Anno domini 2019 Tue, 12 Nov 09:05:30 +0100
Adam Borowski scripsit:
> On Mon, Nov 11, 2019 at 08:06:26PM -0500, Steve Litt wrote:
> > > I really don't think data-loss is an acceptable compromise just to
> > > reduce wear.
> >
> > Lack of a journal doesn't necessarily mean data loss. It just means
> > that you might need to run fsck.ext4 on the drive after unmounting.
> > Occasionally you do lose data: I think it has to do with crash
> > shutdowns twice in a row without an intervening fsck.ext4.
>
> No: it's not just "might need to run fsck". It's data corruption of
> unknown parts of the filesystem. Not on a pair of crashes: a single
> crash is enough. If the disk was quiescent the damage might be negligible
> or non-existant, but in the normal case, I'd be looking at recovery of
> newest writes and restore the rest from backups.

I just wonder how we surved in the last millenia with ext2.

What you describe was the "default" behaviour of reiserfs. It had the additional feature of truncating all files that were opened for writing to size=0 in case of a crash - no matter if an actual write operation was pending when the crash accured. Don't know if that was ever fixed.

Nik


>
>
> Meow!



--
Please do not email me anything that you are not comfortable also sharing with the NSA, CIA ...