[Lustre-discuss] MDS activity monitoring

[Andreas Davour]

On Monday, March 21, 2011 08:32:59 Andreas Dilger wrote:
> On 2011-03-20, at 9:09 PM, Andreas Davour wrote:
> > I've decided I want to keep an eye on the activity on my MDS, and poking
> > about in /proc I found both /proc/fs/lustre/mds/<fsname>-MDT0000/stats
> > and /proc/fs/lustre/mdt/MDS/mds/stats which both contain stats about
> > metaops, but while similar they are slightly different.
> > 
> > Now, I'm not really sure I see how they are different, and why there are
> > two /proc/fs/lustre/md* directories with stats like that?
> 
> I believe there is actaully a patch in bugzilla that consolidates these two
> /proc files, and removes one of them.  I think one is just left over for
> compatibility reasons.

That would be a good thing, since they are confusing. Any hint on where to find 
that bugzilla entry?

> > Can somebody who knows something about lustre internals maybe shed some
> > light on the layout of the /proc files and where to best look to get a
> > grip on what's going on in the file system?
> 
> There are two main different types of "stats" files - ones for the network
> RPC service (e.g. MDS/{mds,mds_readpage}, OSS/{ost,ost_io,ost_create}),
> and one for the target device (e.g. mds/<fsname>-MDT0000/,
> obdfilter/<fsname>-OSTnnnn).
> 
> Each have slightly different statistics, because they have access to
> different information.  I think some of this is documented in the manual.

Thanks. I'll take a closer look at the documentation.

/andreas
-- 
Systems Engineer
PDC Center for High Performance Computing
CSC School of Computer Science and Communication
KTH Royal Institute of Technology
SE-100 44 Stockholm, Sweden
Phone: 087906658
"A satellite, an earring, and a dust bunny are what made America great!"