On Fri, 2009-11-06 at 17:50 +0100, Martin Nagy wrote:
On Thu, 2009-11-05 at 15:53 -0500, Simo Sorce wrote:
> On Wed, 2009-11-04 at 18:32 +0100, Martin Nagy wrote:
> > On Mon, 2009-11-02 at 16:02 -0500, Simo Sorce wrote:
> > > On Fri, 2009-10-30 at 14:07 +0100, Martin Nagy wrote:
> > > > Hi,
> > > > attached are patches needed for the fail over functionality. The
> > > > service discovery is not there yet, I want to hold of with that
until
> > > > I
> > > > have at least a basic SRV-based one so I can test it properly.
It's
> > > > possible that we will discover something missing when we'll be
> > > > integrating it into providers. Together with Steven we at least
> > > > figured
> > > > out that for ldapi:// for example we need an "extra"
treatment. So I
> > > > made the name resolution optional and you can provide a server with
> > > > user data. The commit messages and header files should explain it
> > > > better.
> > >
> > > Just a formal NACK.
> > > We discussed changes on IRC.
> > >
> > > Simo.
> >
> > New patches attached.
> >
> > Changes:
> > * use a fail over context instead of a global variable
> > * rename fo_get_server_* to fo_resolve_service_*
> > * rename fo_set_service_status() and enum service_status to use
> > "port_status"
> > * don't use a default port for services
> > * always use only one request for resolving and add even the first
> > request as a subrequest into the queue
> > * put user_data in fo_add_server()
> > * use NULL as a server name to indicate we don't want to do resolving
> > instead of having a boolean for it
> > * in fo_resolve_service_send() only return NULL if the
> > tevent_req_create() fails
> > * add a destructor for lookup hooks, use a different memory context
> > * detect if existing server name, port and user_data combination already
> > exists and return EEXIST if they do
> >
> > Hope I didn't miss anything.
>
> Sorry still a NACK, this is the discussion on IRC:
[snip]
> Simo.
OK, should be good now. I also optimized a bit functions for getting
server/port status so they don't always call gettimeofday() if they
don't have to and I also made the code a bit cleaner. New patches
rebased on the current master are attached.
Ok, we have betean this horse enough,
Ack and pushed.
Simo.
--
Simo Sorce * Red Hat, Inc * New York