BTW, in response to previous questions... there are three kinds of AFS
volumes:
- read/write
- clone
- read-only
The first kind (read/write) is the kind you get with a "vos create".
They're the "normal" kind.
The second kind (clone) is basically a "copy-on-write" snapshot of a
read/write volume. "Backup" volumes are a specific kind of clone
volume; you get these with "vos backup". Temporary clones are also
sometimes created automatically (for example, at the start of a "vos
dump" or during "vos release") and removed when no longer needed. You
can also create them manually with "vos clone" but I recommend against
that unless you're absolutely certain you know what you're doing.
Clones must reside on the same server and same partition as the
read/write volume they were derived from. Clone volumes cannot be
modified.
The third kind (read-only) is for load balancing and redundancy; they
are created with "vos addsite" and updated with "vos release". You
can have as many read-only versions of a volume as you like, but only
one read-write version. A few subtle points here:
- Changes are propagated atomically from the read-write volume to
all read-only replicas upon the conclusion of "vos release"
- Clients will ALWAYS prefer read-only copies unless you
specifically indicate (as a flag to "fs mkm") on the mountpoint
that they should use the read-write volume.
- If a read-only replica of a volume exists, clients will NOT "fail
over" to the read-write volume in the event that all read-only
replicas are inaccessible. From the client's perspective the
read/write and read-only volumes have no relationship to each
other (though the server knows better).
These behaviors can be a bit confusing at first, but they start to
makes (some) sense after a while. The typical convention is that if
you make a mountpoint at "/afs/foo/bar" for a volume which has
read-only replicas, you should also make a force-read-write mountpoint
at "/afs/foo/.bar". AFS does this automatically for the synthetic
mountpoints directly beneath /afs/ (aka "dynroot"), but everywhere
else it must be done manually by an administrator.
To summarize:
read/write <==> "vos create"
clone <==> "vos backup", "vos dump -clone", "vos clone"
read-only <==> "vos addsite", "vos release"
So, basically, there is no "vos create user.foo.backup" in create-user
because that volume is a backup (clone) volume and it will get created
by "vos backup". We run "vos backupsys" (equivalent to "vos backup"
repeated for every volume) every night out of /etc/cron.d/ on deleuze.
It might make sense to immediately "vos backup" after creating a new
user just so that the /afs/hcoop/old/*/ mountpoints aren't dead links
(despite its name, "vos backup" is nearly instantaneous).
- a
--
PGP/GPG: 5C9F F366 C9CF 2145 E770 B1B8 EFB1 462D A146 C380
_______________________________________________
HCoop-SysAdmin mailing list
[email protected]
http://hcoop.net/cgi-bin/mailman/listinfo/hcoop-sysadmin
