Hey lists,
Not 100% sure this is NFS related, so I copied LKML.
A couple times a day I get an Apache process stuck in D state while
reading off of NFS. I have adjusted Apache's NFS specific settings to
work around any apache-over-nfs bugs. The directory which the locked-up
httpd process was serving from (/proc/$pid/cwd) isn't accessible VIA
bash either, so it looks to me like it's more an NFS client/server bug
rather than an Apache one. A reboot is required to free the locked
process AFAIK.
Server: 2.6.11.5, nfs-utils 1.0.7
Client: 2.6.11(.7,.11), nfs-utils 1.0.7
The server also has various clients from 2.4.23 to 2.6.11.11 which all
seem to behave fine (however their workloads are different). I've
noticed this happen with almost no load, to multiple occurrences with a
load of ~4.0 or so.
The clients (at least the 2.6 ones) are all connected to the same
network switch and mounted with the following:
nfsserver:/nfsdir/home /home nfs
rw,v3,rsize=32768,wsize=32768,hard,intr,tcp,lock,addr=nfsserver 0 0
Attached is a SysRq process trace.
Thanks,
Dale