On Sunday 16 November 2008 13:51:23 Mick wrote: > > Yes. Unix does some RealSmartThings(tm) when using files. The name is > > just a pointer to the actual file, represented by an inode. Once you have > > an inode open, it stays open until everything using it closes it. So you > > can add/delete/copy/move files by name with impunity as you then just > > move names around. Contrast this with other inferior systems, like say > > Windows for example, which has a built-in self-destruct button when you > > try this... > > Sure, but isn't there a problem with atime mtime metadata when you carry > out a backup in real time and then restore from it?
With a restore, you have really just two options: - consider the file being restored to be a new file and set the *time to now - consider it a full restore and set them the same as what's in the backup root or the file's owner is permitted to do the latter There's a third option which makes little sense: set the *time of the restored file to be the same as whatever file it is repalcing on disk. But this is mostly silly as the file data is now inconsistent with the recorded times > > > I was gravitating towards using LVM snapshot and then tar'ing that to > > > an external USB drive. > > > > This is the preferred way, as you get a consistent snapshot frozen at a > > point in time. This deals nicely with inconsistencies caused by files > > changing while you are backing up other ones. > > Right, that's what I was thinking too. What does restoring from a backed > up snapshot involve? The backup is just a backup, the fact that it was made from a frozen disk snapshot is irrelevant. So you would restore it in the usual manner for the backup format/method in use -- alan dot mckinnon at gmail dot com
