Cluster aware clients
tsantero at gmail.com
Fri Aug 15 12:46:51 EDT 2014
My preference is for each client application node to connect to an instance
of haproxy on the localhost, that way cluster changes only requires
updating and reloading the haproxy config--which is cleaner and safer for
An example haxproxy config can be found on the Riak docs:
On Fri, Aug 15, 2014 at 11:32 AM, Daniel Simms <daniel.simms at gmail.com>
> < 2.0 the java cluster client was actively unhelpful (in that it didn't
> track node health, so if one node was down, the client would regularly try
> it still, IIRC), and a client-local haproxy (or such) is the way to go.
> >= 2.0 the java cluster client looks like it's smarter about tracking node
> health (but I'm not using it in anger and therefore don't know if it's
> otherwise suitable and would be curious if anyone is happily using it
> without haproxy)
> On Fri, Aug 15, 2014 at 6:18 AM, Paul Barry <paul.barry at temetra.com>
>> We’ve been using Riak with Java client for more than two years in
>> production, and best solution is ha-proxy.
>> Your Java client connects to one ha-proxy, configured for round-robin to
>> all the nodes. We run ha-proxy on our front-ends, so clients always connect
>> to localhost.
>> If you’re planning to add or remove a node, just reconfigure ha-proxy and
>> reload it (instant with no down-time).
>> ha-proxy also gives you much better control over how unexpected failures
>> are dealt with automatically.
>> On 15 Aug 2014, at 13:19, Simon Hartley <Simon.Hartley at williamhill.com>
>> So I’m writing a Java client app. I’m running a 5 node Riak cluster and I
>> want my client to know about all the nodes to avoid single points of
>> Is there a best practice for automatically updating clients when new
>> nodes are added to a cluster?
>> riak-users mailing list
>> riak-users at lists.basho.com
> riak-users mailing list
> riak-users at lists.basho.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the riak-users