[pulseaudio-discuss] Making locking nicer for NFS

[Arun Raghavan]

On Tue, 2012-04-03 at 10:08 +0100, Colin Guthrie wrote:
> 'Twas brillig, and David Henningsson at 03/04/12 06:54 did gyre and gimble:
> > On 04/03/2012 06:27 AM, Arun Raghavan wrote:
> >> On Thu, 2011-08-25 at 10:37 -0700, Thomas Bushnell, BSG wrote:
> >>> This method also has the advantage of not relying on lock promotion
> >>> semantics, which (apparently) will make the Windoze version easier.
> >>>
> >>>
> >>> Thomas
> >>>
> >>> On Thu, Aug 25, 2011 at 10:30 AM, David Henningsson
> >>> <david.henningsson at canonical.com>  wrote:
> >>>          On 08/21/2011 04:38 PM, Thomas Bushnell, BSG wrote:
> >>>                  Whoops. They need to repeat the read after obtaining
> >>>                  the write lock and
> >>>                  only update the file if the contents are still bad in
> >>>                  that case.
> >>>
> >>>
> >>>          Would a good handling of this be:
> >>>
> >>>          1) Open the cookie read-only
> >>>          2) read the cookie
> >>>          3) close file
> >>>          4) if we have a correct cookie, do nothing more
> >>>          5) if we have the wrong cookie, do the old handling unchanged:
> >>>          open with write lock, check the contents (again), and write if
> >>>          something is (still) wrong.
> >>
> >> Thomas, David: Any news on this? Looks like we're agreed on an approach
> >> and this "just" needs to be implemented now. :)
> > 
> > As I understand it, Thomas problem was solved somehow (see
> > https://bugs.launchpad.net/ubuntu/+source/pulseaudio/+bug/817269/comments/8
> > ), and thus nobody did anything.
> > 
> > In the long term, maybe the cookie should move to XDG_RUNTIME_DIR [1],
> > which I understand would normally reside on a tmpfs, where this is not
> > an issue in the first place.
> 
> The runtime dir will only affect the native protocol socket (and other
> unix sockets we create I guess) a related NFS fix I pushed the other day
> as you know: https://bugs.freedesktop.org/show_bug.cgi?id=44680
> 
> It would not actually help with the .pulse-cookie file which was what
> this thread started as.
> 
> So I think the general principle probably still stands here.

The original problem appeared to be related locking on NFS. Moving
transient data to local storage (which I presume the runtime directory
would be) should also solve this problem, right?

-- Arun