Juha Jäykkä wrote:
As soon as the client comes across a RW volume the path will be a RW
path. Therefore the clients will not switch. Are you sure your
clients are really using RO versions of the volume?
Yes, I understand this. This is why I wrote "Now users' $HOME is, of
course, the RW version (/afs/blablabla/.username) and programs using $HOME
will complain when A disappears."
What we exactly did was this. We simulated a crashed host by unplugging
the ethernet cable of A (housing RW and RO of everything) so only B
(housing another RO copies) and C (with no volumes) were available. After
that, all RW paths became inaccessible, as they should.
We came about implementing AFS for precisely the reason that losing users'
home dirs in case of a crashed file server is something we want to avoid.
Obviously we'll lose the ability to write to home dirs, but that is just
an inconveniency: users are able to get things done by reading the files
off the RO replica and temporarily writing their files to local discs.
Should this work? We tried accessing the RO replica with "ls -l
/afs/tfy.utu.fi/home/username", but got nothing back. This does, however,
work if we remove the replica from A and only have one RO replica at B. I
even tried fs flushmount and flushvolume before the ls.
You may also want to try "fs checkv" where you're doing the flushes
Kim
-Juha
_______________________________________________
OpenAFS-info mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-info
