lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Shawn Heisey <apa...@elyograg.org>
Subject Re: Poll: Master-Slave or SolrCloud?
Date Sun, 30 Apr 2017 15:35:35 GMT
On 4/25/2017 3:13 PM, Otis Gospodnetić wrote:
> Could one run *only* embedded ZK on some SolrCloud nodes, sans any data?
> It would be equivalent of dedicated Elasticsearch nodes, which is the
> current ES best practice/recommendation.  I've never heard of anyone being
> scared of running 3 dedicated master ES nodes, so if SolrCloud offered the
> same, perhaps even completely hiding ZK from users, that would present the
> same level of complexity (err, simplicity) ES users love about ES.  Don't
> want to talk about SolrCloud vs. ES here at all, just trying to share
> observations since we work a lot with both Elasticsearch and Solr(Cloud) at
> Sematext.

Yes, you could do that ... but I don't see any real value right now. 
You have to learn how to configure a redundant ZK ensemble and apply
that configuration to the embedded servers manually.  Since that's not
any different from what you'd do with an external ensemble, I think it's
better to just use the external install.  As I understand it, elastic
wrote their cluster code themselves ... it's part of ES, not provided by
a separate software package, so their recommendation makes sense for ES.

Using embedded ZK as you have described, there will be at least three
extra Solr nodes that are not intended to host collections.  To keep it
running this way, it will be important to explicitly avoid putting new
collections on those nodes, because that won't happen by default.  With
dedicated external ZK processes, there's no Solr node to worry about,
and no need to create a "master node" capability.

I'm not opposed to automated scripts included with Solr to configure and
start standalone ZK processes, including a way to create an init
script.  That would be very useful and go a long way towards extremely
easy instructions for setting up a fault tolerant SolrCloud installation
on multiple servers.

In situations where ZK is installed on dedicated hardware, a native ZK
will require less heap memory than one embedded in Solr, and probably
will have slightly lower CPU requirements.

SOLR-9386 does make your idea more viable because it brings the full
capability of recent zookeeper configuration options to the embedded
server.  It will be available in version 6.6.

Thanks,
Shawn


Mime
View raw message