hadoop-yarn-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Alejandro Abdelnur <t...@cloudera.com>
Subject Re: job history server UI slowness
Date Mon, 01 Apr 2013 20:25:11 GMT
IMO, a service should never use 0.0.0.0 to resolve the host for a URL for a
 client. And even more if the host in question is of another service. It
seems we should introduce different properties to specify a service bind
address and a service IP address.

Thx


On Wed, Mar 27, 2013 at 7:30 PM, Sandy Ryza <sandy.ryza@cloudera.com> wrote:

> I get the slowness even when not running jobs and no jobs have run.  I dug
> deeper into it and it looks like all the slowness comes from trying to
> resolve the RM address of 0.0.0.0 to a hostname for a link address on the
> page.  I'm not sure why the dns lookup is taking so long for me, but
> there's no reason we need to do it on every page load so I filed
> https://issues.apache.org/jira/browse/MAPREDUCE-5111.
>
> -Sandy
>
> On Wed, Mar 27, 2013 at 7:17 AM, Robert Evans <evans@yahoo-inc.com> wrote:
>
> > I don't think anyone has dug into it.  Are you running jobs at the same
> > time that you are looking at the history server? The history server is
> > configured by default to only hold 5 jobs in memory at any point in time.
> > If you are switching in between jobs there is the possibility that you
> are
> > reading the complete history file each time.  Also if you are running
> jobs
> > in the background the job client may be causing the history server to
> load
> > the job data so it might be pushing jobs out of the cache while you are
> > still looking at them on the UI. I don't think any of those are likely,
> > but they are something to be aware of. I would suggest that you try to
> dig
> > into it some and profile it to see what is going on.
> >
> > --Bobby
> >
> > On 3/26/13 10:41 PM, "Sandy Ryza" <sandy.ryza@cloudera.com> wrote:
> >
> > >When I run the job history server on my local machine, pages take in the
> > >10s of seconds to load, where there MR1 equivalents loaded
> > >instantaneously.
> > > Has anyone else experienced this?  Does anybody know what could be
> > >causing
> > >it?
> > >
> > >thanks for any guidance!
> > >Sandy
> >
> >
>



-- 
Alejandro

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message