RE: SolrCould read-only replicas

[Sandeep Tikoo <st...@digitalriver.com>]

Erick,

Thank you for your response. Yup, when I said it is not possible to have a cross continent data center replica, I meant that we never ever want to do that because of the latency.

What I was hoping is that  I could have Solr cloud in my DataCentre A (DC-A) and get all the benefits of sharding ( scaling/parallel computing) and failover redundancy within the same data center. If I could then have a read-only replica (with no guaranteed consistency of course ) of this entire cloud in my DataCenter B (DC-B), that would make my reads over DC-B faster without making my writes slow. To clarify, all the writes were going to go against DC-A only. The read-only cluster in DC-B could  also be made the master in case the entire DC-A went down.  The DC-B wouldn't be guaranteed to be in sync with the DB-A master but in my use case I could live with that. Seems like that is no possible out-of-the-box if I am using Solr 4.0+ in the cloud mode. It is either Solr Coud or cross data center read only replica. Can't do both at the same time.
I think that is what you confirmed as well. If I have it wrong, please let me know. Also, any thoughts on the most easy way to accomplish the read-only replica of the entire solr cloud cluster?

Thanks!
Tikoo

From: Sandeep Tikoo
Sent: Saturday, September 27, 2014 9:43 PM
To: 'solr-user@lucene.apache.org'
Subject: SolrCould read-only replicas

Hi-

I have been reading up on SolrCloud and it seems that it is not possible to have a cross-datacenter read-only slave anymore but wanted to ask here to be sure.
We currently have a pre Solr 4.0 installation with the master instance in our US mid-west datacenter. The datacenter in Europe has read-replicas which pull data using solr.ReplicationHandler. We wanted to upgrade to SolrCloud. As far as I have been able to figure out, with SolrCloud you cannot have a read-only replica anymore. A replica has to be able to become a leader and writes against all replicas for a shard have to succeed. Because of the a strong consistency model across replicas, it seems that replicas cannot be across datacenters anymore.

So my question is, how can we have a read-ony replica in a remote datacenter in Solr 4.0+ similar to pre Solr 4.0? Is it not possible anymore without doing it all yourself?

cheers,
Tikoo

Re: SolrCould read-only replicas

[Walter Underwood <wu...@wunderwood.org>]

Here is a different approach.

Set up independent Solr Cloud clusters in each data center. Send all updates into a persistent message queue (Amazon SQS, whatever) and have each cluster get updates from the queue.

The two clusters are both live and configured identically, so there is nothing to change in a failover.

wunder
Walter Underwood
wunder@wunderwood.org
http://observer.wunderwood.org/


On Oct 2, 2014, at 7:07 AM, Sandeep Tikoo <st...@digitalriver.com> wrote:

> Erick,
> 
> Thank you for your response. Yup, when I said it is not possible to have a cross continent data center replica, I meant that we never ever want to do that because of the latency.
> 
> What I was hoping is that  I could have Solr cloud in my DataCentre A (DC-A) and get all the benefits of sharding ( scaling/parallel computing) and failover redundancy within the same data center. If I could then have a read-only replica (with no guaranteed consistency of course ) of this entire cloud in my DataCenter B (DC-B), that would make my reads over DC-B faster without making my writes slow. To clarify, all the writes were going to go against DC-A only. The read-only cluster in DC-B could  also be made the master in case the entire DC-A went down.  The DC-B wouldn't be guaranteed to be in sync with the DB-A master but in my use case I could live with that. Seems like that is no possible out-of-the-box if I am using Solr 4.0+ in the cloud mode. It is either Solr Coud or cross data center read only replica. Can't do both at the same time.
> I think that is what you confirmed as well. If I have it wrong, please let me know. Also, any thoughts on the most easy way to accomplish the read-only replica of the entire solr cloud cluster?
> 
> Thanks!
> Tikoo
> 
> From: Sandeep Tikoo
> Sent: Saturday, September 27, 2014 9:43 PM
> To: 'solr-user@lucene.apache.org'
> Subject: SolrCould read-only replicas
> 
> Hi-
> 
> I have been reading up on SolrCloud and it seems that it is not possible to have a cross-datacenter read-only slave anymore but wanted to ask here to be sure.
> We currently have a pre Solr 4.0 installation with the master instance in our US mid-west datacenter. The datacenter in Europe has read-replicas which pull data using solr.ReplicationHandler. We wanted to upgrade to SolrCloud. As far as I have been able to figure out, with SolrCloud you cannot have a read-only replica anymore. A replica has to be able to become a leader and writes against all replicas for a shard have to succeed. Because of the a strong consistency model across replicas, it seems that replicas cannot be across datacenters anymore.
> 
> So my question is, how can we have a read-ony replica in a remote datacenter in Solr 4.0+ similar to pre Solr 4.0? Is it not possible anymore without doing it all yourself?
> 
> cheers,
> Tikoo
>