Mailing-List: contact solr-user-help@lucene.apache.org; run by ezmlm
Precedence: bulk
Reply-To: solr-user@lucene.apache.org
Received-SPF: pass (nike.apache.org: local policy)
From: =?iso-8859-1?Q?Robert_Kr=FCger?= <krueger@signal7.de>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: quoted-printable
Subject: Pragmatic more or less high availability option on 2 servers
Date: Tue, 16 Feb 2010 11:01:21 +0100
Message-Id: <91D61ED8-C773-40E5-8D41-609F54FD793F@signal7.de>
To: solr-user@lucene.apache.org
Mime-Version: 1.0 (Apple Message framework v1077)


Hi,

I have to set up a SOLR cluster with some availability concept (is =
allowed to require manual interaction on fault, however, if there is a =
better way, I'd be interested in recommendations).

I have two servers (A and B for the example) at my disposal.

What I was thinking about was the following setup:

Set up 2 SOLR Instances on each server, i.e.

on A)
- One master (active)
- One slave (always active) replicating index from the master using the =
mechanisms described in http://wiki.apache.org/solr/SolrReplication

on B)
- One master (active but not used for index updates) replicating index =
form the master on A
- One slave (always active) replicating index from the master on A

I'll write the configs with placeholders for properties for the address =
of the master server so I can start the slaves with the master on A or =
on B as the master to replicate from.

When B goes down, nothing happens as the slave on A is still there to =
serve queries as well as the master to serve update requests.

When A goes down updates will begin to fail but queries will still be =
served from the slave on B. A restart of B with the until then inactive =
master as master will bring index update functionality up again.

For simplicity's sake I did not mention any load-balancing or IP address =
switching stuff. I assume that I have to load-balance the access to the =
two slaves and make sure that when the master is switched to B that =
either it assumes the IP address or the clients updating the index need =
to be changed )the forme probably being simpler).

Now, my questions are:
- Will this work?
- Can this be simplified, i.e. does it make sense to have separate =
master/slave instances when they are on the same hardware or is it =
enough or even better to write the config of the slave on B so that it =
can be restarted in "master mode" based on the same index?

Thanks in advance,

Robert