On Thu, Sep 15, 2016 at 12:26:16PM -0700, Andres Freund wrote:
> Hi,
>
> On 2016-09-15 15:57:55 +0200, Marco Pfatschbacher wrote:
> > the current implementation of PostmasterIsAlive() uses a pipe to
> > monitor the existence of the postmaster process.
> > One end of the pipe is held open in the postmaster, while the other end is
> > inherited to all the auxiliary and background processes when they fork.
> > This leads to multiple processes calling select(2), poll(2) and read(2)
> > on the same end of the pipe.
> > While this is technically perfectly ok, it has the unfortunate side
> > effect that it triggers an inefficient behaviour[0] in the select/poll
> > implementation on some operating systems[1]:
> > The kernel can only keep track of one pid per select address and
> > thus has no other choice than to wakeup(9) every process that
> > is waiting on select/poll.
>
> Yikes, that's a pretty absurd implementation.
Not when you take into account that it's been written over 20 years ago ;)
> Does openbsd's kqueue implementation have the same issue? There's
> http://archives.postgresql.org/message-id/CAEepm%3D37oF84-iXDTQ9MrGjENwVGds%2B5zTr38ca73kWR7ez_tA%40mail.gmail.com
Looks ok, but your milage may vary. I've seen strange subtle bugs
using kqeue..
struct kevent { __uintptr_t ident; /* identifier for this event */ short filter;
/* filter for event */ u_short flags; u_int fflags; quad_t data;
void *udata; /* opaque user data identifier */
};
> > Attached patch avoids the select contention by using a
> > separate pipe for each auxiliary and background process.
>
> I'm quite unenthusiastic about forcing that many additional file
> descriptors onto the postmaster...
In our use case it's about 30.
> I'm not quite sure I understand why this an issue here - there shouldn't
> ever be events on this fd, so why is the kernel waking up all processes?
> It'd kinda makes sense it'd wake up all processes if there's one
> waiting, but ... ?
Every read is an event, and that's what PostmasterIsAlive does.
Marco