Re: [PATCH] xargs with large (or zero) max-procs

[Marc Aurele La France]

On Sat, 5 Jan 2008, James Youngman wrote:
> On Dec 26, 2007 10:51 PM, Marc Aurele La France <address@hidden> wrote:
> > The -p1 patch below causes `xargs` to reap its zombied children before
> > attempting to fork() another child process.  This is intended to make it
> > more difficult for a user to fill up the process table, potentially
> > causing other system services to fail.

> First thanks for reporting the bug.   I have recorded it at
> http://savannah.gnu.org/bugs/index.php?21960

> I opted for a fix that is somewhat different to your proposal.  I
> wanted to have xargs reap children a little more frequently than
> provided for in your code change (for example, before fork when -P50
> is in effect).   Also though, xargs had a number of variables which
> were signed but not indended to hold a negative quantity; some of
> these related to counting the number of outstanding child processes.
> I took the opportunity to fix that too.

> The change I prepared was emailed (as usual) to
> address@hidden and is attached to the bug at the above URL.
> I'll apply it to the CVS tree in a couple of days, once people have
> had an opportunity to review it.

Looks good from here.

> This means that despite your work to fix this problem, the code change
> you suggested wasn't the one I used.   I hope you don't mind.

Not a problem.  Expected, in fact.  Although, I'm ever so slightly 
disappointed at the removal of support for ancient systems without waitpid().

> Nevertheless your change was a good one and should you want to further
> contribute to findutils, we'd be glad of the help.

I already know this to be quite true.  I have submitted patches to many 
projects over the years, and this one has been one of the most positively 
responsive.

> > This is keyed on whether or not <sys/wait.h> #define's WNOHANG, but you
> > might want to key off of the existence of waitpid(), using the automangle
> > suite.  I am submitting this in its current form given that waitpid() is
> > just about as old as wait(), older than wait3() and wait4() even, and that
> > there is no point in WNOHANG without waitpid() or vice-versa.

> I agree that waitpid is well-established.   Every bit of
> autoconfiscation imposes a small extra maintenance burden.   Partly
> for that reason, I decided to just assume the existence of both
> waitpid() and WNOHANG.

Alright.  Not worth resurrecting my old 7th Edition system for ;-)

Thanks.

Marc.

+----------------------------------+----------------------------------+
|  Marc Aurele La France           |  work:   1-780-492-9310          |
|  Academic Information and        |  fax:    1-780-492-1729          |
|    Communications Technologies   |  email:  address@hidden         |
|  352 General Services Building   +----------------------------------+
|  University of Alberta           |                                  |
|  Edmonton, Alberta               |    Standard disclaimers apply    |
|  T6G 2H1                         |                                  |
|  CANADA                          |                                  |
+----------------------------------+----------------------------------+
XFree86 developer and VP.  ATI driver and X server internals.