[collectd] df plugin and statfs/statvfs

Martin B. Smith smithmb at ufl.edu
Fri Mar 11 23:14:40 CET 2011

Hi Florian,

On 03/11/2011 03:45 PM, Florian Forster wrote:
> Hi Martin,
> On Fri, Mar 11, 2011 at 01:33:36PM -0500, Martin B. Smith wrote:
>> >  Can anyone wager any other reason why this might not work? Any advice?
> the "df" plugin reads a list of mountpoints (using getmntent(3) when on
> Linux) and calls statvfs(2) only for returned directories. I assume that
> unless the filesystem is in fact mounted, possible targets of
> automounter are not returned by getmntent(3).

Ah, that'll do it. I must have missed getmntent() somewhere when 
scanning through df.c. Thank you for pointing that out :)

I'm actually explicitly specifying the 'automountable' directories, so I 
wasn't expecting the 'df' plugin to check if the mount was currently 
mounted -- I was hoping it would just call statvfs() on what I told it 
should be checked.

Would you think that a reasonable behavior when mount points are 
explicitly given in the config would be that they shouldn't be checked 
against getmntent? Or that statvfs() should be called before getmntent() 
and then again after? I'd be happy to provide a patch either way.

> If you use the file system and it is actually mounted, collectd should
> pick it up and report statistics. I don't know if this counts as "usage"
> of the file system, so it might be that it these file systems are not
> unmounted automatically when collectd is running [*].

Indeed, on automounted filesystems that go up and down, I get slices of 
time where the df plugin returns interesting data split between slices 
of time where it doesn't see the mount.

> Hope this helps,
> regards,
> —octo
> [*] Given that nobody has complained about this yet, I guess this is not
>      likely.

This leaves me in a weird situation. I have some NFS appliances that I 
cannot run collectd on, and I'd prefer to have the same style and format 
of output data on those as I do on regular Linux machines.

What do you think of my two options of calling statvfs() once before 
getmntent() or ignoring getmntent() when a mount point is explicitly 

Martin B. Smith
smithmb at ufl.edu - (352) 273-1374
CNS/Open Systems Group
University of Florida

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5497 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://mailman.verplant.org/pipermail/collectd/attachments/20110311/57bdac86/attachment.bin>

More information about the collectd mailing list