zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alexander Shraer <shra...@gmail.com>
Subject Re: Question about recovering from a surviving minority in 3.5.0
Date Sun, 23 Nov 2014 00:26:03 GMT
Hi,

If a majority permanently fails you may loose state. But if that's really
what you want to do then yes,
the same thing would work in 3.5.0.

When modifying configuration files, you need to decide what to do about the
configuration version (nothing / remove it / increase). When two servers
talk the configuration with the higher version wins. Usually the
configuration version
equals to the zxid of the reconfig that created it, or the zxid of the
NEWLEADER message for the
first version, but that's not enforced, so theoretically you could bump it
up yourself (its now part of the filename)
or you could remove it and then zookeeper would create a new one equal to
the zxid of the NEWLEADER message when it comes up. If you make sure that
the configuration version is higher than it was before and the old servers
do come up for some reason, they will adopt the new config and everything
should be ok.

Alex

On Sat, Nov 22, 2014 at 4:00 PM, zk questions <zkquestions@gmail.com> wrote:

> Hi,
>
> I was wondering what the recommend practice is for recovering from the
> scenario where the majority of your nodes are permanently failed and we
> need to reconfigure the minority surviving nodes to let those nodes know
> that they are the only ones left.
>
> I believe that in 3.4.x, you would shutdown all the remaining surviving
> nodes, make changes to their configuration files to indicate that they are
> the only nodes left, and then start them up again.
>
> Is the same thing doable in 3.5.0 where the dynamic config is managed by
> ZK?
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message