change feed without single point of failure

[Timothy Riley <ri...@gmail.com>]

We are running couchdb version 1.6.1 in a 3 node environment and are
replicating changes to an indexing engine (e.g. Elastic Search).

How can we use the couchdb change feed to replicate changes to our indexing
engine without relying on a single couchdb node (and having a single point
of failure)? Our concern is we will miss or lose changes if we switch
between couchdb nodes in the event of a node failure (e.g. hardware or
network failure).

Will the clustering capabilities in couchdb 2.0 solve this problem and
produce a consistent change feed for all nodes in a cluster or is this
achievable with version 1.6.1 and beyond?

Sincerely
Tim

Re: change feed without single point of failure

[Jan Lehnardt <ja...@apache.org>]

Heya Tim,

> On 07 Apr 2015, at 20:59, Timothy Riley <ri...@gmail.com> wrote:
> 
> We are running couchdb version 1.6.1 in a 3 node environment and are
> replicating changes to an indexing engine (e.g. Elastic Search).
> 
> How can we use the couchdb change feed to replicate changes to our indexing
> engine without relying on a single couchdb node (and having a single point
> of failure)? Our concern is we will miss or lose changes if we switch
> between couchdb nodes in the event of a node failure (e.g. hardware or
> network failure).

you *could* build a changes feed proxy that reads from all nodes and sends
the feed along to ES and then run multiple instances of that so you can do
hot failover. Or wait for 2.0.


> Will the clustering capabilities in couchdb 2.0 solve this problem and
> produce a consistent change feed for all nodes in a cluster or is this
> achievable with version 1.6.1 and beyond?

2.0 will give you a cluster-wide _changes feed that would solve your problem.

Best
Jan
-- 
Professional Support for Apache CouchDB:
http://www.neighbourhood.ie/couchdb-support/