[OpenAFS] Re: Stumped: openAFS client "Connection Timed Out"

Andrew Perrin clists@perrin.socsci.unc.edu
Thu, 20 Feb 2003 10:40:27 -0500 (EST)


Thanks for everyone's responses. Based on information from UNC's
administrators and Todd DeSantis, I decided the simplest thing to do was
to leave the machine offline for over two hours to reset the server's
record. So I turned it off last night and back on this morning (which
ruined my uptime but otherwise is no problem).  It now works fine.

Looking back in my logs, I think what happened is this:
- My cable modem connection went down, breaking the AFS connection
- The server noted the unavailability and started trying to use UDP
packets to 7100 (is this right?) to check for availability
- Because of the NAT, UDP to 7100 from the server just gets swallowed,
since the NAT firewall isn't running an AFS client.
- The server is therefore unavailable.
- Being offline for two hours resets this behavior.

One thought, if this all makes sense, is that perhaps running an AFS
client on the firewall might "trick" the servers into responding, since
the client would then respond to the 7100 packets.

This experience brings up a bug/feature request, too: it would be nice if
a client could:
- Tell the server to delete the record without going offline for two
hours; or, at least,
- Stop the AFS client without rebooting the machine.

Thanks again to all who replied.

----------------------------------------------------------------------
Andrew J Perrin - http://www.unc.edu/~aperrin
Assistant Professor of Sociology, U of North Carolina, Chapel Hill
clists@perrin.socsci.unc.edu * andrew_perrin (at) unc.edu