In the project I'm working on, snapshots as backups are more critical then the "live" data. What I'd like is the original volume to get write errors if I run out of snapshot room (with the corresponding printk causing my log-watcher to lvextend the snapshot and re-activate the original) Obviously, this would be a panic situation. But the reasons for this are many: If I'm doing something really dangerous on my database, I'd shutdown or quiesce it, snapshot, and activate it again. Do the dangerous merge, and drop the snapshot if everything worked out. If I miscalculate how much disk space would be needed, what could happen is I screwup the master, AND the snapshot breaks. I'd much rather the master was stuck in disk-wait waiting for a free COW block. I have not yet looked into LVM2/device-mapper, so if it handles this situation I'll just upgrade to that instead. P.S.: I know you can always allocate as much snapshot space as the original, but when you're talking about a 250gig filesystem it's not always possible to have a spare quarter terr on the same machine. --Dan
Attachment:
pgp00538.pgp
Description: PGP signature
