View Issue Details
|ID||Project||Category||View Status||Date Submitted||Last Update|
|0004984||CentOS-6||autofs||public||2011-07-18 20:37||2011-10-17 21:33|
|Target Version||Fixed in Version|
|Summary||0004984: trying to access lots of automount points fails|
|Description||trying to access a lot of automount points in rapid succession (like might happen for bulk email delivery) causes a failure where automount just stops mounting anything for some undetermined amount of time. this is usually around 100-200 mounts. no log messages are generated.|
|Additional Information||this works fine with 5.6 and there are no obvious config differences that i've found. it isn't any kind of mountpoint count because if i wait a bit, then i can get some more mounted until it hits the problem again. i have run tests to limit it to one server, changed the sysconfig/autofs file to use nfs3 by default (not running 4 yet), all without any effect. i haven't been able to identify anything upstream yet that could address this, but maybe there is something pending from 6.1?|
|Tags||No tags attached.|
|i guess i should clarify, that when this issue arises, it doesn't hang, it just stops making the mounts and so access fails with a 'no such file or directory' type error.|
|i have just updated to the autofs from the cr repo. no change to this bug yet. maybe there is a kernel update pending.|
just tested with new kernel and it still fails.
is there someone watching these with access to an official rhel box to test?
i have 1800+ entries in my automount map for /home. i test by doing something like:
ypcat -k auto.home | cut -d' ' -f1 | while read x; do df /home/"$x"; done
after about 50 directories, it starts failing the mount. wait a few minutes and it will clear the issue and get through another batch.
|did some more testing after latest kernel. bug still present. but i'm pretty sure that the actual issue is with the mount() system call for nfs. it looks like perhaps it is running out of low number ports to use for mounting. c5 doesn't show this problem, so perhaps some values have been changed in the kernel as to how many ports are available for nfs mounts, or something of that sort. i'll continue digging and updating this bug, unless a new bug should be opened based on my theory that it is not automount itself that is the problem.|
|aha! the bug is because nfs v4 is now tried first on mount. and v4 doesn't use a separate mount interface, it all goes over the nfs rpc and it uses tcp and it wants a privileged port. so it quickly burns up all the privileged ports and then things go downhill quickly. since i don't have v4 capable servers, i am not sure yet if this is a problem with a fully function v4 network. as a workaround i have added nfsvers=3 into my auto.master file and that seems to have fixed things.|