[Pvfs2-users] pvfs2 stability

[Rob Ross]

There's no bit anywhere where we get to say "we're a network file 
system". They may skip file systems with the fstype "nfs" or "afs" or 
something along those lines...

Rob

Mark Bartelt wrote:
>>>  A slightly better solution is to update the PRUNEFS setting
>>>  if one exists on your system. That way you can add pvfs2 to
>>>  the list of filesystem types to always exclude, rather than
>>>  relying on the mount path to stay the same.
> 
> Yep, I agree; this is better (I'd say "much better", instead of
> "slightly better", in fact).  The reason I didn't suggest it in
> my previous message is that back when we first got slammed by a
> bunch of systems all firing off cron jobs which did a recursive
> descent into our PVFS hierarchy, I wanted to confirm that if we
> specified a PVFS filesystem type, it really would prune things
> correctly.  I actually tried it with "find" rather than trying
> it with "updatedb"; but the result should be the same.
> 
> So I noticed that the "-fstype ... -prune" wasn't preventing it
> from walking through the PVFS filesystems.  This morning, after
> reading your message, I realized what I'd probably done:  When
> I tried it before, I used "-fstype pvfs", not "-fstype pvfs2".
> D'oh!  I just tried it with the correct "-fstype" argument and
> all is well.
> 
> So yeah, that's definitely the way to go.  One question, though:
> In SuSE's /etc/sysconfig/locate (which defines things which the
> /etc/cron.daily/updatedb uses) there's a comment which says ...
> 
> # uptdatedb normally only scans local harddisks, but can include net paths
> # in the database as well. If you specify directories here, they will be
> # scanned.
> 
> ... which implies that, unless UPDATEDB_NETPATHS gets defined to
> include non-local filesystems, "updatedb" _should_ restrict the
> search to local disks.
> 
> Without digging into the "updatedb" source code, I have no idea
> how it decides about "local filesystem" vs "network filesystem".
> But it clearly gets it wrong with regard to PVFS, because if it
> correctly identified PVFS as non-local, it wouldn't have started
> looking through those filesystems.
> 
> So ...  Rob (et al.), do you know the answer to this offhand, and
> is it a problem with PVFS2, or some kernel issue (or a combination
> of the two, or something totally different) which causes "updatedb"
> to think the PVFS filesystems are local?  (I suppose it could just
> be something as simple as "updatedb" having a table of filesystem
> types it knows about as being "network" filesystems, without PVFS
> being included.)
> _______________________________________________
> Pvfs2-users mailing list
> Pvfs2-users at beowulf-underground.org
> http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
>