Bug#174521: Threading bug leading to zombies? / licq problems

[Will Aoki <waoki@umnh.utah.edu>]

[resuming this thread after months. Preserving quoted material because
the list archives have a hole between 18 Sep 2002 and 1 Oct 2002.]

On Sun, Sep 29, 2002 at 06:05:21PM +0200, Christian Jaeger wrote:
> Hello
> 
> I've problems running licq on my powerbook G3. Frequently it leaves 
> zombies around until at 33 threads (including the zombies) it opens a 
> dialog saying that it can't create new threads because of no 
> resources being available. (Strangely, ulimit -u says 2560, and a 
> perl test script can in fact create that many zombies, so the limit 
> must be set somewhere else but I don't care now about that)
> 
> I'm noticing some other strange things on my machine:
> 
> chris@lombi chris > ps aux|grep ' Z '
> chris    26793      0    0 ?        Z    Sep24   0:00 [nc <defunct>]
> mysql    28948      0    0 ?        Z    06:25   0:00 [mysqld <defunct>]
> mysql    28956      0    0 ?        Z    06:25   0:00 [mysqld <defunct>]
> chris    23281      0    0 ?        Z    15:26   0:00 [netstat <defunct>]
> (trimmed to fit the line width)
> 
> mysqld leaving zombies behind? That must be a bug on my system, it 
> would be known pretty fast if that would be a mysql bug.
> 
> (netstat is forked off by galeon-bin (huh, what for?? but anyway) so 
> that's probably only a bug of galeon)
> 
> I'm also seing segfaults of nedit since about the 7th of august. I've 
> already reinstalled libmotif and recompiled/upgraded nedit without 
> any change: every time it opens a dialog it crashes. I thought that 
> maybe some library is damaged but I ran debsums and except libc6 
> which doesn't have md5sums everything seemed ok, and even after a 
> recent libc6 security upgrade it's still happening (the nedit crash 
> as well as the zombies).
> 
> So my hypothesises:
> 
> - libc6 might have a ppc dependent bug in thread handling.
> - the kernel I'm running has a bug.
> 
> Has anyone else seen those problems?

I've been investigating this problem. I need more data for a bug report.
If anyone else is still observing it, please do the following:

* Compile the attached pthread-test.c program with:
  'gcc -o pthread-test pthread-test.c -lpthreaad'

* Run it while the zombie problem is occuring. Count how many zombies it
  creates. Zombies are reported at the end of the run.

* If you're in X, kill X, run pthread-test again, and see how many
  zombies it reports.

* If it's now reporting no zombies, or only one or two zombies:
  Create /tmp/foo.sh with the contents:
     #!/bin/sh
     sleep 1 &
     sleep 30
  chmod +x it, run it, and while it's running, wait a second, then run
  pthread-test in another console, and see how many zombies it reports.

* Send me mail (off-list; I'll summarize) with zombie counts, the kernel
  version that you're running, your kernel configuration (if it's not a
  pre-compiled Debian kernel), and the results of 'dpkg -l libc6' and
  'lsmod'. If there's any other information you think pertinent, please
  send it, too.

Thanks in advance!

-- 
William Aoki     waoki@umnh.utah.edu       /"\  ASCII Ribbon Campaign
B1FB C169 C7A6 238B 280B  <- key change    \ /  No HTML in mail or news!
99AF A093 29AE 0AE1 9734   prev. expired    X
                                           / \

#include <pthread.h>
#include <errno.h>

#define HOW_MANY_THREADS 10

void threadguts(void *d) {

	int i;

	printf("Thread %i spawned.\n", (int *) d);
	for (i = 0; i < 3; i++) {
		printf("Thread %i doing stuff.\n", (int *) d);
	}

	pthread_exit(0);

}

/* Make a few threads that do stuff, then die. */
int main() {
	pthread_t thread[HOW_MANY_THREADS];
	pthread_attr_t attr;
	void * s;
	int i;
	int e;

	pthread_attr_init(&attr);

	
	for (i = 0; i < HOW_MANY_THREADS; i++) {
		printf("Spawning thread %i\n", i);
		if (e = pthread_create(&thread[i], &attr, (void *) threadguts, (void *) i)) {
			//perror("pthread_create:");
			printf("pthread_create: %s\n", strerror(e));
		}
	}

	/* Make sure they're dead. */
	for (i = 0; i < HOW_MANY_THREADS; i++) {
		printf("Waiting to join thread %i\n", i);
		if (pthread_join(thread[i], NULL)) {
			printf("Can't join thread %i\n", i);
		} else {
			printf("Joined thread %i\n", i);
		}
	}

	pthread_attr_destroy(&attr);


	/* This won't work if too many threads were created - it will
	   be unable to fork
	 */
	system("/bin/ps | grep defunct");

	return 0;

}