hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrew Purtell <apurt...@apache.org>
Subject Re: HBASE -- Regionserver and QuorumPeer ?
Date Mon, 02 Jul 2012 00:41:32 GMT
On Sun, Jul 1, 2012 at 2:05 PM, Jay Wilson
<registration@circle-cross-jn.com> wrote:
> Can a regionserver and quorumpeer reside on the same node?

It can, but you want to consider how disk is allocated in the cluster.

A typical and recommended configuration is HBase RegionServer and HDFS
DataNode colocated on the nodes. The DataNode will use locally
attached disk to store and serve blocks.

A ZooKeeper quorum peer must record transactions to its local log
before acking writes as part of the agreement protocol. Therefore you
will want to dedicate a storage device for this independent of other
use to minimize latency. If you are also putting a quorum peer aside a
RegionServer aside a DataNode, then you lose one block device.

Otherwise, during periods of heavy filesystem I/O the latency of
ZooKeeper writes may become quite large. Often heavy filesystem I/O
and ZooKeeper write demand coincide with HBase region transitions or
node failure recovery, so you are impacted most by this when you least
would want to be.

IMO, it is better to run a separate ZooKeeper ensemble, point HBase to
it, and then it is also available as an independent coordination
service for your applications because HBase use of it will mostly be
light.

Best regards,

   - Andy

Problems worthy of attack prove their worth by hitting back. - Piet
Hein (via Tom White)

Mime
View raw message