With a QEMU bug, it can happen that the QEMU process doesn't react to a
'quit' QMP command. If we got an exception during previous QMP
communication (e.g. iotests Timeout expiring), we could also be in an
inconsistent state where after sending 'quit' we immediately read an old
response and close the socket even though the 'quit' command wasn't
processed yet. Both cases would lead to a hanging test.

Fix this by waiting for the QEMU process to exit after sending 'quit'
with a timeout, and if it doesn't happen within three seconds, send
SIGKILL.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 python/qemu/machine.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/python/qemu/machine.py b/python/qemu/machine.py
index 183d8f3d38..c837ee8723 100644
--- a/python/qemu/machine.py
+++ b/python/qemu/machine.py
@@ -358,6 +358,7 @@ class QEMUMachine(object):
                     if not has_quit:
                         self._qmp.cmd('quit')
                     self._qmp.close()
+                    self._popen.wait(timeout=3)
                 except:
                     self._popen.kill()
             self._popen.wait()
-- 
2.20.1

On 3/13/20 9:36 AM, Kevin Wolf wrote:
> With a QEMU bug, it can happen that the QEMU process doesn't react to a
> 'quit' QMP command. If we got an exception during previous QMP
> communication (e.g. iotests Timeout expiring), we could also be in an
> inconsistent state where after sending 'quit' we immediately read an old
> response and close the socket even though the 'quit' command wasn't
> processed yet. Both cases would lead to a hanging test.
> 
> Fix this by waiting for the QEMU process to exit after sending 'quit'
> with a timeout, and if it doesn't happen within three seconds, send
> SIGKILL.
> 
> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> ---
>   python/qemu/machine.py | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/python/qemu/machine.py b/python/qemu/machine.py
> index 183d8f3d38..c837ee8723 100644
> --- a/python/qemu/machine.py
> +++ b/python/qemu/machine.py
> @@ -358,6 +358,7 @@ class QEMUMachine(object):
>                       if not has_quit:
>                           self._qmp.cmd('quit')
>                       self._qmp.close()
> +                    self._popen.wait(timeout=3)
>                   except:
>                       self._popen.kill()
>               self._popen.wait()
> 

Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>

On 3/13/20 4:36 AM, Kevin Wolf wrote:
> With a QEMU bug, it can happen that the QEMU process doesn't react to a
> 'quit' QMP command. If we got an exception during previous QMP
> communication (e.g. iotests Timeout expiring), we could also be in an
> inconsistent state where after sending 'quit' we immediately read an old
> response and close the socket even though the 'quit' command wasn't
> processed yet. Both cases would lead to a hanging test.
> 
> Fix this by waiting for the QEMU process to exit after sending 'quit'
> with a timeout, and if it doesn't happen within three seconds, send
> SIGKILL.
> 
> Signed-off-by: Kevin Wolf <kwolf@redhat.com>

Reviewed-by: John Snow <jsnow@redhat.com>

> ---
>  python/qemu/machine.py | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/python/qemu/machine.py b/python/qemu/machine.py
> index 183d8f3d38..c837ee8723 100644
> --- a/python/qemu/machine.py
> +++ b/python/qemu/machine.py
> @@ -358,6 +358,7 @@ class QEMUMachine(object):
>                      if not has_quit:
>                          self._qmp.cmd('quit')
>                      self._qmp.close()
> +                    self._popen.wait(timeout=3)
>                  except:
>                      self._popen.kill()
>              self._popen.wait()
> 

-- 
—js