On 9/9/14 5:57 PM, Sean Caron wrote:
Hey, just sharing some hard-won (believe me) professional experience.
I have seen xfs_repair take a bad situation and make it worse many
times. I don't know that a filesystem fuzzer or any other simulation
can ever provide true simulation of users absolutely pounding the tar
out of a system. There seems to be a real disconnect between what
developers are able to test and observe directly, and what happens in
the production environment in a very high-throughput environment.
Fair enough, but I don't want to let stand an assertion that you should
avoid xfs_repair at all (most) costs. It, like almost any software,
has some bugs, but they don't get fixed if they don't get well reported.
We do our best to improve it when we get useful reports from
users - usually including a metadata dump - and we beat on it as best
we can in the lab.
"pounding the tar out of a filesystem" should not, in general, require
an xfs_repair run. ;)
Yes, it's always good advice to do a dry run before committing to a
repair, in case something goes off the rails. But most times I've seen
things go very very badly was when the storage device under the filesystem
was no longer consistent, and the filesystem really had no pieces to