Re: [Gluster-devel] distributed locking

[Krishna Srinivas]

On Dec 5, 2007 9:54 PM, Székelyi Szabolcs <address@hidden> wrote:
> Krishna Srinivas wrote:
> > As a temporary workaround can you make all the clients use the same afr
> > translator and see if it works? i.e in client spec you need to specify the
> > same server for all the clients (instead of localhost) and afr should be on
> > the server side.
>
> It works this way, however a workaround is not required, since this is
> not a production environment; I can wait until you find and fix the
> problem (if there's any).
>
> Thanks for the advice anyway.
>
> I'm thinking that it seems the problem arises only with multiple
> posix-locks instances. Which is not a surprise in itself, because to do
> proper locking this way, the locking information between each instance
> should be synchronized at any and all times, which cannot be guaranteed
> if AFRs ask them for locking in different orders. But if the orders are
> the same, there should be no problem.
>
> Is that correct? Just to make sure I'm not bothering you with some very
> dumb thing...

That is correct.