[libvirt] [PATCH 2/2] qemu: emit 'defined' event after reverted to snapshot

[John Ferlan]

On 12/15/2015 04:00 AM, Dmitry Andreev wrote:
> 
> 
> On 14.12.2015 23:10, John Ferlan wrote:
>>
>>
>> On 12/09/2015 03:29 AM, Dmitry Andreev wrote:
>>> Config file is changed. VIR_DOMAIN_EVENT_DEFINED should be emitted
>>> ---
>>>   src/qemu/qemu_driver.c | 5 +++++
>>>   1 file changed, 5 insertions(+)
>>>
>>> diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c
>>> index ae1d8e7..b32172a 100644
>>> --- a/src/qemu/qemu_driver.c
>>> +++ b/src/qemu/qemu_driver.c
>>> @@ -15293,6 +15293,7 @@
>>> qemuDomainRevertToSnapshot(virDomainSnapshotPtr snapshot,
>>>       virDomainSnapshotObjPtr snap = NULL;
>>>       virObjectEventPtr event = NULL;
>>>       virObjectEventPtr event2 = NULL;
>>> +    virObjectEventPtr define_event = NULL;
>>>       int detail;
>>>       qemuDomainObjPrivatePtr priv;
>>>       int rc;
>>> @@ -15401,6 +15402,9 @@
>>> qemuDomainRevertToSnapshot(virDomainSnapshotPtr snapshot,
>>>           config = virDomainDefCopy(snap->def->dom, caps,
>>> driver->xmlopt, true);
>>>           if (!config)
>>>               goto endjob;
>>> +        define_event = virDomainEventLifecycleNewFromObj(vm,
>>> +                                     VIR_DOMAIN_EVENT_DEFINED,
>>> +                                    
>>> VIR_DOMAIN_EVENT_DEFINED_FROM_SNAPSHOT);
>>>       }
>>>
>>>       switch ((virDomainState) snap->def->state) {
>>> @@ -15627,6 +15631,7 @@
>>> qemuDomainRevertToSnapshot(virDomainSnapshotPtr snapshot,
>>>       } else if (snap) {
>>>           snap->def->current = false;
>>>       }
>>> +    qemuDomainEventQueue(driver, define_event);
>>
>> I think this can go right after the generation of the event. That way it
>> can come "in order"...
>>
>> Following the rest of the logic, event can be queued as it happens and
>> then "reused" in a few instances prior to event2 being generated. Then
>> at cleanup the most recent event and event2 get queued.
> 
> As I can see from the code - 'stopped' event is the only event that is
> queued as it happens. The VM state is changed to STOPPED inside the
> switch for few cases. If you queue the 'defined' event before the
> switch (or right after the generation) the order will be 'defined' ->
> 'stopped' -> event -> event2 when the natural order is 'stopped' before
> 'defined'.
> 
> API client will get 'defined' event for VM with cached running state
> before you send him 'stopped' event.
> 

hmm... true. Following the transitions is a bit of a challenge. Not part
of this issue, but I think there's also a bug in there now.  I'll send a
different patch on that...

As for this new event, I think there is an odd path after the comment
about transitions 1, 4, 7, where the STOP event is created, but before
the "if (config)", we goto cleanup where the DEFINED event could be
generated before the STOP event gets queued, but the redefinition never
occurs.

So, I think perhaps, prior to the "goto cleanup;" there could be
virObjectUnref(define_event); define_event = NULL; similar to what
happens in the code after the comments about transitions 2, 5, 8 when
the event is cleared.

Alternatively, whenever there is:

    if (config)
        virDomainObjAssignDef(...)

that's when the "define_event" could be queued, especially since once
that ObjAssignDef code runs - regardless of what other paths are taken
in the code, the domain is redefined.

John

>>
>> If you're OK with me moving that, then I can do so and push the two
>> patches.
>>
>> John
>>>       if (event) {
>>>           qemuDomainEventQueue(driver, event);
>>>           qemuDomainEventQueue(driver, event2);
>>>