[ILUG] Maximum value of a PID

[Niall]

On Sun, Mar 12, 2000 at 07:22:24PM +0000, Nick Hilliard wrote:
> > I read in some post to comp.unix.programmer that the maximum pid
> > is 30000, and that's a historical limit, going back to AT&T SysV.
> > It'll probably need a kernel hacker to shed some light ...
> 
> It's only a historic limit in the same sense that, for example, file names
> are limited to 14 characters.  The limit died a long, long time ago.
> 
> POSIX states that pid_t should be of type int, but most versions of unix
> wrap pid's at either 32000 or eles 32767.  FreeBSD goes up to 99999 before
> wrapping (for legibility), and there may be some other regional varieties.

As I mentioned, I just noticed the wraparound happening (and no, I didn't
force it - it was pure happenstance) and I wondered how the next PID is
chosen. I saw PID 32766 and then PID 413 in my file serial numbers. Now, the
production of one of these files in itself uses up a few PIDs and this is on
a web server. There was AFAIR 8 minutes between these two, so there could
easily have been 400 processes in that time. But, I did what we're supposed
to do - I read the source. I don't do that as often as I should before
asking a question (there's such a lot of it :-) ) but in this case, I lucked
out because it was in the first place I looked (kernel/fork.c) and it wraps
to 300 - quite an arbitrary number, just chosen to be probably bigger than
all startup daemons etc. Oh and I looked - it does wrap at 32768, but
there's no such thing as a comment to indicate why this is so.



Kindest regards,


Niall  O Broin