[OpenAFS] ADS communications issue?
Eric Chris Garrison
Tue, 08 Sep 2009 16:06:06 -0400
We're doing a cutover from MIT Kerberos to ADS, and while it works most
everywhere else, one of our supercomputers' nodes is giving us problems.
AFS works fine with MIT Kerberos.
AFS "hangs" with ADS. That is, it'll try to reach the /afs/iu.edu
directory for 10-15 minutes, then time out, saying:
ls: /afs/iu.edu: No such file or directory
Interestingly, if I do a "bos status -noauth" from the client it works
fine. If I do it with a proper (ADS) token, it hangs and times out with
the following error:
bos: failed to contact host's bosserver (communications failure (-1)).
Watching a tcpdump of the session from the server's point of view, it
appears that the client never responds to rx requests:
bos: rx data bos call enumerate-instance 0 (36)
15:51:54.485660 IP RFSHOST.afs3-bos > CLIENT.53057: rx challenge (44)
15:51:54.586685 IP RFSHOST.afs3-bos > CLIENT.53057: rx ack first 1 serial
0 reason delay acked 16777216 (66)
15:51:56.486625 IP RFSHOST.afs3-bos > CLIENT.53057: rx challenge (44)
15:51:58.487591 IP RFSHOST.afs3-bos > CLIENT.53057: rx ack first 1 serial
0 reason ping acked 16777216 (66)
15:51:58.487610 IP RFSHOST.afs3-bos > CLIENT.53057: rx challenge (44)
15:51:58.488850 IP CLIENT.53057 > RFSHOST.afs3-bos: rx ack first 1 serial
4 reason ping response (65)
...and so on.
Again, this works fine with the MIT Kerberos realm, so it does not seem to
be a firewall or router filter type issue, and we're able to get Kerberos
tickets and tokens just fine with either realm. It just has
communications issues when we have tokens gotten with ADS.
Any ideas what would cause this behavior? I had them update to a recent
(1.4.11) openafs client, just in case our current dual realm status was
the problem, but that does not seem to make a difference.
Eric Chris Garrison | Principal Mass Storage Specialist
email@example.com | Indiana University - Research Storage
W: 317-278-1207 M: 317-250-8649 | Jabber IM: firstname.lastname@example.org