vg's 0, 1, and 2 are on different sets of spindles to give higher IOPS. The problem is that the 4 snapshots need to be created atomically or else I'm going to get my database and transaction logs out of sync as there could be a few ms between the snapshot of lv2 and lv3.
AFAIK you can't do that.
Thanks for the confirmation... that's pretty much what google told me but I thought maybe someone else might know something more.
In IIRC Scalix there was an option to say "get your shit in order and be quiescent until further notice, or fifteen seconds, whichever elapses first." Then you'd make your snapshots, then (since it was probably <<15s) say to scalix "OK, you can buffer writes again".
I have no idea how well that actually worked, since I (thankfully) never had to restore that box from backups.
This is where Windows is _really_ good. Pretty much all the backup infrastructure these days is built around VSS which allows you to take, at the very least, a crash-consistent copy of all volumes in the system. On top of that, application-specific VSS Writers ensure that their databases are shutdown-consistent before the snapshot is taken. From the snapshot you can back up files or blocks as you see fit.
Plan B, of course, would be to dump the database, then backup that.
I want to take a consistent snapshot of the whole machine and then bring it up in another VM for testing. A procedure more complex than 'take snapshot, start VM' is not the desired outcome. Thanks again James