[OpenAFS] Re: Disk cache not updating

Charles W Ball cball@bu.edu
Fri, 14 Jan 2005 15:20:03 -0500 (EST)


We had a similar symptom occur for our apache docroot this week. This
occured as file servers providing web content were upgraded from 1.2.9 to
1.2.13. In this case, I'm prepared to attribute the problem to loss of
callbacks across the fileserver restart/upgrade, but it is quite unusual
for clients to lose state in this way.

Client (webservers) which showed evidence of the problem were:

AFS version:  OpenAFS 1.2.9 built  2003-05-12	(os: solaris 7 & 8)
AFS version: Base configuration afs3.6 2.36	(solaris 8, test system) 

This incident affected both memcache and diskcache clients. Use of fs
flushvol resolved the affected volume. The critical (to us) volume was
modified and released 20 minutes after it's fileserver was restarted; most
clients were fine, but several required a flushvol more than 3 hours after
the release (3hr 20min after restart).  This seems like a long time, but
unfortunately the immediate symptom was addressed before I could dump the
cm stats.

Is there a procedure that can be followed either before or after a
fileserver restart to detect any existing, defunct callbacks and refresh
them before volumes are modified and released?

Charles Ball
Information Technology
Boston University