Does someone know what all needs to happen to allow a real [non-Mini] cluster to run on fully random ports? Currently it looks like:
ThriftBlurControllerServer - binding needs fixed when multiple instances are running ThriftBlurShardServer - binding needs fixed when multiple instances are running BlurClient - needs a way to accept a ZK quorum as a connection string and discover controllers from ZK.
The controllers already rely on ZK for the shard connections so that should "just work" I reckon.
I don't think the blur gui ports are currently stored in ZK, so one would either add them to ZK or poll each shard/controller for the port its running its gui on. On Tue, Jul 15, 2014 at 9:05 AM, Tim Williams <[EMAIL PROTECTED]> wrote:
On Tue, Jul 15, 2014 at 9:11 AM, Garrett Barton <[EMAIL PROTECTED]> wrote: My vote is to look them up via thrift (and cache as needed). The config value is "blur.http.status.running.port" and can be retrieved by calling getConfiguration() on the given process. We could likely build a nice little java object that would cache things as needed and make the results available via a thrift admin call.