 | | | NFS client VM bug? | NFS client VM bug? 2004-11-04 - By Garrick Staples
Back
I 've seen a strange behaviour a few times and I 'm wondering if anyone else has
noticed this...
I have some overzealous users that like to chew up all available VM.
Unfortunately, any time the machine, however breifly, is out of VM, I 'll get
some processes stuck in IO wait forever. Invariably, these processes are
accessing something on NFS. Furthermore, any future access of those files also
get stuck in IO wait, forever.
It seems like any NFS request that fails to allocate memory won 't successfully
recover, and the files or directories associated with that NFS filehandle are
permamently locked.
If it 's just a file that is locked, I can unlink it and make a new one. But if
a directory is stuck, all I can do is rename it. Sometimes only part of a
directory is locked where I can access the oldest files, but not the newer
ones; so I 'm convinced it has something to do with reclaiming pages used by NFS
requests. After unlinking or renaming, I can make new ones and let the users
continue on with their work, but we eventually leak more and more memory this
way.
Also, I haven 't seen this happen on another machine since I booted it with the
hugemem kernel, but I 've not actually tried to produce the problem either.
This is using 2.4.21-20.ELsmp on a quad Xeon with 3G of physical ram and 2G
swap. I don 't see anything on bugzilla. Anyone else see this happening? Any
thoughts?
--
Garrick Staples, Linux/HPCC Administrator
University of Southern California
--
Taroon-list mailing list
Taroon-list@(protected)
http://www.redhat.com/mailman/listinfo/taroon-lis
t
|
|
|