maybe there is a better way to do this, so i'll lay out my plan first,
then my problem, and then my solution.
to avoid 100s or 1000s of nfs mounts i would like to have one mount per
exported file system and then bind mounts (or symlinks) for each directory
within each mount point. like so:
real mounts:
/disks/foo.0 -> foo:/0
/disks/foo.1 -> foo:/1
/disks/bar.0 -> bar:/0
etc.
and then bind mounts:
/home/user -> /disks/foo.0/user
/home/other -> /disks/foo.1/other
/home/another -> /disks/bar.0/another
now this kinda works, but since there is one global lock file for all
automount daemons the first reference to /home/user will try to touch
/disks/foo.0 and that will hang for 10 seconds waiting for the lock file.
once the real mounts are done, things work well. and since there are a
few thousand entries in /home it keeps the nfs mounts from getting out of
control.
so my solution would be to make one lock per daemon. that will keep each
daemon happy and they won't interfere with each other.
is that a reasonable idea? is there a better way to cope with this? i
know that long ago you used to be able to do things like:
user foo:/0:user
user1 foo:/0:user1
and somehow it would mount foo:/0 once and then use symlinks for user and
user1 (or something like that). but modern automounts just do a full
mount for each one.
part of this plan is based on old 2.4 behaviour where too many mounts
would kill the system. but it still doesn't feel right to have a few
thousand mounts on a box (other than bind mounts). am i just being
paranoid?
_______________________________________________
autofs mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/autofs
