Hang with migration multi-thread compression under high load

On Wed, Apr 27, 2016 at 03:29:30PM +0100, Dr. David Alan Gilbert wrote:
> ccing in Liang Li
> 
> * Daniel P. Berrange (berrange@redhat.com) wrote:
> > for some reason it isn't shown in the stack thrace for thread
> > 1 above, when initially connecting GDB it says the main thread
> > is at:
> > 
> > decompress_data_with_multi_threads (len=702, host=0x7fd78fe06000, f=0x55901af09950) at /home/berrange/src/virt/qemu/migration/ram.c:2254
> > 2254	        for (idx = 0; idx < thread_count; idx++) {
> > 
> > 
> > Looking at the target QEMU, we see  do_data_decompress method
> > is waiting in a condition var:
> > 
> >         while (!param->start && !quit_decomp_thread) {
> > 	    qemu_cond_wait(&param->cond, &param->mutex);
> >             ....do stuff..
> > 	    param->start = false
> >         }
> > 
> > 
> > Now the decompress_data_with_multi_threads is checking param->start without
> > holding the param->mutex lock.
> > 
> > Changing decompress_data_with_multi_threads to acquire param->mutex
> > lock makes it work, but isn't ideal, since that now blocks the
> > decompress_data_with_multi_threads() method on the completion of
> > each thread, which defeats the point of having multiple threads.

FWIW, the following patch also appears to "fix" the issue, presumably
by just making the race much less likely to hit:


Incidentally IIUC, this decompress_data_with_multi_threads is just
busy waiting for a thread to become free, which seems pretty wasteful
of CPU resources. I wonder if there's a more effective way to structure
this, so that instead of having decompress_data_with_multi_threads()
choose which thread to pass the decompression job to, it just puts
the job into a queue, and then let all the threads pull from that
shared queue. IOW whichever thread the kerenl decides to wakeup would
get the job, without us having to explicitly assign a thread to the
job.


Regards,
Daniel

> On Wed, Apr 27, 2016 at 03:29:30PM +0100, Dr. David Alan Gilbert wrote:

> > ccing in Liang Li

> >

> > * Daniel P. Berrange (berrange@redhat.com) wrote:

> > > for some reason it isn't shown in the stack thrace for thread

> > > 1 above, when initially connecting GDB it says the main thread is

> > > at:

> > >

> > > decompress_data_with_multi_threads (len=702, host=0x7fd78fe06000,

> f=0x55901af09950) at /home/berrange/src/virt/qemu/migration/ram.c:2254

> > > 2254	        for (idx = 0; idx < thread_count; idx++) {

> > >

> > >

> > > Looking at the target QEMU, we see  do_data_decompress method is

> > > waiting in a condition var:

> > >

> > >         while (!param->start && !quit_decomp_thread) {

> > > 	    qemu_cond_wait(&param->cond, &param->mutex);

> > >             ....do stuff..

> > > 	    param->start = false

> > >         }

> > >

> > >

> > > Now the decompress_data_with_multi_threads is checking param->start

> > > without holding the param->mutex lock.

> > >

> > > Changing decompress_data_with_multi_threads to acquire param-

> >mutex

> > > lock makes it work, but isn't ideal, since that now blocks the

> > > decompress_data_with_multi_threads() method on the completion of

> > > each thread, which defeats the point of having multiple threads.

> 

> FWIW, the following patch also appears to "fix" the issue, presumably by just

> making the race much less likely to hit:

> 

> diff --git a/migration/ram.c b/migration/ram.c index 3f05738..be0233f 100644

> --- a/migration/ram.c

> +++ b/migration/ram.c

> @@ -2271,6 +2271,7 @@ static void

> decompress_data_with_multi_threads(QEMUFile *f,

>          if (idx < thread_count) {

>              break;

>          }

> +        sched_yield();

>      }

>  }

> 

> 

> Incidentally IIUC, this decompress_data_with_multi_threads is just busy

> waiting for a thread to become free, which seems pretty wasteful of CPU

> resources. I wonder if there's a more effective way to structure this, so that

> instead of having decompress_data_with_multi_threads()

> choose which thread to pass the decompression job to, it just puts the job

> into a queue, and then let all the threads pull from that shared queue. IOW

> whichever thread the kerenl decides to wakeup would get the job, without

> us having to explicitly assign a thread to the job.

> 

> 


Thanks for reporting this issue, I will take a look and get back to you.

Liang

> Regards,

> Daniel

> --

> |: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|

> |: http://libvirt.org              -o-             http://virt-manager.org :|

> |: http://autobuild.org       -o-         http://search.cpan.org/~danberr/ :|

> |: http://entangle-photo.org       -o-       http://live.gnome.org/gtk-vnc :|