lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Erick Erickson <erickerick...@gmail.com>
Subject Re: keeping master server indexes in sync after failover recovery
Date Fri, 11 Nov 2011 14:17:24 GMT
You've pretty much nailed getting back in sync, although there's
a third way, just re-index all the changed content to M1 since M1
went down. Whether this is viable depends a lot on your
particular situation. You'd want to disable replication while the
reindexing happened.

Note that you have one other problem: any data that was
either uncommitted on the master or didn't replicated to M2
will have to be re-indexed on M2.

Actually, there's a 4th way. Just let M2 become the new primary
master permanently (depends on M2 being comparable to M1) and
when M1 gets repaired, it now becomes your master backup.

Making the masters slaves of each other is just *asking* for
trouble.

I guess my summary is that there's no magic to allow that to happen,
although we'll see with the SolrCloud stuff.

Best
Erick

On Thu, Nov 10, 2011 at 12:35 PM, Robert Stewart <bstewart.ny@gmail.com> wrote:
> If I have 2 masters in a master-master setup, where one master is the
> live master and the other master acts as backup in slave mode, and
> then during failover the slave master accepts new documents, such that
> indexes become out of sync, how can original live master index get
> back into sync with the slave master index?
>
> To be more clear:
>
> M1 - master SOLR server in master mode accepting all new document posts
> M2 - master SOLR server in slave mode pulling index replicas from M1
>
> M1 fails, so that new documents are posted to M2.
>
> M1 comes back to live, but now has index without new documents added
> to M1 index.
>
> How do I get M1 index back in sync with M2 index?
>
> Of course, it can be some manual copy of index directory from M2 to
> M1, or a re-config of M1 as a slave to pull replicas from M2 and then
> re-configure, etc.
>
> Is there a better or simpler way that is automatic or some other best
> practice?  Is it possible for both M1 and M2 be configured as slaves
> of each other so that they try to keep indexes in sync between
> themselves?  I am not sure how index "versioning" works with
> replication.
>
> Thanks
> Bob
>

Mime
View raw message