solrcloud without faceting, i.e. for failover only

[Will Milspec <wi...@gmail.com>]

Hi all,

We have a smallish index that performs well for searches and are
considering using solrcloud --but just for high availability/redundancy,
i.e. without any sharding.

The indexes would be replicated, but not distributed.

I know that "there are no stupid questions..Only stupid people"...but here
goes:

-is solrcloud w/o sharding done?( I.e. "it's just not done!!" )
-any downside (i.e. aside from the lack of horizontal scalability )

will

Re: solrcloud without faceting, i.e. for failover only

[Chris Hostetter <ho...@fucit.org>]

: #1 is a trade off against being possibly more available to writes in the case
: of a single down node. In the cloud case, you're still open for business. In
: the classical replication case, you're no longer available for writes if the
: downed node is the master.

or to put it another way: classic replication lets you use N nodes for 
high availability reads, but you have a single point of failure for 
writes.

solr cloud gives you high availability for reads and writes -- including 
NRT support -- at the expense of more network overhead when writes happen.

: > -is solrcloud w/o sharding done?( I.e. "it's just not done!!" )
: > -any downside (i.e. aside from the lack of horizontal scalability )

it is certainly done -- specifically it is a matter of creating a 
collection with numShards=1 and replicationFactor=N.



-Hoss
http://www.lucidworks.com/

Re: solrcloud without faceting, i.e. for failover only

[Michael Della Bitta <mi...@appinions.com>]

The downsides that come to mind:

1. Every write gets amplified by the number of nodes in the cloud. 1000 
write requests end up creating 1000*N HTTP calls as the leader forwards 
those writes individually to all of the followers in the cloud. Contrast 
that with classical replication where only changed index segments get 
replicated asynchronously.

2. Slightly more complicated infrastructure in terms of having to run a 
zookeeper cluster.

#1 is a trade off against being possibly more available to writes in the 
case of a single down node. In the cloud case, you're still open for 
business. In the classical replication case, you're no longer available 
for writes if the downed node is the master.

My two cents.

On 1/6/15 16:30, Will Milspec wrote:
> Hi all,
>
> We have a smallish index that performs well for searches and are
> considering using solrcloud --but just for high availability/redundancy,
> i.e. without any sharding.
>
> The indexes would be replicated, but not distributed.
>
> I know that "there are no stupid questions..Only stupid people"...but here
> goes:
>
> -is solrcloud w/o sharding done?( I.e. "it's just not done!!" )
> -any downside (i.e. aside from the lack of horizontal scalability )
>
> will
>