[ovs-dev] [RFC/PATCH v2] Make the PID part of socket path configurable

[Ben Pfaff]

On Mon, Aug 08, 2016 at 10:05:03PM +0200, Christian Svensson wrote:
> On Mon, Aug 8, 2016 at 7:59 PM, Ben Pfaff <blp at ovn.org> wrote:
> 
> > OK, let's add that then.
> 
> While it's an alternative, I feel like going down that path is hasty and
> for my use case I will probably just patch our local fork of OpenvSwitch.
>
> Why so hostile towards this change? Is there a better forum I can explain
> why I believe this problem should be reconsidered?

Including the PID allows multiple daemons of a single type to run.  I
also don't think you're solving a real problem, because reading a
pidfile isn't difficult.

> Is there really anyone that can justify the way socket paths are created
> should contain the PID? You clearly saw the problem with it in the Windows
> code base, but yet pushed on to keep it in the Linux one. I simply do not
> see why you're not taking this patch as a way to encourage people to
> migrate towards a saner default, and somewhere down the line maybe even
> switch the default behavior.

I think that Windows doesn't support getpid().  MSDN documents it that
way, at least.  That is probably why we omit it on Windows, though I do
not know for sure.  Guru, is that the reason?

I don't understand why you want to change this so badly.  It's not hard
to read a pidfile.

I definitely don't want to fork OVS behavior here based on a
configuration flag, as I already explained.

Basically: I see little benefit to your change, and some drawbacks.