You are viewing a plain text version of this content. The canonical link for it is here.
Posted to solr-user@lucene.apache.org by Emmanuel GOURAUD <eg...@jouve.fr> on 2013/11/06 13:52:19 UTC

solr cloud bad replication

Hi all,

I have a solr cloud (solr 4.3.1) configuration that is composed of 8 instances:

4 shards with their replicas (solrcloud replication, not legacy one)

Due to a wrong assignment by solrcloud(the leader shard and its replica was on the same physical server), i must set the shardId parameter (the first shard is not concerned).

Now i have a file in the index directory which is successfuly replicated to the replica on the shard1 but not on the 3 others (shard 2 to 4 did not replicate correctly this file), this also happens when i do a full sync  (when i delete all files to force a full solr sync)

did someone already have this kind of problem?

thanks,

Emmanuel

Re: solr cloud bad replication

Posted by Emmanuel GOURAUD <eg...@jouve.fr>.
Hi shawn,

I configure 2 solr instances per server.

Concerning the file, i mean shard by shard (shard1 leader to shard 1 replica, shard2 leader to shard2 replica, etc...) and not shard2 leader to shard3 replica :)

It seams  that having a shardId forced make a difference (because it is the only difference between shard1 and the others)

thanks ,

Emmanuel
Le 6 nov. 2013 à 17:32, Shawn Heisey a écrit :

> On 11/6/2013 5:52 AM, Emmanuel GOURAUD wrote:
>> I have a solr cloud (solr 4.3.1) configuration that is composed of 8 instances:
>> 
>> 4 shards with their replicas (solrcloud replication, not legacy one)
>> 
>> Due to a wrong assignment by solrcloud(the leader shard and its replica was on the same physical server), i must set the shardId parameter (the first shard is not concerned).
> 
> If you set up Solr properly so there is one Solr instance per server,
> SolrCloud cannot make this mistake.  If you have more than one Solr
> instance per server, SolrCloud only knows that there are multiple
> instances, the fact that they live on the same server is not something
> it knows about.  Solr is more efficient if there is only one instance
> per server, which is why it's recommended.  One Solr instance can run
> many cores.  Shard replicas are implemented as cores.
> 
>> Now i have a file in the index directory which is successfuly replicated to the replica on the shard1 but not on the 3 others (shard 2 to 4 did not replicate correctly this file), this also happens when i do a full sync  (when i delete all files to force a full solr sync)
> 
> If you have files in the index directory for one of your shard1
> replicas, then those files can only be replicated to other replicas of
> shard1.  Shards 2 through 4 are completely separate indexes from shard1
> and will never receive shard1's files or documents.
> 
> Thanks,
> Shawn
> 


Re: solr cloud bad replication

Posted by Shawn Heisey <so...@elyograg.org>.
On 11/6/2013 5:52 AM, Emmanuel GOURAUD wrote:
> I have a solr cloud (solr 4.3.1) configuration that is composed of 8 instances:
> 
> 4 shards with their replicas (solrcloud replication, not legacy one)
> 
> Due to a wrong assignment by solrcloud(the leader shard and its replica was on the same physical server), i must set the shardId parameter (the first shard is not concerned).

If you set up Solr properly so there is one Solr instance per server,
SolrCloud cannot make this mistake.  If you have more than one Solr
instance per server, SolrCloud only knows that there are multiple
instances, the fact that they live on the same server is not something
it knows about.  Solr is more efficient if there is only one instance
per server, which is why it's recommended.  One Solr instance can run
many cores.  Shard replicas are implemented as cores.

> Now i have a file in the index directory which is successfuly replicated to the replica on the shard1 but not on the 3 others (shard 2 to 4 did not replicate correctly this file), this also happens when i do a full sync  (when i delete all files to force a full solr sync)

If you have files in the index directory for one of your shard1
replicas, then those files can only be replicated to other replicas of
shard1.  Shards 2 through 4 are completely separate indexes from shard1
and will never receive shard1's files or documents.

Thanks,
Shawn