[
https://issues.apache.org/jira/browse/ZOOKEEPER-335?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13043441#comment-13043441
]
Vishal K commented on ZOOKEEPER-335:
------------------------------------
{quote}
vishal, i'm not sure if i understand your proposal correctly. when i originally
read it i thought you meant to have a client make sure to do an innocuous first
transaction, but you would run into timing issues there because another client
might be first.
{quote}
You are right about the timing issue. To avoid another client from
doing a transaction I do the null transactions from all clients :-)
Not efficient, but avoids this race. However, while testing this
client side workaround I ran into another problem. Our application
uses watchdogs and reboots the machines when zookeeper is disconnect
for long periods of time. As a result, the client cannot guarantee
that the first transaction to be logged will be always the null
transaction. It could be a createSession. So two nodes can still have
log divergence. If a node misses createSession from its transaction
log, then it will not expire it. doh! should have realized this. Now,
I can go a step ahead and make sure that the first session creation is
innocuous. This will work, but does not sound good. I will probably
give up on my idea for client side work around.
{quote}
your comment got me thinking about another way, which you were probably
proposing all along: if the leader issues the first innocuous transaction, and
makes sure it is committed before it processes any more it might be easier than
doing things at the quorum packet layer. i'm going to see if i can map it to
the flow in the tech report. it would help with the backwards compatibility.
{quote}
Yes, that was what I was proposing (and hence confused about backward
compatibility).
> zookeeper servers should commit the new leader txn to their logs.
> -----------------------------------------------------------------
>
> Key: ZOOKEEPER-335
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-335
> Project: ZooKeeper
> Issue Type: Bug
> Components: server
> Affects Versions: 3.1.0
> Reporter: Mahadev konar
> Assignee: Mahadev konar
> Priority: Blocker
> Fix For: 3.4.0
>
> Attachments: ZOOKEEPER-790.travis.log.bz2, faultynode-vishal.txt,
> zk.log.gz, zklogs.tar.gz
>
>
> currently the zookeeper followers do not commit the new leader election. This
> will cause problems in a failure scenarios with a follower acking to the same
> leader txn id twice, which might be two different intermittent leaders and
> allowing them to propose two different txn's of the same zxid.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira