You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hbase.apache.org by Dongsu Lee <do...@gmail.com> on 2014/08/14 23:13:06 UTC

Four nodes HBase cluster configuration

Hi,



Could you help me to find a guideline or recommendation for standing up
4-node HBase cluster?

I have read HBase in Action book and the book recommended not to have fewer
than 10 nodes in a production cluster. However, due to budget constraint,
we would like to begin with small cluster and expand it later depends on
traffic and volume. I am thinking of putting roles on each nodes like below:



*[1 Master Node]*

 NameNode

 Secondary Node

 Zookeeper

 HBase Master

 Resource Manager

 JohHistory Server





*[3 Data Nodes]*

 DataNode

 Region Server

 NodeManager



We will not run any heavyweight m/r jobs in backend to keep consistent
low-latency, but I am concerning availability as above configuration
doesn't have multi masters. Do you think it is safe to have HA
configuration later when we can add more nodes or should I do it from the
beginning?



1. Name Node HA

2. Three Zookeepers

3. Backup HBase Master



Thank you,

Dongsu

Re: Four nodes HBase cluster configuration

Posted by Dongsu Lee <do...@gmail.com>.
Thanks, Nick!


On Thu, Aug 14, 2014 at 2:50 PM, Nick Dimiduk <nd...@gmail.com> wrote:

> This is a fine role assignment if HA is not required. For true HBase HA
> you'll need at least HA namenode, multiple HBase masters, and a zookeeper
> quorum.
>
> On Thursday, August 14, 2014, Dongsu Lee <do...@gmail.com> wrote:
>
> > Hi,
> >
> >
> >
> > Could you help me to find a guideline or recommendation for standing up
> > 4-node HBase cluster?
> >
> > I have read HBase in Action book and the book recommended not to have
> fewer
> > than 10 nodes in a production cluster. However, due to budget constraint,
> > we would like to begin with small cluster and expand it later depends on
> > traffic and volume. I am thinking of putting roles on each nodes like
> > below:
> >
> >
> >
> > *[1 Master Node]*
> >
> >  NameNode
> >
> >  Secondary Node
> >
> >  Zookeeper
> >
> >  HBase Master
> >
> >  Resource Manager
> >
> >  JohHistory Server
> >
> >
> >
> >
> >
> > *[3 Data Nodes]*
> >
> >  DataNode
> >
> >  Region Server
> >
> >  NodeManager
> >
> >
> >
> > We will not run any heavyweight m/r jobs in backend to keep consistent
> > low-latency, but I am concerning availability as above configuration
> > doesn't have multi masters. Do you think it is safe to have HA
> > configuration later when we can add more nodes or should I do it from the
> > beginning?
> >
> >
> >
> > 1. Name Node HA
> >
> > 2. Three Zookeepers
> >
> > 3. Backup HBase Master
> >
> >
> >
> > Thank you,
> >
> > Dongsu
> >
>

Re: Four nodes HBase cluster configuration

Posted by Nick Dimiduk <nd...@gmail.com>.
This is a fine role assignment if HA is not required. For true HBase HA
you'll need at least HA namenode, multiple HBase masters, and a zookeeper
quorum.

On Thursday, August 14, 2014, Dongsu Lee <do...@gmail.com> wrote:

> Hi,
>
>
>
> Could you help me to find a guideline or recommendation for standing up
> 4-node HBase cluster?
>
> I have read HBase in Action book and the book recommended not to have fewer
> than 10 nodes in a production cluster. However, due to budget constraint,
> we would like to begin with small cluster and expand it later depends on
> traffic and volume. I am thinking of putting roles on each nodes like
> below:
>
>
>
> *[1 Master Node]*
>
>  NameNode
>
>  Secondary Node
>
>  Zookeeper
>
>  HBase Master
>
>  Resource Manager
>
>  JohHistory Server
>
>
>
>
>
> *[3 Data Nodes]*
>
>  DataNode
>
>  Region Server
>
>  NodeManager
>
>
>
> We will not run any heavyweight m/r jobs in backend to keep consistent
> low-latency, but I am concerning availability as above configuration
> doesn't have multi masters. Do you think it is safe to have HA
> configuration later when we can add more nodes or should I do it from the
> beginning?
>
>
>
> 1. Name Node HA
>
> 2. Three Zookeepers
>
> 3. Backup HBase Master
>
>
>
> Thank you,
>
> Dongsu
>