On Sun, Feb 3, 2013 at 8:46 AM, Jimmy Xiang <[EMAIL PROTECTED]> wrote:
> The web UI came with Hadoop provides some useful information about each
> server: for example, jmx, metrics, stacks, log level, conf, and so on.
> I was wondering should we add a web UI to REST and Thrift too, as we do for
> master and region server.
Would be useful I'd say.
> REST is a little different. It already hasa web server. In working on
> HBASE-7738, I tried to add jmx/metrics servlet to the existing REST web
Did it work? Can you mount the servlets into a new context or as a new
> We have to add several knobs to avoid path mapping conflicts between the
> new servlets and the existing REST paths.
Mount the hadoop stuff on to '/this' or '/proc' or '/ui' or '/insight'
> I was wondering should we
> use a separate web server on a different port for example 8085. For Thrift,
> we can use 9095. That's default port, can be configured.
I suppose it makes sense having separate port for thrift (unless we want to
do some hackery) but would be interested in why we'd need to do similar
when we already have a web context going on the REST server?
> The benefit to have another web server in REST is that we can get all
> Hadoop http server features easily, and will never worry about path mapping
> conflicts. The cons is that it needs another port.
Could the context you mount the hadoop servlets on be something that makes
no sense as a REST query?