[pnfs] [PATCH 0/6] pNFS file layout write path - help with testing

[Benny Halevy]

Andy, I submitted the patchset onto the andros-write-path branch.
I pass connectathon tests over it with panlayout (with the 3 fixes I
submitted today:
http://git.linux-nfs.org/?p=bhalevy/linux-pnfs.git;a=commitdiff;h=05962a25af175d0a7a8faeb686242200a290cb5d
http://git.linux-nfs.org/?p=bhalevy/linux-pnfs.git;a=commitdiff;h=3e6b91897aefaffdea5a33867137c6d7f3194a2c
http://git.linux-nfs.org/?p=bhalevy/linux-pnfs.git;a=commitdiff;h=c5d8cff478aa7092453dd70cda1fd70f003829c5

However, there seems to be something fishy in pnfs_writeback_done_update
that's being called for all DS writes.

	if (resp->verf->committed == NFS_FILE_SYNC ||

Shouldn't that be != NFS_FILE_SYNC?

	    (data->how == FLUSH_STABLE && data->call_ops == NULL))
		pnfs_need_layoutcommit(nfsi, argp->context);

I don't quite understand this conditional term...
1. can data->call_ops be NULL?
2. why does it matter if the write was stable or not?
Do you assume that if the DS write was FLUSH_STABLE metadata
on the MDS is already up-to-date and therefore no layoutcommit is needed?
If so that should be implemented as !((data->how & FLUSH_STABLE)

Benny

On Jan. 18, 2008, 17:01 +0200, "William A. (Andy) Adamson" <andros at citi.umich.edu> wrote:
> The first 5 patches are a resend - benny didn't receive them.
> 
> -->Andy
> 
> ---------- Forwarded message ----------
> From: William A. (Andy) Adamson <andros at citi.umich.edu>
> Date: Jan 17, 2008 2:22 PM
> Subject: [PATCH 0/5] pNFS file layout write path - help with testing
> To: pnfs at linux-nfs.org
> 
> OK - I've responded to Benny's code review. Here are the refactored patches,
> they exhibit the same behaviour - failing the special cthon test at
> 
> second check for lost reply on non-idempotent requests
> testing 50 idempotencies in directory "testdir"
> stat 1: bad file type/size 0100611/137438953473
> 
> and failing the general connectathon tests on the GPFS DSserver:
> 
> Four simultaneous large compiles test - fails with Stale NFS Handle.
> 
> Please help debug....
> 
> Thanks
> -->Andy
> 
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> pNFS mailing list
> pNFS at linux-nfs.org
> http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs