Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
HBase, mail # dev - Metrics in 0.96


Copy link to this message
-
Re: Metrics in 0.96
Elliott Clark 2012-06-15, 17:04
Sounds great.
I also think we should start moving away from just averages, and move
towards using the histograms.  Right now the histograms are broken, they
don't get exported into jmx but I've got a fix I think (HBASE-6211).  I
would also like to look into tying Server/Region Load in with metrics.
 Right now they are very similar to
RegionServerMetrics/RegionServerStatistics and seems like there is a lot of
overlap.

On Fri, Jun 15, 2012 at 8:54 AM, Lars George <[EMAIL PROTECTED]> wrote:

> Hi Andy,
>
> Yes, that is exactly the plan. I am sure we have areas where we simply
> lack metrics, I have noted this down somewhere and will dig up the details.
> It would also consolidate (and close out) all open "Improve HBase Metrics"
> JIRAs.
>
> Lars
>
> On Jun 15, 2012, at 5:41 PM, Andrew Purtell wrote:
>
> > +1
> >
> > And at the same time IMHO it's worth asking if we have everything
> covered with a meaningful metric, and add if missing, as well as remove
> those not useful.
> >
> > Metrics2 is in Hadoop core 1 and 2. There was a backport into 1 at some
> point but the latest (1.0.3) has it for sure.
> >
> >    - Andy
> >
> > On Jun 15, 2012, at 8:07 AM, Stack <[EMAIL PROTECTED]> wrote:
> >
> >> On Fri, Jun 15, 2012 at 1:07 AM, Lars George <[EMAIL PROTECTED]>
> wrote:
> >>> Hi,
> >>>
> >>> I propose that we clean up the metrics for 0.96 as well (if that is
> not already underway. This includes switching to Hadoop Metrics v2 and
> adding all missing stats, as well as fixing those that are misleading or
> useless. For that reason I talked to our friends at Sematext during the
> Buzzwords HBase workshop, as well as to our internal HBase team. We all
> seem to agree and I'd be happy to setup some sort of call/task force to
> discuss what we need to work on etc. Obviously I would love for someone
> from the FB team to also partake as they have done heaps lately on that end
> - but I also understand if that is a timing issue.
> >>>
> >>> What do y'all think? My reasoning is that 0.96 might become 1.0, but I
> would love for the metrics to be stable by then.
> >>>
> >>
> >> +1 and +1 on you being Ring Master (tell us what to do!)
> >>
> >> Some metrics are near useless.
> >>
> >> Metrics2 is in hadoop 1.0.x?  0.96.0 will require 1.0.x so that should
> be fine.
> >> St.Ack
>
>