Re: [PATCH 5/6] migration/qemu-file.c: Don't ratelimit a shutdown fd

[Dr. David Alan Gilbert]

* Lukas Straub (address@hidden) wrote:
> On Mon, 18 May 2020 12:55:34 +0100
> "Dr. David Alan Gilbert" <address@hidden> wrote:
> 
> > * Zhanghailiang (address@hidden) wrote:
> > > > This causes the migration thread to hang if we failover during 
> > > > checkpoint. A
> > > > shutdown fd won't cause network traffic anyway.
> > > >   
> > > 
> > > I'm not quite sure if this modification can take side effect on normal 
> > > migration process or not,
> > > There are several places calling it.
> > > 
> > > Maybe Juan and Dave can help ;)
> > >   
> > > > Signed-off-by: Lukas Straub <address@hidden>
> > > > ---
> > > >  migration/qemu-file.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > > 
> > > > diff --git a/migration/qemu-file.c b/migration/qemu-file.c index
> > > > 1c3a358a14..0748b5810f 100644
> > > > --- a/migration/qemu-file.c
> > > > +++ b/migration/qemu-file.c
> > > > @@ -660,7 +660,7 @@ int64_t qemu_ftell(QEMUFile *f)  int
> > > > qemu_file_rate_limit(QEMUFile *f)  {
> > > >      if (f->shutdown) {
> > > > -        return 1;
> > > > +        return 0;
> > > >      }  
> > 
> > This looks wrong to me; I'd be curious to understand how it's hanging
> > for you.
> > '1' means 'stop what you're doing', 0 means carry on; carrying on with a
> > shutdown fd sounds wrong.
> > 
> > If we look at ram.c we have:
> > 
> >         while ((ret = qemu_file_rate_limit(f)) == 0 ||
> >                 !QSIMPLEQ_EMPTY(&rs->src_page_requests)) {
> >             int pages;
> >         ....
> > 
> > so if it returns '1', as it does at the moment it should cause it to
> > exit the ram_save_iterate loop - which is what we want if it's failing.
> > Thus I think you need to find the actual place it's stuck in this case -
> > I suspect it's repeatedly calling ram_save_iterate and then exiting it,
> > but if that's happening perhaps we're missing a qemu_file_get_error
> > check somewhere.
> 
> Hi,
> the problem is in ram_save_host_page and migration_rate_limit, here is a 
> backtrace:

Ah...

> #0  0x00007f7b502921a8 in futex_abstimed_wait_cancelable (private=0, 
> abstime=0x7f7ada7fb3f0, clockid=0, expected=0, futex_word=0x55bc358b9908) at 
> ../sysdeps/unix/sysv/linux/futex-internal.h:208
> #1  do_futex_wait (sem=sem@entry=0x55bc358b9908, 
> abstime=abstime@entry=0x7f7ada7fb3f0, clockid=0) at sem_waitcommon.c:112
> #2  0x00007f7b502922d3 in __new_sem_wait_slow (sem=0x55bc358b9908, 
> abstime=0x7f7ada7fb3f0, clockid=0) at sem_waitcommon.c:184
> #3  0x000055bc3382b6c1 in qemu_sem_timedwait (sem=0x55bc358b9908, ms=100) at 
> util/qemu-thread-posix.c:306
> #4  0x000055bc3363950b in migration_rate_limit () at 
> migration/migration.c:3365

OK, so how about:

diff --git a/migration/migration.c b/migration/migration.c
index b6b662e016..4e885385a8 100644
--- a/migration/migration.c
+++ b/migration/migration.c
@@ -3356,6 +3356,10 @@ bool migration_rate_limit(void)
     bool urgent = false;
     migration_update_counters(s, now);
     if (qemu_file_rate_limit(s->to_dst_file)) {
+
+        if (qemu_file_get_error(mis->from_src_file)) {
+            return false;
+        }
         /*
          * Wait for a delay to do rate limiting OR
          * something urgent to post the semaphore.

Does that work?
I wonder if we also need to kick the rate_limit_sem when we yank the
socket.

Dave

> #5  0x000055bc332b70d3 in ram_save_host_page (rs=0x7f7acc001a70, 
> pss=0x7f7ada7fb4b0, last_stage=true) at /home/lukas/qemu/migration/ram.c:1696
> #6  0x000055bc332b71fa in ram_find_and_save_block (rs=0x7f7acc001a70, 
> last_stage=true) at /home/lukas/qemu/migration/ram.c:1750
> #7  0x000055bc332b8bbd in ram_save_complete (f=0x55bc36661330, 
> opaque=0x55bc33fbc678 <ram_state>) at /home/lukas/qemu/migration/ram.c:2606
> #8  0x000055bc3364112c in qemu_savevm_state_complete_precopy_iterable 
> (f=0x55bc36661330, in_postcopy=false) at migration/savevm.c:1344
> #9  0x000055bc33641556 in qemu_savevm_state_complete_precopy 
> (f=0x55bc36661330, iterable_only=true, inactivate_disks=false) at 
> migration/savevm.c:1442
> #10 0x000055bc33641982 in qemu_savevm_live_state (f=0x55bc36661330) at 
> migration/savevm.c:1569
> #11 0x000055bc33645407 in colo_do_checkpoint_transaction (s=0x55bc358b9840, 
> bioc=0x7f7acc059990, fb=0x7f7acc4627b0) at migration/colo.c:464
> #12 0x000055bc336457ca in colo_process_checkpoint (s=0x55bc358b9840) at 
> migration/colo.c:589
> #13 0x000055bc336459e4 in migrate_start_colo_process (s=0x55bc358b9840) at 
> migration/colo.c:666
> #14 0x000055bc336393d7 in migration_iteration_finish (s=0x55bc358b9840) at 
> migration/migration.c:3312
> #15 0x000055bc33639753 in migration_thread (opaque=0x55bc358b9840) at 
> migration/migration.c:3477
> #16 0x000055bc3382bbb5 in qemu_thread_start (args=0x55bc357c27c0) at 
> util/qemu-thread-posix.c:519
> #17 0x00007f7b50288f27 in start_thread (arg=<optimized out>) at 
> pthread_create.c:479
> #18 0x00007f7b501ba31f in clone () at 
> ../sysdeps/unix/sysv/linux/x86_64/clone.S:95
> 
> It hangs in ram_save_host_page for at least 10 Minutes.
> 
> Regards,
> Lukas Straub
> 
> > Dave
> > 
> > > >      if (qemu_file_get_error(f)) {
> > > >          return 1;
> > > > --
> > > > 2.20.1  
> > >   
> > --
> > Dr. David Alan Gilbert / address@hidden / Manchester, UK
> > 
> 


--
Dr. David Alan Gilbert / address@hidden / Manchester, UK