On Tue, 2015-12-29 at 19:06 +0100, David Sterba wrote:
> > Both open of course many questions (how to deal with crashes,
> > etc.)...
> > maybe having a look at how mdadm handles similar problems could be
> > worth.
> 
> The crash consistency should remain, other than that we'd have to
> enhance the balance filters to process only the unconverted chunks to
> continue.

What about nodatacow'ed files? I'd expect that in case of a crash
during reshaping, these files are (likely) garbage then right?
Not particularly desirable...


But probably that just goes in the direction of the issues/questions I
brought up in the other thread where I've asked the devs for
possibilities in terms of checksumming on nodatacowed areas:

i.e. stability/integrity of such files


For me, speaking with the sysadmin's hat, that was always the main
reason not to do reshapes so far, especially when data is quite
precious, which at least one typical use case for nodatacow is, namely
DBs.

So having crash resistance for CoW + nodataCoW during RAID reshape
should be desired.

Time for a journal in btrfs? O;-)


Cheers,
Chris.