[
https://issues.apache.org/jira/browse/ZOOKEEPER-1987?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14073457#comment-14073457
]
Alexander Shraer commented on ZOOKEEPER-1987:
---------------------------------------------
to sum up some offline discussion with [~phunt]:
comments 1 and 2 - its possible to disable reconfig when the user doesn't
upgrade the config file or alternatively to avoid changing the config file and
return an error message without bringing up the server. I'd prefer the latter
because of the additional code that will be needed to patch the dynamic config
using to clientPort, disable reconfig, and creating a state where different
servers have different configs with the same config version. I'd rather clearly
state somewhere that users should add the ports to config file like I did in
the doc. Having said that, I understand that users may not be reading the docs,
so its up to you. We should warn the user that the reconfig is disabled only on
the server where config file wasn't upgraded, so if he upgrades part of the
config files but not all of them, things may go wrong.
comment 3 - since we can't write bak files atomically with changing the config
file, I suggest to number them like bak3, bak4, etc, I think this is what you
meant too. Writing it should probably be done in QuorumPeerConfig right before
the dynamic config file is updated.
comment 4 - please see ZOOKEEPER-1660, which also talks about upgrading from
3.4.6
Thanks,
Alex
> unable to restart 3 node cluster
> --------------------------------
>
> Key: ZOOKEEPER-1987
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-1987
> Project: ZooKeeper
> Issue Type: Bug
> Components: tests
> Affects Versions: 3.5.0
> Reporter: Patrick Hunt
> Assignee: Alexander Shraer
> Priority: Blocker
> Fix For: 3.5.0
>
> Attachments: ZOOKEEPER-1987-ver1.patch, ZOOKEEPER-1987-ver2.patch,
> ZOOKEEPER-1987-ver3.patch, ZOOKEEPER-1987.patch, f1.jstack, l3.jstack,
> test3.tar.gz
>
>
> I tried a fairly simple test, start a three node cluster, bring it down, then
> restart it. On restart the servers elect the leader and send updates, however
> the negotiation never completes - the client ports are never bound for
> example.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
