Dean> On Mon, 12 Nov 2007, John Stoffel wrote: >> Well, the general issue is open files. And files created after the >> backup process starts. >> >> For example, if you are doing an incremental backup of just the files >> that changed since the previous backup, you need to do two passes. >> Well, not quite, you can optimize it, but there's definitely a window >> where a file can be created (or deleted!) between the initial scan of >> the directory and the actual reading of the file to be sent to tape, >> disk, whatever.
Dean> This is why dump manpage says the filesystem should be quiecent. Dump/restore was all written way back in the mists of time when people didn't mind that you shutdown the system to single user mode for backups and such. Thankfully, it's long gone. :] Dean> Unmounted and clean is best, but not required. Shutdown the Dean> database before dump; start it after dump. Other user activity Dean> is usually just hit or miss, and the users don't mind just too Dean> much (well, don't tell them that when they've just lost their Dean> important file, unless behind 3 inches of Dean> plexiglass--bofh). This is why people want snapshots in Dean> filesystems... Snapshots are awesome, I love my NetApps for this. Linux filesystems and volume manager still have a way to go in this regard unfortunately. Dean> Fix ext3, ffs, and ufs to have even just one snapshot (like Dean> afs), and they'd be a hero loved by all... But its probably Dean> easier to rig the lottery... ext4 might be able to do this, I doubt ext3 could. And you really do want to make sure you handle all the corner cases gracefully. John _______________________________________________ bblisa mailing list [email protected] http://www.bblisa.org/mailman/listinfo/bblisa
