Re: [Avocado-devel] avocado-vt hangs with: Creating new, empty env file 2

[Lukáš Doktor]

Hello Andrei,





```
diff --git a/virttest/qemu_monitor.py b/virttest/qemu_monitor.py
index 8ff872b..48e7ef2 100644
--- a/virttest/qemu_monitor.py
+++ b/virttest/qemu_monitor.py

     else:
         raise MonitorConnectError(monitor_name)

+class VM(object):
+    def __init__(self, name):
+        self.name = name

 class Monitor:

@@ -188,7 +191,7 @@ class Monitor:

         :raise MonitorConnectError: Raised if the connection fails
         """
-        self.vm = vm
+        self.vm = VM(vm.name)
         self.name = name
         self.filename = filename
         self._lock = threading.RLock()
```



Lukáš

Dne 10.2.2017 v 14:05 Andrei Stepanov napsal(a):

this was a tough one, but we have circular dependency in pickled files again (I remember cleaning probably the same one ~5 years ago). An ugly workaround is: @@ -167,6 +167,9 @@ def wait_for_create_monitor(vm, monitor_name, monitor_params, timeout): but I'll come up with a cleaner solution and send a bugfix then... In the mean time, can you verify this fix works for you as well?

Hi

For a long time SpiceQE from time to time had this error:

Exception thrown while loading env
(AttributeError("'VM' object has no attribute 'name'",), <class
virttest.qemu_monitor.HumanMonitor at 0x3992a78>, (<virttest.qemu_vm.VM
object at 0x3e25950>, 'hmp1',
'/var/tmp/avocado_bFZBC6/monitor-hmp1-20170210-134859-mVznyQoq', False))
Creating new, empty env file 2

After which Avocado test process hangs forever. It was very hard to explain.

I finally found the cause.

It is easy reproduce even with "boot" test.

1. Change boot.py to raise any exception at the beginning of run()
function: "raise E...."
2. Run this test with:

avocado run --show-job-log boot --vt-extra-params 'kill_vm=no'
'kill_vm_on_error=no'

This test will stay active all time no matter what.

The problem is: there is active thread:

[<_MainThread(MainThread, started 140040446437184)>,
<Thread(tail_thread_MwKAuDx9_QEMU_AUDIO, started 140040077772544)>]

tail_thread_xxxxxx thread is created at:

./lib/python2.7/site-packages/aexpect/client.py:624:
 self.tail_thread = threading.Thread(target=self._tail,

This is a cause that Avocado process stays active.

It reaches exit point:

--Return--

/mnt/tests/spice/qe-tests/virtualenv/bin/avocado(83)<module>()->None

-> sys.exit(app.run())
(Pdb)

But running thread doesn't allow to exit:

# strace -f -p 22879
Process 22879 attached with 2 threads
[pid 22879] select(10, [9], [], [], {0, 17975} <unfinished ...>
[pid 22353] futex(0x45e0760, FUTEX_WAIT_PRIVATE, 0, NULL <unfinished ...>
[pid 22879] <... select resumed> )      = 0 (Timeout)
[pid 22879] select(10, [9], [], [], {0, 50000}) = 0 (Timeout)
[pid 22879] select(10, [9], [], [], {0, 50000}) = 0 (Timeout)
[pid 22879] select(10, [9], [], [], {0, 50000}) = 0 (Timeout)
[pid 22879] select(10, [9], [], [], {0, 50000}) = 0 (Timeout)


Okay. We can set "kill_vm" to "yes".  But...... There is something wrong
with:

kill_vm = no + kill_vm_on_error=no

signature.asc
Description: OpenPGP digital signature