Home | About | Sematext search-lucene.com search-hadoop.com
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB
 Search Hadoop and all its subprojects:

Switch to Threaded View
Zookeeper >> mail # dev >> quorum of 4 servers


Copy link to this message
-
Re: quorum of 4 servers
Hi Ted,

It is always OK to have an even number of servers. The thing is that 4
servers is not more fault-tolerant than 3 servers - both allow only 1
failure.
It may still be better in terms of read throughput.

It all depends on the quorum system you use. If you use the majority
quorum system (the default in zookeeper) then you can only tolerate
the failure of a minority of servers,
so yes, if you want to tolerate N failures you need at least 2N+1servers.

Best Regards,
Alex

On Fri, Jul 20, 2012 at 2:05 PM, Ted Yu <[EMAIL PROTECTED]> wrote:
> Hi,
> I want to get people's opinion on the following statements:
>
> ----------
> While zk nodes are out of service (e.g. OS upgrade), it should be OK to
> have an even number of nodes.
>
> For zk or any other kind of voting system, it is misleading to focus on
> having an odd number of voters to avoid a tie vote.  The real focus should
> be "I want to tolerate at least N wrong or missing votes, so I need to have
> N + (N+1 or MORE) voters to make sure that N wrong or missing votes will
> not kill my service."  Typically people just build out (N + N+1) which
> means you are always starting with an odd number of voters.
> ----------
>
> Cheers
NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB