuck on the one failed
server.  This would only be an issue in an ensemble larger than 2 of course, because failing
1 out of 2 would lead to quorum loss anyway.
> There are other harmonics possible if every other server in the list is failed, but this
is simplest to reproduce in a 3 server ensemble where the client only knows about 2 servers,
one of which then fails.  There are probably some elegant fixes here, but I think the simplest
is to add a flag to track whether a server has been accessed before, and if it hasn't, don't
call zoo_cycle_next_server() at the top of the zookeeper_interest() function.

This message was sent by Atlassian JIRA

View raw message