> I am looking for general advice on replicated volumes. Any suggestions
> would be appreciated.
>
> In particular, I am trying to answer these questions:
>
> 1) I want to replicate the volume mounted at /afs/mycell/usr/local/sws.
> What non-obvious things do I need to consider? Do other volumes
> mounted above this volume need to be replicated for this to work?
That's about it. The cache manager will only look for readonly volumes
if you're still on a "read-only" path - that is, if all volumes traversed
from /afs on down have been read-only volumes. This means that once you
go through a "read/write" mount point (e.g., a mount point for "%root.afs"
instead of "#root.afs"), you always get read/write volumes.
If you want to actually be able to _change_ the contents of a replicated
volume, you'll have to have such a "read/write" mount point. A common
convention seems to be to create "/afs/.cell-name" as a read-write
mount point for root.cell. You still need a way to get at a read-write
version of root.afs as well; there doesn't seem to be a common convention
for what to call that mount point (we call it /afs/.WRITE).
Also, note that if you want more than one read-only copy of a volume
(you almost always do), then they must be on different fileservers.
You can't, for example, have two RO clones on different partitions on
the same server.
> 2) This volume's contents change occasionally. Is there any way to
> automate the release of changes to its ReadOnly clones? Do I really
> have to do "vos release" after editing any file?
Yes. Anytime you make a change to the volume, you have to do "vos release"
before any changes will be visible. Also, some clients may not see the
change for up to an hour, unless someone on that client does "fs checkv".
Many cells take advantage of this behaviour to allow coordinated releases
of lots of files (for example, an entire software package). This means
you can be installing a new version without affecting the old until the
last minute. Beware, though, that if you replace a binary and then
release that volume, copies of that program running on a client will
almost certainly crash.
> Why is mounting of this volume different than mounting of root.cell,
> which has different mount points for each type of root.cell volume?
No; this is exactly the same concept. It's worth noting, however, that
there's another way to get onto a read-only path besides simply traversing
RO volumes from /afs. A "cellular" mount point (one that points to a
volume in another cell) will always get you onto a read-only path in
that cell (provided, of course, that that volume is replicated).
-- Jeffrey T. Hutzelman (N3NHS) <[EMAIL PROTECTED]>
Systems Programmer, CMU SCS Research Facility
Please send requests and problem reports to [EMAIL PROTECTED]
