Re: [PULL 00/12] Block patches

[Kevin Wolf]

Am 09.07.2020 um 20:41 hat Eduardo Habkost geschrieben:
> On Thu, Jul 09, 2020 at 05:02:06PM +0200, Kevin Wolf wrote:
> > Am 08.07.2020 um 00:05 hat Eduardo Habkost geschrieben:
> > > On Tue, Jul 07, 2020 at 05:28:21PM +0200, Philippe Mathieu-Daudé wrote:
> > > > On 6/26/20 12:25 PM, Stefan Hajnoczi wrote:
> > > > > On Thu, Jun 25, 2020 at 02:31:14PM +0100, Peter Maydell wrote:
> > > > >> On Wed, 24 Jun 2020 at 11:02, Stefan Hajnoczi <stefanha@redhat.com> 
> > > > >> wrote:
> > > > >>>
> > > > >>> The following changes since commit 
> > > > >>> 171199f56f5f9bdf1e5d670d09ef1351d8f01bae:
> > > > >>>
> > > > >>>   Merge remote-tracking branch 
> > > > >>> 'remotes/alistair/tags/pull-riscv-to-apply-20200619-3' into staging 
> > > > >>> (2020-06-22 14:45:25 +0100)
> > > > >>>
> > > > >>> are available in the Git repository at:
> > > > >>>
> > > > >>>   https://github.com/stefanha/qemu.git tags/block-pull-request
> > > > >>>
> > > > >>> for you to fetch changes up to 
> > > > >>> 7838c67f22a81fcf669785cd6c0876438422071a:
> > > > >>>
> > > > >>>   block/nvme: support nested aio_poll() (2020-06-23 15:46:08 +0100)
> > > > >>>
> > > > >>> ----------------------------------------------------------------
> > > > >>> Pull request
> > > > >>>
> > > > >>> ----------------------------------------------------------------
> > > > >>
> > > > >> Failure on iotest 030, x86-64 Linux:
> > > > >>
> > > > >>   TEST    iotest-qcow2: 030 [fail]
> > > > >> QEMU          --
> > > > >> "/home/petmay01/linaro/qemu-for-merges/build/alldbg/tests/qemu-iotests/../../x86_64-softmmu/qemu-system-x86_64"
> > > > >> -nodefaults -display none -accel qtest
> > > > >> QEMU_IMG      --
> > > > >> "/home/petmay01/linaro/qemu-for-merges/build/alldbg/tests/qemu-iotests/../../qemu-img"
> > > > >> QEMU_IO       --
> > > > >> "/home/petmay01/linaro/qemu-for-merges/build/alldbg/tests/qemu-iotests/../../qemu-io"
> > > > >>  --cache writeback --aio threads -f qcow2
> > > > >> QEMU_NBD      --
> > > > >> "/home/petmay01/linaro/qemu-for-merges/build/alldbg/tests/qemu-iotests/../../qemu-nbd"
> > > > >> IMGFMT        -- qcow2 (compat=1.1)
> > > > >> IMGPROTO      -- file
> > > > >> PLATFORM      -- Linux/x86_64 e104462 4.15.0-76-generic
> > > > >> TEST_DIR      --
> > > > >> /home/petmay01/linaro/qemu-for-merges/build/alldbg/tests/qemu-iotests/scratch
> > > > >> SOCK_DIR      -- /tmp/tmp.8tgdDjoZcO
> > > > >> SOCKET_SCM_HELPER --
> > > > >> /home/petmay01/linaro/qemu-for-merges/build/alldbg/tests/qemu-iotest/socket_scm_helper
> > > > >>
> > > > >> --- /home/petmay01/linaro/qemu-for-merges/tests/qemu-iotests/030.out
> > > > >>  2019-07-15 17:18:35.251364738 +0100
> > > > >> +++ 
> > > > >> /home/petmay01/linaro/qemu-for-merges/build/alldbg/tests/qemu-iotests/030.out.bad
> > > > >>   2020-06-25 14:04:28.500534007 +0100
> > > > >> @@ -1,5 +1,17 @@
> > > > >> -...........................
> > > > >> +.............F.............
> > > > >> +======================================================================
> > > > >> +FAIL: test_stream_parallel (__main__.TestParallelOps)
> > > > >> +----------------------------------------------------------------------
> > > > >> +Traceback (most recent call last):
> > > > >> +  File "030", line 246, in test_stream_parallel
> > > > >> +    self.assert_qmp(result, 'return', {})
> > > > >> +  File 
> > > > >> "/home/petmay01/linaro/qemu-for-merges/tests/qemu-iotests/iotests.py",
> > > > >> line 848, in assert_qmp
> > > > >> +    result = self.dictpath(d, path)
> > > > >> +  File 
> > > > >> "/home/petmay01/linaro/qemu-for-merges/tests/qemu-iotests/iotests.py",
> > > > >> line 822, in dictpath
> > > > >> +    self.fail(f'failed path traversal for "{path}" in "{d}"')
> > > > >> +AssertionError: failed path traversal for "return" in "{'error':
> > > > >> {'class': 'DeviceNotActive', 'desc': "Block job 'stream-node8' not
> > > > >> found"}}"
> > > > >> +
> > > > >>  
> > > > >> ----------------------------------------------------------------------
> > > > >>  Ran 27 tests
> > > > >>
> > > > >> -OK
> > > > >> +FAILED (failures=1)
> > > > > 
> > > > > Strange, I can't reproduce this failure on my pull request branch or 
> > > > > on
> > > > > qemu.git/master.
> > > > > 
> > > > > Is this failure deterministic? Are you sure it is introduced by this
> > > > > pull request?
> > > > 
> > > > Probably not introduced by this pullreq, but I also hit it on FreeBSD:
> > > > https://cirrus-ci.com/task/4620718312783872?command=main#L5803
> > > > 
> > > >   TEST    iotest-qcow2: 030 [fail]
> > > > QEMU          --
> > > > "/tmp/cirrus-ci-build/build/tests/qemu-iotests/../../aarch64-softmmu/qemu-system-aarch64"
> > > > -nodefaults -display none -machine virt -accel qtest
> > > > QEMU_IMG      --
> > > > "/tmp/cirrus-ci-build/build/tests/qemu-iotests/../../qemu-img"
> > > > QEMU_IO       --
> > > > "/tmp/cirrus-ci-build/build/tests/qemu-iotests/../../qemu-io"  --cache
> > > > writeback --aio threads -f qcow2
> > > > QEMU_NBD      --
> > > > "/tmp/cirrus-ci-build/build/tests/qemu-iotests/../../qemu-nbd"
> > > > IMGFMT        -- qcow2 (compat=1.1)
> > > > IMGPROTO      -- file
> > > > PLATFORM      -- FreeBSD/amd64 cirrus-task-4620718312783872 12.1-RELEASE
> > > > TEST_DIR      -- /tmp/cirrus-ci-build/build/tests/qemu-iotests/scratch
> > > > SOCK_DIR      -- /tmp/tmp.aZ5pxFLF
> > > > SOCKET_SCM_HELPER --
> > > > --- /tmp/cirrus-ci-build/tests/qemu-iotests/030.out     2020-07-07
> > > > 14:48:48.123804000 +0000
> > > > +++ /tmp/cirrus-ci-build/build/tests/qemu-iotests/030.out.bad   
> > > > 2020-07-07
> > > > 15:05:07.863685000 +0000
> > > > @@ -1,5 +1,17 @@
> > > > -...........................
> > > > +.............F.............
> > > > +======================================================================
> > > > +FAIL: test_stream_parallel (__main__.TestParallelOps)
> > > >  ----------------------------------------------------------------------
> > > > +Traceback (most recent call last):
> > > > +  File "030", line 246, in test_stream_parallel
> > > > +    self.assert_qmp(result, 'return', {})
> > > > +  File "/tmp/cirrus-ci-build/tests/qemu-iotests/iotests.py", line 848,
> > > > in assert_qmp
> > > > +    result = self.dictpath(d, path)
> > > > +  File "/tmp/cirrus-ci-build/tests/qemu-iotests/iotests.py", line 822,
> > > > in dictpath
> > > > +    self.fail(f'failed path traversal for "{path}" in "{d}"')
> > > > +AssertionError: failed path traversal for "return" in "{'error':
> > > > {'class': 'DeviceNotActive', 'desc': "Block job 'stream-node8' not 
> > > > found"}}"
> > > > +
> > > > +----------------------------------------------------------------------
> > > >  Ran 27 tests
> > > 
> > > Looks like a race condition that can be forced with a sleep call.
> > > With the following patch, I can reproduce it every time:
> > > 
> > >   diff --git a/tests/qemu-iotests/030 b/tests/qemu-iotests/030
> > >   index 1cdd7e2999..ee5374fc22 100755
> > >   --- a/tests/qemu-iotests/030
> > >   +++ b/tests/qemu-iotests/030
> > >   @@ -241,6 +241,7 @@ class TestParallelOps(iotests.QMPTestCase):
> > >                result = self.vm.qmp('block-stream', device=node_name, 
> > > job_id=job_id, base=self.imgs[i-2], speed=512*1024)
> > >                self.assert_qmp(result, 'return', {})
> > >   
> > >   +        time.sleep(3)
> > >            for job in pending_jobs:
> > >                result = self.vm.qmp('block-job-set-speed', device=job, 
> > > speed=0)
> > >                self.assert_qmp(result, 'return', {})
> > 
> > We can "fix" it for probably all realistic cases by lowering the speed
> > of the block job significantly. It's still not fully fixed for all
> > theoretical cases, but the pattern of starting a block job that is
> > throttled to a low speed so it will keep running for the next part of
> > the test is very common.
> > 
> > Kevin
> > 
> > diff --git a/tests/qemu-iotests/030 b/tests/qemu-iotests/030
> > index 256b2bfbc6..31c028306b 100755
> > --- a/tests/qemu-iotests/030
> > +++ b/tests/qemu-iotests/030
> > @@ -243,7 +243,7 @@ class TestParallelOps(iotests.QMPTestCase):
> >              node_name = 'node%d' % i
> >              job_id = 'stream-%s' % node_name
> >              pending_jobs.append(job_id)
> > -            result = self.vm.qmp('block-stream', device=node_name, 
> > job_id=job_id, base=self.imgs[i-2], speed=512*1024)
> > +            result = self.vm.qmp('block-stream', device=node_name, 
> > job_id=job_id, base=self.imgs[i-2], speed=1024)
> >              self.assert_qmp(result, 'return', {})
> > 
> >          for job in pending_jobs:
> 
> Sounds good to me.  This would change the expected job completion
> time for the 2-4 MB images from 4-8 seconds to ~30-60 minutes,
> right?

I'm not sure about the granularity in which it really happens, but the
theory is that we have 2 MB of data in each image, so with 1024 bytes
per second, it should take 2048 seconds = ~34 minutes. And if we don't
manage to start and unthrottle four jobs within 34 minutes, we'll have
more problems that just that. :-)

> This is also a nice way to be sure (block-job-set-speed speed=0)
> is really working as expected.

speed=0 means unlimited, so this doesn't work for avoiding to make any
progress. It's what the next loop does to actually get the jobs
completed without waiting that long.

Kevin