[3039] in SIPB-AFS-requests
Volumes with only one rep site
daemon@ATHENA.MIT.EDU (Greg Hudson)
Wed Jun 24 15:27:42 1998
Date: Wed, 24 Jun 1998 15:27:33 -0400
From: Greg Hudson <ghudson@MIT.EDU>
To: sipb-afsreq@MIT.EDU
A recently discovered AFS problem is that volumes with only one RO
site (on the same partition as the RW site, presumably) cannot be
safely released; clients may time out accessing the volume during a
release and not recover for a while (even after the release is
complete) unless an "fs flushv" is done on the volume. Following is a
list of the 4 volumes in the sipb cell with only one replication site.
Adding a rep site and releasing the volume will cause the kind of
outage described above, so this state is not trivial to recover from.
One option (which I haven't tested) might be to remove the RO site
from the VLDB, wait for clients to shift over to the RW site (give
them a week or so), zap the RO site, then add two RO sites and
release.
system.inbsd.srvd.80
system.inbsd.xfree32
system.linux.usrandrew
system.linux.usrathena
