Re: Red Hat needs this patch (was Re: handling NFSERR_JUKEBOX)

[Marcelo Tosatti]

On 6 Nov 2001, Trond Myklebust wrote:

> >>>>> " " == Bob Smart <[e-mail]> writes:
> 
>     >> Take a look at the nfs sourceforge mailing list.  There's a
>     >> long thread about this in the context of a Solaris HSM server.
>     >> I wrote a patch to try to do the right thing on a linux client,
>     >> but it is not perfect for many reasons:
>     >> http://devrandom.net/lists/archives/2001/6/NFS/0135.html
> 
>      > Thanks again for this patch. I've appended our current version
>      > to save others the trouble of minor modifications. I've just
>      > tested it against 2.4.13-ac7.
> 
>      > Your patch may not be perfect but it is a million times better
>      > than nothing. We have had no problems with it. Linux has a long
>      > tradition of utilizing imperfect software till something better
>      > comes along.
> 
>      > This patch is ESSENTIAL to make Linux useable in large
>      > enterprises, because they increasingly commonly utilize
>      > hirearchically managed storage. I strongly urge
>      > Alan/Linus/Macello to include this patch in the stable kernel
>      > tree (but marked experimental). Or at least can the kernel
>      > maintainers at Red Hat and other distributions put it in their
>      > offerings: surely you see that you need this for your big
>      > customers.
> 
> The patch may appear to work, but it fails to understand the nature of
> the problem. The problem with NFSERR_JUKEBOX is that is can appear in
> more or less *any* NFSv3 RPC call. That includes everything from
> lookups to writes.
> 
> If all you do is intercept read and setattr, then you certainly
> haven't achieved 'enterprise quality'.

Trond, 

Should'nt we make Linux retry on NFSERR_JUKEBOX _by default_ ? 

Is there any reason why we already don't do that already except "nobody
coded it yet" ?