"That which is overdesigned, too highly specific, anticipates outcome; the anticipation of outcome guarantees, if not failure, the absence of grace."
-- William Gibson, All Tomorrow's Parties
IPMI hosing your NFS client.

Our build files live on a Solaris 10 NFS server. The build client lives in a zone on a separate host. The build files are exported via v3 and tcp to the client.

Periodically the client would hang and require a zone reboot. Needless to say, this was astoundingly annoying if you didn't realize it had hung until you had started your build or push processes. An init 6 always fixed it... for a while.

Looking at snoop on the NFS server, it looked like a bunch of tcp:664 packets came in and go... nowhere. They hit the interface and vanish. Gee, I thought. That's odd.

Finally I got sick of this, and Googled around and found some references to port 623, a Linux bug that sounded pretty similar, and other Solaris users experiencing the same problem.

The first post is really the most useful. Different port, but same behavior.

After creating the rmcp dummy service in inetd, and restarting the zone, the problem has not resurfaced.

It's pretty interesting that this particular bug manifests because a chip on the motherboard eats traffic silently. "Interesting", anyway.

August 29, 2009 6:05 AM