[diskless] pkg takes 100% of a CPU

[Rick Macklem]

BERTRAND Joël wrote:
[stuff snipped]
>        All filesystems are mounted from NetBSD server :
>
>root at pythagore:/var # mount
>192.168.10.128:/srv/pythagore on / (nfs, asynchronous)
>devfs on /dev (devfs, local, multilabel)
>procfs on /proc (procfs, local)
>fdescfs on /dev/fd (fdescfs)
>192.168.10.128:/home on /home (nfs, asynchronous)
Just fyi, "nfsstat -m" on the client will show you what NFS mount options
are actually in use.

>root at pythagore:/var # cat /etc/fstab
># Device        Mountpoint      FStype  Options Dump    Pass#
>192.168.10.128:/srv/pythagore / nfs nfsv3,tcp,soft,intr,rw,async,nolockd
I'd suggest you try without the "soft,intr" options. When those are set,
a system call like write(2) can return with EINTR and very few (if any)
applications handle that correctly.

[more stuff snipped]
>        On server, /var/log/messages are full of :
>Apr  8 13:40:49 legendre rpc.lockd: duplicate lock from hilbert.45141
>Apr  8 13:40:49 legendre rpc.lockd: no matching entry for hilbert
>Apr  8 13:40:52 legendre rpc.lockd: duplicate lock from pythagore.68734
>Apr  8 13:40:52 legendre rpc.lockd: duplicate lock from pythagore.68734
>Apr  8 13:40:52 legendre rpc.lockd: no matching entry for pythagore
>Apr  8 13:40:52 legendre rpc.lockd: no matching entry for pythagore
>Apr  8 13:40:55 legendre rpc.lockd: duplicate lock from pythagore.68734
>
>even if all filesystems are mounted with nolockd option.
If you use "nolockd" on all mounts, you do not need to run rpc.lockd or
rpc.statd, which is what I prefer. (The NLM and NSM protocols were undocumented
and fundamentally broken. Eventually they were published in an X/Open XNFS
manual, but it was a vague 2 page summary. The only real doc would be the
OpenSolaris implementation. NFSv4 does a much better job of locking, so if you need the locks to be visible to other clients, that is what I'd suggest.

rick