Meanwhile, I was able to roughly estimate which table is getting traffic by
executing the following commands:

1) Store ngrep output in a file (for few seconds)
ngrep -W byline port 60020 > temp.out

2) Find out all the tables that region server has from HBase user interface.
For each table execute the following commands:
grep 'TableName,' temp.out | wc -l

This was enough for us as even which table was getting hit would be very
useful information for us. I am guessing there should be a way to grep
region name too.

Regards,
Vaibhav,
GumGum



On Wed, Nov 17, 2010 at 9:22 AM, Jean-Daniel Cryans <[email protected]>wrote:

> AFAIK most monitoring systems don't like dynamically-named metrics,
> for example in ganglia you would end up with an ever growing number of
> metrics for req/regions (one for each region that the region server
> ever had). At the very least it should be included in the region
> server report so that the master can take action and plan accordingly,
> the new master has better facilities for that.
>
> J-D
>
> On Wed, Nov 17, 2010 at 8:15 AM, Lars George <[email protected]>
> wrote:
> > JD,
> >
> > Should we create a metric for it so that it dynamically counts per
> > region its usage? That can then be exposed via Ganglia context or JMX.
> > Just wondering.
> >
> > Lars
> >
> > On Wed, Nov 17, 2010 at 5:04 PM, Vaibhav Puranik <[email protected]>
> wrote:
> >> hi,
> >>
> >> Thanks for the suggestions JD & Michael.
> >> The region servers serving ROOT & META regions are fine.
> >>
> >> I will try analysing tcpdump output.
> >>
> >> Regards,
> >> Vaibhav
> >> GumGum
> >>
> >>
> >>
> >> On Tue, Nov 16, 2010 at 7:15 AM, Michael Segel <
> [email protected]>wrote:
> >>
> >>>
> >>> Beyond this... which region is serving your ROOT and meta data?
> >>>
> >>> That node will probably get a higher load.
> >>> Also, how many disks do you have and how many nodes?
> >>> You could see higher CPU loads if you're I/O bound.
> >>>
> >>> > Date: Mon, 15 Nov 2010 18:24:31 -0800
> >>> > Subject: Re: Correlating traffic with regions
> >>> > From: [email protected]
> >>> > To: [email protected]
> >>> >
> >>> > Yeah this is one area where HBase could do a much better job...
> >>> > because there's not really a way to do it within the database. One
> >>> > thing you can do is to tcpdump a few seconds of traffic on that node
> >>> > and decipher which tables (shown in the region name) are being used.
> >>> >
> >>> > J-D
> >>> >
> >>> > On Mon, Nov 15, 2010 at 5:17 PM, Vaibhav Puranik <[email protected]
> >
> >>> wrote:
> >>> > > Hi all,
> >>> > >
> >>> > > We are running 0.20.6 in production.
> >>> > >
> >>> > > On one of our nodes, we are seeing CPU (all 8 CPUS) hovering near
> 60%.
> >>> But
> >>> > > the node has many tables and many regions on it.
> >>> > >
> >>> > > Is there an easy way to find out which of these regions or tables
> are
> >>> > > getting most of the traffic?
> >>> > >
> >>> > > Regards,
> >>> > > Vaibhav Purnaik
> >>> > > GumGum
> >>> > >
> >>>
> >>>
> >>
> >
>

Reply via email to