[OpenAFS] Callback/Cache Issues with 1.3.82 on FC3

Jason McCormick jasonmc@cert.org
Mon, 02 May 2005 17:32:54 -0400


  This is just a preliminary heads-up to a problem we're seeing.  I
deployed 1.3.82 on Friday to Fedora 3 hosts and have almost immediately run
into a strange volume consistency problems across hosts as well as
consistency between RW volumes and RO volumes.  The two problems are:

1) Contents differ between a RW volume and it's RO replica on a .82 host,
even after a 'vol release -f' and attempts to flush the volume.  Looking at
the volumes on a .81 or an AS3 box running 1.2.13 shows the volumes
consistent amongst the RW and RO volumes.  A reboot was required to get
consistency.

2) An FC3 host running .82 box had stale contents of a volume and after a
vol release that changed 2 files, the .82 box wiped out intermediate
changes to those same files from an FC3 host running .81.  The FC3 box w/
.81 immediately saw that the files where changed out from under it.  After
re-changing the files (on the .81 box) to include the wiped-out data, the
FC3 box running .82 still couldn't see those changes after a release.  The
AS3 box again say the changes immediately.

After I get this situation stabilized, I'll try and create a test-case for
this situation.

-- 
Jason McCormick <jasonmc@cert.org>
CERT Infrastructure Team