Re: [RFC] Checksum of the parity

[Chris Murphy <lists@colorremedies.com>]

On Mon, Aug 14, 2017 at 2:18 PM, Goffredo Baroncelli <kreijack@inwind.it> wrote:

>> Anyway, I do wish I read the code better, so I knew exactly where, if
>> at all, the RMW code was happening on disk rather than just in memory.
>> There very clearly is RMW in memory code as a performanc optimizer,
>> before a stripe gets written out it's possible to RMW it to add in
>> more changes or new files, that way raid56 isn't dog slow CoW'ing
>> literally a handful of 16KiB leaves each time, that then translate
>> into a minimum of 384K of writes.
>
> In case of a fully stripe write, there is no RMW cycle, so no "write hole".

That is conflating disk writes and in-memory RMW. They have to be
separated. For sure there's in-memory RMW + CoW of the entire stripe
to disk, for a tiny (1 byte) change to a file because I've seen it.
What I don't know, and can't tell from the code, is if there's ever
such a thing as partial stripe RMW (and over write of just a data
strip and a corresponding over write for parity). Any overwrite is
where the write hole comes into play.

But inferring from the work Liu is apparently working on for a
journal, it must be true that there is such a thing as a overwrites
with Btrfs raid56.



>
> Just of curiosity, what is "minimum of 384k" ? In a 3 disks raid5 case, the minimum data is 64k * 2 (+ 64kb of parity).....

Bad addition on my part.


-- 
Chris Murphy