killing NFS mode? objections?

[komtanoo.pinpimai at livetext.com]

Yes, kill it, don't think anybody would want to use nfs over MogileFS, but
if decide to keep it, there should be communication abstraction interface
that could handle HTTP, NFS, or other protocols, and I don't think it's
easy to make them compatible, especially error handling.

-kem

On Fri, September 29, 2006 1:22 pm, Brad Fitzpatrick wrote:
> Would anybody care if I removed the
> all-storage-nodes-locally-mounted-everywhere code from Mogile?  (both
> server and clients)
>
> This is more commonly known as "NFS mode".
>
>
> Rather than assigning HTTP urls to callers, absolute paths are given, so
> people have to all have the mogile storage nodes mounted at the same place.
>
>
> And Linux NFS doesn't even let this work, due to not being able to use
> noshadow mounts w/ export netmask wildcards.
>
> It was just a royal pain in the ass, which is why we moved to HTTP.
>
>
> In the meantime, I think NFS support might've rotted, but I hear maybe it
>  still works?
>
> In any case, it'd clean up the code to remove it, and remove support
> costs.  ("you did what?  no, no, don't use /that/ ...")
>
> If it's still desired from the client-side, I can see a client-side
> transform/check being done... given a URL, see if has a local NFS mount,
> and use that instead.  But that's unrelated to removing it from the
> server.  The server would then only replicate using HTTP, which is a lot
> better anyway, since we can tell which side failed, since we're involved
> in the copy, whereas NFS mounts just kinda hang and are useless.
>
> - Brad
>
>
>