zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alvaro Gareppe <agare...@gmail.com>
Subject Re: Zookeeper in two data centers and disaster recovery
Date Fri, 09 Dec 2016 20:10:53 GMT
Thank you very much for your response..

Maybe the read only mode can be an option in this case. But how does it
works ? It olny allows conections that are read only ? or allows any
connection but fails on write operations ?






On Thu, Dec 8, 2016 at 10:05 PM, Michael Han <hanm@cloudera.com> wrote:

> If "keep running" is defined as serving both read and write requests then
> no there is no way to build a reliable ZooKeeper ensemble across only two
> data centers simply because we can't guarantee that the single data center
> remaining after another one is lost contains the majority number of servers
> that can form a quorum. In this case you need at least 3 DC. If "keep
> running" is defined as serving only read requests then you could enable
> read only mode of ZK server so if a DC failed and remaining servers fail to
> form a quorum, your application can still read data out of ZK, but no write
> would be done after the point.
>
> In general cross DC deployment design depends on your use case and trade
> offs you'd like to make. Similar topic was well discussed before [1], and
> there is a recent paper [2] which also provides good reference on this
> topic.
>
> [1]
> http://zookeeper-user.578899.n2.nabble.com/zookeeper-
> deployment-strategy-for-multi-data-centers-td7582358.html#a7582364
> [2]
> https://www.usenix.org/system/files/conference/atc16/atc16_
> paper-lev-ari.pdf
>
> On Thu, Dec 8, 2016 at 10:16 AM, Alvaro Gareppe <agareppe@gmail.com>
> wrote:
>
> > I have Zookeeper running in a PRD environment with 3 instances running. 2
> > in 1 data center and 1 in other. But as zookeeper ensamble runs with a
> > majority if a loose the data center that runs 2 nodes of zookeeper, I
> loose
> > the cluster.
> >
> > Is there any way to configure zookeeper (even adding more nodes is an
> > option) to keep running if some of the data center is offline (any one of
> > them). The only restiction, of course, is the number of datacenters
> >
> > --
> > Ing. Alvaro Gareppe
> > agareppe@gmail.com
> >
>
>
>
> --
> Cheers
> Michael.
>



-- 
Ing. Alvaro Gareppe
agareppe@gmail.com

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