From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from eggs.gnu.org ([2001:4830:134:3::10]:49795)
	by lists.gnu.org with esmtp (Exim 4.71)
	(envelope-from <ehabkost@redhat.com>) id 1dYCoL-0007ea-N6
	for qemu-devel@nongnu.org; Thu, 20 Jul 2017 10:52:19 -0400
Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71)
	(envelope-from <ehabkost@redhat.com>) id 1dYCoI-00085o-Et
	for qemu-devel@nongnu.org; Thu, 20 Jul 2017 10:52:17 -0400
Received: from mx1.redhat.com ([209.132.183.28]:48956)
	by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32)
	(Exim 4.71) (envelope-from <ehabkost@redhat.com>) id 1dYCoI-00083i-50
	for qemu-devel@nongnu.org; Thu, 20 Jul 2017 10:52:14 -0400
Received: from smtp.corp.redhat.com (int-mx02.intmail.prod.int.phx2.redhat.com
	[10.5.11.12])
	(using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits))
	(No client certificate requested)
	by mx1.redhat.com (Postfix) with ESMTPS id AFAB67A17D
	for <qemu-devel@nongnu.org>; Thu, 20 Jul 2017 14:43:49 +0000 (UTC)
Date: Thu, 20 Jul 2017 11:43:40 -0300
From: Eduardo Habkost <ehabkost@redhat.com>
Message-ID: <20170720144340.GP2757@localhost.localdomain>
References: <20170720091902.22476-1-apahim@redhat.com>
	<20170720091902.22476-3-apahim@redhat.com>
	<87d18v1frq.fsf@dusky.pond.sub.org>
	<CALAZnb0pu7Q6BmLgggigThx0Xdy0NcO3EpRo3FY9Q6mSYY88Og@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CALAZnb0pu7Q6BmLgggigThx0Xdy0NcO3EpRo3FY9Q6mSYY88Og@mail.gmail.com>
Subject: Re: [Qemu-devel] [PATCH v3 2/3] qemu.py: include debug information
 on launch error
List-Id: <qemu-devel.nongnu.org>
List-Unsubscribe: <https://lists.nongnu.org/mailman/options/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=unsubscribe>
List-Archive: <http://lists.nongnu.org/archive/html/qemu-devel/>
List-Post: <mailto:qemu-devel@nongnu.org>
List-Help: <mailto:qemu-devel-request@nongnu.org?subject=help>
List-Subscribe: <https://lists.nongnu.org/mailman/listinfo/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=subscribe>
To: Amador Pahim <apahim@redhat.com>
Cc: Markus Armbruster <armbru@redhat.com>, qemu-devel@nongnu.org, kwolf@redhat.com, =?utf-8?B?THVrw6HFoQ==?= Doktor <ldoktor@redhat.com>, mreitz@redhat.com, Cleber Rosa <crosa@redhat.com>, Fam Zheng <famz@redhat.com>

On Thu, Jul 20, 2017 at 03:14:28PM +0200, Amador Pahim wrote:
> On Thu, Jul 20, 2017 at 1:58 PM, Markus Armbruster <armbru@redhat.com> wrote:
> > Amador Pahim <apahim@redhat.com> writes:
> >
> >> When launching a VM, if an exception happens and the VM is not
> >> initiated, it is useful to see the qemu command line that was executed
> >> and the output of that command.
> >>
> >> Before the patch:
> >>
> >>     >>> VM = qemu.QEMUMachine('../aarch64-softmmu/qemu-system-aarch64')
> >>     >>> VM.launch()
> >>     Traceback (most recent call last):
> >>       File "<stdin>", line 1, in <module>
> >>       File "qemu.py", line 137, in launch
> >>         self._post_launch()
> >>       File "qemu.py", line 121, in _post_launch
> >>         self._qmp.accept()
> >>       File "qmp/qmp.py", line 145, in accept
> >>         self.__sock, _ = self.__sock.accept()
> >>       File "/usr/lib64/python2.7/socket.py", line 206, in accept
> >>         sock, addr = self._sock.accept()
> >>     socket.timeout: timed out
> >>
> >> After the patch:
> >>
> >>     >>> VM = qemu.QEMUMachine('../aarch64-softmmu/qemu-system-aarch64')
> >>     >>> VM.launch()
> >>     Traceback (most recent call last):
> >>       File "<stdin>", line 1, in <module>
> >>       File "qemu.py", line 156, in launch
> >>         raise RuntimeError(msg)
> >>     RuntimeError: Error launching VM.
> >>     Original Exception:
> >>     Traceback (most recent call last):
> >>       File "qemu.py", line 138, in launch
> >>         self._post_launch()
> >>       File "qemu.py", line 122, in _post_launch
> >>         self._qmp.accept()
> >>       File "qmp/qmp.py", line 145, in accept
> >>         self.__sock, _ = self.__sock.accept()
> >>       File "/usr/lib64/python2.7/socket.py", line 206, in accept
> >>         sock, addr = self._sock.accept()
> >>     timeout: timed out
> >>     Command:
> >>     /usr/bin/qemu-system-aarch64 -chardev socket,id=mon,
> >>     path=/var/tmp/qemu-23958-monitor.sock -mon chardev=mon,mode=control
> >>     -display none -vga none
> >>     Output:
> >>     qemu-system-aarch64: No machine specified, and there is no default
> >>     Use -machine help to list supported machines
> >>
> >> Also, if the launch() faces an exception, the 'except' now will use args
> >> to fill the debug information. So this patch assigns 'args' earlier,
> >> assuring it will be available for the 'except'.
> >>
> >> Signed-off-by: Amador Pahim <apahim@redhat.com>
> >> Reviewed-by: Fam Zheng <famz@redhat.com>
> >> ---
> >>  scripts/qemu.py | 18 ++++++++++++++++--
> >>  1 file changed, 16 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/scripts/qemu.py b/scripts/qemu.py
> >> index f0fade32bd..2707ae7f75 100644
> >> --- a/scripts/qemu.py
> >> +++ b/scripts/qemu.py
> >> @@ -18,6 +18,7 @@ import os
> >>  import sys
> >>  import subprocess
> >>  import qmp.qmp
> >> +import traceback
> >>
> >>
> >>  class QEMUMachine(object):
> >> @@ -129,17 +130,30 @@ class QEMUMachine(object):
> >>          '''Launch the VM and establish a QMP connection'''
> >>          devnull = open('/dev/null', 'rb')
> >>          qemulog = open(self._qemu_log_path, 'wb')
> >> +        args = self._wrapper + [self._binary] + self._base_args() + self.args
> >>          try:
> >>              self._pre_launch()
> >> -            args = self._wrapper + [self._binary] + self._base_args() + self._args
> >>              self._popen = subprocess.Popen(args, stdin=devnull, stdout=qemulog,
> >>                                             stderr=subprocess.STDOUT, shell=False)
> >>              self._post_launch()
> >
> > Lifting the assignment out of the try makes sense, because exceptions
> > there are programming errors.
> >
> > Why are ._pre_launch() and ._post_launch() inside the try?
> 
> The _pre_launch()/_post_launch() are probably the main source of
> exceptions here and I assume we want to handle the exceptions coming
> from them.
> 
> The call to subprocess.Popen() is unlikely to raise exceptions. I
> mean, it's easy to trigger an 'OSError' exception providing a
> non-existent command to Popen, but once we provide something
> executable, we will not see an exception here if the command fails.
> The CalledProcessError is only raised by subprocess.check_call() or
> subprocess.check_output().
> 
> 

That's correct.  QEMU exiting won't cause an exception, it
doesn't matter the error code.  Failing to connect to QMP or
qtest sockets on _post_launch() will.

However, another purpose of the try/except block is to ensure we
clean up the process and log file in case _any_ exception happens
(programming error or not).  That's why everything after
open(self._qemu_log_path) is inside the try block.  We should be
careful to not break that.

I think this code really needs to be cleaned up.  The existing
try/except block has a huge number of responsibilities:
1) Ensuring we kill the process in case _post_launch() raises an
   exception.
2) Ensuring we load the log file even if Popen() or _post_launch()
   raise an exception.
3) Ensuring we remove the monitor socket if anything after
   _pre_launch() raises an exception (through _post_shutdown())
4) Ensuring the log file is deleted if anything after
   open(_qemu_log_path) raises an exception (through
   _post_shutdown())
5) Ensuring _post_shutdown() is called if anything after
   _pre_launch() raises an exception (e.g. QEMUQTestMachine
   deletes the socket created by _pre_launch() inside
   _post_shutdown()
6) (New, added by this patch) Generating a more useful exception
   message if something (what exactly?) fails.

Some of the cases above are just programming errors, but it's
still nice to do proper cleanup in case of programming errors.
e.g.: if _base_args() raise an exception, we can ensure we delete
the log file anyway (this patch breaks that).

If we don't clean this up, future cleanups (e.g. moving
_pre_launch() outside the try block) would easily break one or
more responsibilities above, and this would be very easy to miss
on patch review.


> >
> >>          except:
> >
> > Why do we catch all exceptions, and not just
> > subprocess.CalledProcessError?

See above.

> >
> >> +            self._load_io_log()
> >>              if self.is_running():
> >
> > I suspect we can't get here only for "odd" exceptions, not for
> > subprocess.CalledProcessError.

It depends on what you mean by "odd".

> >
> >>                  self._popen.kill()
> >>                  self._popen.wait()
> >> -            self._load_io_log()
> >> +            else:
> >> +                exc_type, exc_value, exc_traceback = sys.exc_info()
> >> +                msg = ('Error launching VM.\n'
> >> +                       'Original Exception: \n%s'
> >> +                       'Command:\n%s\n'
> >> +                       'Output:\n%s\n' %
> >> +                       (''.join(traceback.format_exception(exc_type,
> >> +                                                           exc_value,
> >> +                                                           exc_traceback)),
> >> +                        ' '.join(args),
> >> +                        self._iolog))
> >> +                self._post_shutdown()
> >> +                raise RuntimeError(msg)
> >>              self._post_shutdown()
> >>              raise
> >
> > Why do we print a traceback when subprocess.CalledProcessError?
> 
> subprocess.CalledProcessError will not happen with subprocess.Popen,
> even if the command fails. But if the virtual machine does not start,
> we will, for example, see a 'socket.timeout' exception coming from
> _post_launch() when it tries to connect to the qemu monitor.

Exactly.

-- 
Eduardo