Rich,

Peter Tribble wrote:
> > I've been using fsstat (glad to see it made it into update 3!) a little.
>
> >
> > I've stumbled across bug 6508225, which I see on a nevada box.
> > I'm seeing a related problem on S10U3 - instead of claiming there
> > aren't any statistics for /, most of them are just zero with SVM
> > root. This is particularly annoying as I don't believe in having
> > more partitions that I need, so if I can get away with just a
> > large root partition then I will.
>
> Are you looking at the kstats directly or is fsstat reporting
> mostly zero numbers for "/" (on an SVM device).
>
> If you're looking at the kstats directly, then these are "zombie"
> kstats due to 6508225.


That's what I see on snv_54:

% fsstat /
No statistics available for /

On S10U3 I see mostly zeros:

% fsstat /
 new  name   name  attr  attr lookup rddir  read read  write write
 file remov  chng   get   set    ops   ops   ops bytes   ops bytes
    0     0     0  362K     0      0     0     0     0     0     0 /

I get essentially the same out of jfsstat, which looks at the raw kstats.

> Another observation: I notice that the kstats are generated for
> > all the entries in a direct automount map, even though none
> > of the filesystems are actually mounted. Is this desirable?
> > (I've checked, and it doesn't seem to keep historical data -
> > the values seem to reset if a filesystem is mounted or
> > unmounted.)
>
> Hmm... Sounds like you found a bug where the kstat isn't cleaned
> up when the file system is unmounted.  Would you please submit
> a bug report on this?
>

I don't think that's it. The kstats are cleaned up when the real filesystem
is unmounted (at least, they're reset); the question is why there's a
kstat for every underlying autofs mount? And, because the dev id doesn't
change, it can get very confusing:

 - there's a kstat associated with an autofs mount that is just a
placeholder

 - when the real filesystem gets mounted the kstat has the same
name

 - when the real filesystem gets unmounted, there's still a kstat
with the same name but with zeroed data

It just seems inefficient. And my scheme of simply looking for new kstats
to appear to spot new filesystems being added doesn't work.

Ultimately, should autofs mounts have fsstat kstats?

-- 
-Peter Tribble
http://www.petertribble.co.uk/ - http://ptribble.blogspot.com/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
<http://mail.opensolaris.org/pipermail/observability-discuss/attachments/20070111/c0e0d692/attachment.html>

Reply via email to