Hello Andrei,

first, can you please confirm you are using the `keep_guest_running` to minimize the environment differences.

Then to your reproducer, I'm not sure how to trigger it. I use a modified `boot` test where I run the pre-process twice with modified params. This way I get your "Old vm is destroyed, but, it is still present in env." message, but this message only means the instance is reused. It does not mean it is used to boot the machine. The important part is that `start_vm` is set to `True` which means that around line `173` the old `params` are replaced with the new fresh ones so at least in my understanding it should work properly. Anyway mistakes happen so would you please give me a simple reproducer or at least more info about where this does not work.

Regards,
Lukáš


Dne 2.2.2017 v 12:53 Andrei Stepanov napsal(a):
Lukáš Hi!

I added next debug code:

diff --git a/virttest/env_process.py b/virttest/env_process.py
index d05976e..64c78ac 100644
--- a/virttest/env_process.py
+++ b/virttest/env_process.py
@@ -162,6 +162,12 @@ def preprocess_vm(test, params, env, name):
                     vm.devices = None
                     start_vm = True
                     old_vm.destroy(gracefully=gracefully_kill)
+                    for key1 in env.keys():
+                        vm1 = env[key1]
+                        if not isinstance(vm1, virt_vm.BaseVM):
+                            continue
+                        if vm1.name <http://vm1.name> == old_vm.name
<http://old_vm.name>:
+                            logging.warn("Old vm is destroyed, but, it
is still present in env.")
                     update_virtnet = True

     if start_vm:



Than logs have message:  "Old vm is destroyed, but, it is still present
in env."

So, VM was destroyed, VM object is still in env.

Let's go to line 690 in the same file:

        if vm.name <http://vm.name> not in requested_vms:

VM for next test has the same name.

As a result: next test uses VM object from previous test.  VM is started
using params from previous test.

And this behavior is serious bug.


On Wed, Feb 1, 2017 at 3:05 PM, Lukáš Doktor <[email protected]
<mailto:[email protected]>> wrote:

    Hello Andrei,

    if this happens than there is something really wrong because Avocado
    should re-create the VM for 2 reasons:

    1) by default VMs are not shared between tests (can be enabled in
    cfg by setting `keep_guest_running = True` in `vt.setup` section)
    2) when the params of the existing VM and the current params are
    different, it's recreated.

    The (2) is checked in `virttest.env_process` on line `159` where it
    executes `vm.needs_restart`. The implementation of this function is
    defined mainly in `virttest.virt_vm` and unless overridden it uses
    the `virttest.virt_vm.make_create_command` to compare the original
    and the new command line to create the VM. If they are the same it
    reuses the VM (when (1) is enabled), otherwise it destroys the old
    VM and creates a new one.

    The question is how different your machines are. The
    `make_create_command` does not compares the extra dynamic stuff like
    migration. More info about this can be found in
    `virttest.qemu_vm.create()` function (if using qemu as a backend).

    Would you please (publicly or in private) share more details I might
    be able to identify why the machine is not being re-created.

    Regards,
    Lukáš

    Dne 31.1.2017 v 18:15 Andrei Stepanov napsal(a):

        Hi.

        It seems that avocado-vt has a serious bug.

        Test case: run a few avocado-vt tests in a bunch. For example
        two tests.
        Test1 starts just right after Test2.

        Test1.
        Test2.

        Test1 & Test2 use the same name for VM in cartesian configs:
        vms = guest

        Other options for VM() objects are different, for example port
        VNC port,
        some device config, etc....

        The problem is that: KVM from Test2 uses VM() object that was
        created
        for Test1.

        For Test2:
        virttest/env_process.py:

        def preprocess_vm(test, params, env, name):

              vm = env.get_vm(name)  <--- returns VM that was created
        for Test1.
              create_vm == False

        It can be fixed by:

        diff --git a/virttest/env_process.py b/virttest/env_process.py
        index d05976e..7c08df4 100644
        --- a/virttest/env_process.py
        +++ b/virttest/env_process.py
        @@ -687,9 +687,8 @@ def preprocess(test, params, env):
                 vm = env[key]
                 if not isinstance(vm, virt_vm.BaseVM):
                     continue
        -        if vm.name <http://vm.name> <http://vm.name> not in
        requested_vms:
        -            vm.destroy()
        -            del env[key]
        +        vm.destroy()
        +        del env[key]

             if (params.get("auto_cpu_model") == "yes" and
                     vm_type == "qemu"):


        Could you please confirm that bug exists in real?




Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to