Ian,
Thanks for your reply, I think from what you say I am indeed getting the
expected results. I read earlier about the autofs bug, I have 2 systems
(i386) running 2.6.18-53.1.14.el5, which still exhibit this behavior,
however my x64's are ok (same kernel). I think I also read that the
patch does not go live until RHEL 5.2. Since these only get booted after
a kernel update or once a year (just because I can!), the problem does
not really have any impact. The other odd behavior on the x64 machines,
maybe related is that for some time after install they could not cd
~user1. However this morning they can!
Thanks again,
Neil.
Ian Kent wrote:
On Tue, 2008-04-15 at 11:11 +0100, Neil Marjoram wrote:
OK I have just installed 5.1 on 4 machines, one of which is my file
server, running ldap, samba etc. The installs are from fresh, but I am
using the old configs for NFS and autofs from RHEL 4.
Make sure you update the kernel.
There was a known problem where a couple of autofs patches missed in the
initial release kernel. You can tell if you're seeing this problem as
the first attempt to automount a directory will return a fail but the
mount will happen.
All my servers use autofs to mount user directories from the one central
file server and all seems fine. Last night I completed the final upgrade
to the main file server, all is working just about fine, but I notice
that df does not show any of the filesystems mounted with autofs on the
file server, however cat /proc/mounts shows all the mounted directories
correctly. df -a also shows the directories mounted!
I can only imagine that this is a new feature so df does not show autofs
mounted from the local machine.
Am I missing something?
I'm not sure about the output of df for other than for the autofs
mounts, I would expect it to show the other mounts. The autofs file
system mounts aren't done using mount in version 5 so they shouldn't
normally show up in df output.
To quote myself responding to a similar question on the autofs list:
There are a few reasons.
The foremost reason is because version 5 uses lots of mounts as
triggers, for direct mounts and for the lazy mount/expire of trees of
mounts (aka the multi-mounts and the hosts map). If they were added to
the mtab this would lead to considerable confusion. And if you've used a
system that does add these to the mtab (such as old versions of AIX) and
your map had, even just a few hundred direct mounts, you would
understand how annoying it is. I guess we could use a pseudo mount
option, like "hide", so the entries aren't displayed (but that doesn't
really help with the locking issues below).
There is also the issue of mtab locking, under heavy mount activity,
leading to mtab corruption, which we've seen often in the past. Not
using the mtab for autofs mounts themselves substantially reduces the
pressure on mtab locking.
Finally there's the issue of scanning the mtab which I need to do fairly
often when checking if something is mounted. A few hundred entries isn't
that bad but the number of entries can order in the thousands for quite
a few users which leads to performance problems and aggravates the
locking issues.
Ian
_______________________________________________
rhelv5-list mailing list
rhelv5-list@redhat.com
https://www.redhat.com/mailman/listinfo/rhelv5-list
--
Neil Marjoram
Systems Manager
Adastral Park Campus
University College London
Ross Building
Adastral Park
Martlesham Heath
Ipswich - Suffolk
IP5 3RE
Tel: 01473 663711
Fax: 01473 635199
Reclaim Your Inbox!
http://www.mozilla.org/products/thunderbird
_______________________________________________
rhelv5-list mailing list
rhelv5-list@redhat.com
https://www.redhat.com/mailman/listinfo/rhelv5-list
