Re: [Qemu-devel] [Qemu-block] [PATCH v2 4/4] qemu-iotests: Block migration test

[Kevin Wolf]

Am 30.05.2017 um 20:48 hat Jeff Cody geschrieben:
> On Tue, May 30, 2017 at 06:57:05PM +0200, Kevin Wolf wrote:
> > Seems you are right, but I fail to see why the other formats shouldn't
> > work? Are these just bugs or do I make any assumption in the test script
> > that is invalid for other image formats?
> >
> 
> Well, the following formats have live migration blockers:
> 
> vmdk, vhdx, vdi, vpc, qcow, vvfat

Oh, good point. We also need to fix 181 then.

> So that leaves qed, dmg, quorum.
> 
> Of those, qed is the only one that fails.  And I would guess it is a
> bug?

It is. Manual testing of a simple block migration fails, too, with qed.
It's hanging while it tries to activate the image:

#0  0x00007f5770684eef in __GI_ppoll (fds=0x7f577e322790, nfds=1, 
timeout=<optimized out>, address@hidden, address@hidden)
    at ../sysdeps/unix/sysv/linux/ppoll.c:56
#1  0x00007f577d11280b in qemu_poll_ns (__ss=0x0, __timeout=0x0, 
__nfds=<optimized out>, __fds=<optimized out>) at /usr/include/bits/poll2.h:77
#2  0x00007f577d11280b in qemu_poll_ns (fds=<optimized out>, nfds=<optimized 
out>, address@hidden) at util/qemu-timer.c:322
#3  0x00007f577d1144ca in aio_poll (address@hidden, blocking=<optimized out>) 
at util/aio-posix.c:622
#4  0x00007f577d083bb1 in qed_read_l1_table_sync (address@hidden) at 
block/qed-table.c:183
#5  0x00007f577d082ac7 in bdrv_qed_do_open (address@hidden, flags=8194, 
address@hidden, options=0x0) at block/qed.c:519
#6  0x00007f577d082cc7 in bdrv_qed_invalidate_cache (bs=0x7f577e317b50, 
errp=0x7f57593ffbe0) at block/qed.c:1646
#7  0x00007f577d054cb4 in bdrv_invalidate_cache (bs=0x7f577e317b50, 
address@hidden) at block.c:4041
#8  0x00007f577d08f77c in blk_invalidate_cache (address@hidden, address@hidden) 
at block/block-backend.c:1537
#9  0x00007f577cfff729 in block_load (f=<optimized out>, opaque=<optimized 
out>, version_id=<optimized out>) at migration/block.c:930
#10 0x00007f577cff4d97 in vmstate_load (f=0x7f577e2dc750, se=0x7f577e3109e0, 
version_id=1) at migration/savevm.c:728
#11 0x00007f577cff5232 in qemu_loadvm_state_main (mis=0x7f577d76d860 
<mis_current.32100>, f=0x7f577e2dc750) at migration/savevm.c:1917
#12 0x00007f577cff5232 in qemu_loadvm_state_main (address@hidden, 
address@hidden <mis_current.32100>) at migration/savevm.c:1948
#13 0x00007f577cff7e26 in qemu_loadvm_state (address@hidden) at 
migration/savevm.c:2014
#14 0x00007f577cff0113 in process_incoming_migration_co (opaque=0x7f577e2dc750) 
at migration/migration.c:391
#15 0x00007f577d127c6a in coroutine_trampoline (i0=<optimized out>, 
i1=<optimized out>) at util/coroutine-ucontext.c:79
#16 0x00007f57705decf0 in __start_context () at /lib64/libc.so.6

> So I would make _supported_fmt be 'qcow2 raw qed dmg quorum', and the
> failure of qed is an actual failure.

Yes, sounds good.

> > What would you think about squashing this in:
> > 
> > --- a/tests/qemu-iotests/183
> > +++ b/tests/qemu-iotests/183
> > @@ -96,11 +96,8 @@ if echo "$reply" | grep "compiled without old-style" > 
> > /dev/null; then
> >      _notrun "migrate -b support not compiled in"
> >  fi
> >  
> > -while _send_qemu_cmd $src "{ 'execute': 'query-migrate' }" "return"  |
> > -      grep '"status": "active"' > /dev/null
> > -do
> > -    sleep 0.1
> > -done
> > +QEMU_COMM_TIMEOUT=0.1 qemu_cmd_repeat=50 silent=yes \
> > +    _send_qemu_cmd $src "{ 'execute': 'query-migrate' }" '"status": 
> > "completed"'
> >  _send_qemu_cmd $src "{ 'execute': 'query-status' }" "return"
> >  
> >  echo
> >
> 
> I like that approach.  It also worked well with the qed failure case.
> Rather than hanging during the query-migrate, it times out nicely.  All
> other formats expected to work, still worked for me.

Okay, I'll send a v3.

Kevin