On Wed, Jun 05, 2019 at 02:42:00PM -0600, Tom Hromatka
<[email protected]> wrote:
> I created a simple double-forking test program to try and reproduce the
> issue, but I was unable to get it to happen. Is your system heavily
> loaded? Any recommendations on reproducing the issue?
I reproduce this when cgrulesengd is slowed down relatively, the overall load
or renice helps this. And the double forking program is multithreaded.
> Otherwise I think your changes look reasonable, but I would really like
> to test them out before giving my Reviewed-by.
Actually, when I reran my reproducer again, I noticed on more failure path for
a short-lived process, hence I'm sending v2 of the patch.
## My reproducer
cat >/etc/cgconfig.conf <<EOD
mount {
cpuset = /cgroup/cpuset;
cpu = /cgroup/cpu;
cpuacct = /cgroup/cpuacct;
memory = /cgroup/memory;
devices = /cgroup/devices;
freezer = /cgroup/freezer;
net_cls = /cgroup/net_cls;
blkio = /cgroup/blkio;
}
group testgroup {
cpu {
cpu.cfs_period_us=100000;
cpu.cfs_quota_us=50000;
cpu.shares=1024;
}
cpuacct {
cpuacct.usage=0;
}
memory {
memory.limit_in_bytes=1073741824;
}
}
EOD
cat >/etc/cgrules.conf <<EOD
root:/root/repro/sleep-server-double-mt cpu,memory,cpuacct testgroup/
EOD
cat >/root/repro/sleep-server-double-mt.c <<EOD
#include <unistd.h>
#include <stdlib.h>
#include <string.h>
#include <pthread.h>
void *worker_fn(void *arg) {
long duration = (long) arg;
if (duration > 1)
duration -= 1;
sleep(duration);
return NULL;
}
int main(int argc, char *argv[]) {
int n, i;
int duration;
if(fork()) { // parent
usleep(100000);
} else { // child
n = (argc > 2) ? atoi(argv[2]) : 2;
duration = atoi(argv[1]);
pthread_t *threads = malloc(sizeof(pthread_t) * n);
for (i = 0; i < n; ++i) {
pthread_create(threads + i, NULL, worker_fn, (void
*)duration);
}
sleep(duration);
for (i = 0; i < n; ++i) {
pthread_join(threads[i], NULL);
}
free(threads);
}
return 0;
}
EOD
cat >/root/repro/Makefile <<EOD
CFLAGS+=-pthread
LDFLAGS+=-lpthread
all: sleep-server-double-mt
EOD
cat >/root/repro/loop.sh <<EOD
#!/bin/bash
DAEMON=/root/repro-public/sleep-server-double-mt
CGROUP=testgroup
killproc $DAEMON || :
counter=0
while true ; do
#strace -f -tt -o strace.$counter start_daemon $DAEMON 10000 4 &
start_daemon $DAEMON 10000 4 &
while true ; do
sleep 0.2
pid=$(pidof $DAEMON)
if [ $? -ne 0 ] ; then
# echo "fail"
continue
fi
if [[ "$pid" =~ " " ]] ; then
# echo ">$pid<"
continue
fi
break
done
file=/proc/$pid/cgroup
if [ $(grep -c $CGROUP $file) -lt 3 ] ; then
echo "Too little cgroups (PID=$pid, strace.$counter): "
echo "---"
cat $file
echo "---"
break;
fi
echo -n "."
killproc $DAEMON
counter=$(($counter+1))
done
EOD
Michal Koutný (1):
cgrulesengd: Do not ignore changes of short-lived processes
src/daemon/cgrulesengd.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)
--
2.21.0
_______________________________________________
Libcg-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libcg-devel
