Re: [nfsv4] LAYOUTERROR in draft-ietf-nfsv4-minorversion2-40

You might consider reporting multiple ranges by using multiple
LAYOUTERRORs in a COMPOUND, given that each
LAYOUTERROR only allows a single range to be reported.

On Sat, Jan 23, 2016 at 10:45 AM, Trond Myklebust <trondmy@gmail.com> wrote:

> Hi,
>
> When looking to implement LAYOUTERROR from
> https://tools.ietf.org/html/draft-ietf-nfsv4-minorversion2-40, I
> noticed there is a major difference between it and the way we're doing
> error passing in LAYOUTRETURN. While the latter allows you to send an
> array of several error reports for different byte ranges on the same
> file, the LAYOUTERROR spec only allows you to specify a single byte
> range.
>
> Being able to specify errors as an array of ranges is somewhat of a
> requirement for the flexfiles environment. In the case where the data
> is mirrored across several DSes, reporting the actual byte ranges
> where errors occurred can significantly cut down on the amount of work
> that the server has to perform to fix the mirroring when errors start
> to occur; instead of rewriting the whole file, it can just rewrite
> those chunks that had errors.
>
> A typical scenario where this might happen is if the file is fenced,
> and so the WRITEs start receiving EACCES calls; unless you are
> extraordinarily lucky, the fencing will leave one set of successful
> writes on one mirror image, and a different set on the other (due to
> timing, RPC reordering, etc).
>
> While I can continue to use LAYOUTRETURN for most cases, I was hoping
> to use LAYOUTERROR for those cases where we have queued up lots of
> errors (due to parallel WRITEs), and so want to split the error
> reporting into multiple RPC calls.
>
> Cheers
>   Trond
>
> _______________________________________________
> nfsv4 mailing list
> nfsv4@ietf.org
> https://www.ietf.org/mailman/listinfo/nfsv4
>