quorum, consistency and datacenter awareness

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

quorum, consistency and datacenter awareness

HubertChang

ElasticSearch works in the cluster.
Is there a quorum concept? If the replication factor is 2 or 3 and one
replication fails, is the content still available?
Is there a param to read from _all_replications and read from
_one_replication to maintain the consitency?
Will the replication be datacenter awared in the future?
Thanks.
--
View this message in context: http://n3.nabble.com/quorum-consistency-and-datacenter-awareness-tp696373p696373.html
Sent from the ElasticSearch Users mailing list archive at Nabble.com.
Reply | Threaded
Open this post in threaded view
|

Re: quorum, consistency and datacenter awareness

kimchy
Administrator
Date center has been discussed here: http://groups.google.com/a/elasticsearch.com/group/users/browse_thread/thread/44e8e96420d78201/0275c454ae80ad01?lnk=gst&q=data+center#0275c454ae80ad01.

ElaticSearch clustering module is different than cassandra and other dynamo based systems. You don't have to read from all/quorum replicas in order to maintain consistency. When something is indexed, it is indexed on a primary shard and then replicated to all its replicas synchronously (if you have more than one replica, it is done in parallel using async io).

Note, that elasticsearch provides near real time for indexed documents to be visible when searching/getting them. But, if (when) it provides real time support, then once you indexed, it will be provided immediately and no need for quorum based reads.

When doing a read / search, one of the replicas is chosen and the reads is done from it. Search is never done on more than one replica of the same shard. This gives you better performance and better read/search scalability just by increasing the number of replicas.

-shay.banon

On Sun, Apr 4, 2010 at 7:59 PM, HubertChang <[hidden email]> wrote:

ElasticSearch works in the cluster.
Is there a quorum concept? If the replication factor is 2 or 3 and one
replication fails, is the content still available
Is there a param to read from _all_replications and read from
_one_replication to maintain the consitency?
Will the replication be datacenter awared in the future?
Thanks.
--
View this message in context: http://n3.nabble.com/quorum-consistency-and-datacenter-awareness-tp696373p696373.html
Sent from the ElasticSearch Users mailing list archive at Nabble.com.