[Kimchi-devel] [PATCH V2 6/6] Fix Kimchi vm tests according to new messages sent from backend
Rodrigo Trujillo
rodrigo.trujillo at linux.vnet.ibm.com
Wed Feb 26 20:55:27 UTC 2014
Hi Cristian,
your analyses about the flow and root of the problem are correct.
However I think you are only seeing the case where the flow just works,
you did not took in account that
rollbacks are used in case of errors/crashes/fails in the middle of
test. For instance:
1- inst.vms_create(params_1)
2- inst.vms_create(params_2)
3- vms = inst.vms_get_list()
4- inst.vm_start('kimchi-vm1')
5- info = inst.vm_lookup('kimchi-vm1')
...
- If the code fails/breaks in (3), then (1) and (2) must be 'rollbacked'.
- If the code fails/breaks in (5) , then (1), (2) and (4) must be
'rollbacked' ... (not really sure if 4 is needed, but the code was
already there)
For these reasons, I cannot remove the rollback after create the vms,
otherwise, they will remains in the system.
If the code breaks in (4), while starting the vm (receive an exception),
how would you remove the vms created above (1)(2) ? I asked myself this.
The option I could think know is to add try/except in the code, but, the
with/rollback was implemented exactly for this (or to avoid this).
I also tried to split with/rollback, but this does not work.
On 02/26/2014 04:52 PM, Crístian Viana wrote:
> AFAIU, here is what happens in the method "test_vm_update": the VMs
> "kimchi-vm1" and "kimchi-vm2" are created; the commands to *delete
> those VMS by their names* are deferred in the transaction; *the VM
> "kimchi-vm1" is renamed to "?e?-??"*; the command to *delete "?e?-??"
> by its name* is deferred in the transaction.
>
> Notice how the VM is deleted twice. Both "kimchi-vm1" and "?e?-??" are
> the same VMs, with different names in different moments. The command
> "vm_delete" fails for two reasons: 1) the VM "kimchi-vm1" does not
> exist anymore when "vm_delete" is executed (NotFoundError); 2) even if
> it did, you cannot delete a VM twice (also NotFoundError).
> So the correct solution would be to not delete the VM twice, and keep
> track of the names they have during the tests. If the VMs are renamed,
> the deferred commands will fail in the future when they execute.
>
> The same analysis can be done to work out a better solution for the
> "vm_stop" case. That function "_rollback_wrapper" looks very hacky and
> there should be a way not to need it.
>
> Am 26-02-2014 16:05, schrieb Rodrigo Trujillo:
>> With new backend messaging system, when a VM is not found the backend
>> raises an exception which was breaking Rollback functions. This patch
>> fixes this problem.
>>
>> Signed-off-by: Rodrigo Trujillo<rodrigo.trujillo at linux.vnet.ibm.com>
>> ---
>> tests/test_model.py | 29 +++++++++++++++++++++++------
>> 1 file changed, 23 insertions(+), 6 deletions(-)
>>
>> diff --git a/tests/test_model.py b/tests/test_model.py
>> index 74e2424..d48377d 100644
>> --- a/tests/test_model.py
>> +++ b/tests/test_model.py
>> @@ -547,14 +547,17 @@ class ModelTests(unittest.TestCase):
>> params_2 = {'name': 'kimchi-vm2', 'template': '/templates/test'}
>> inst.vms_create(params_1)
>> inst.vms_create(params_2)
>> - rollback.prependDefer(inst.vm_delete, 'kimchi-vm1')
>> - rollback.prependDefer(inst.vm_delete, 'kimchi-vm2')
>> + rollback.prependDefer(self._rollback_wrapper, inst.vm_delete,
>> + 'kimchi-vm1')
>> + rollback.prependDefer(self._rollback_wrapper, inst.vm_delete,
>> + 'kimchi-vm2')
>>
>> vms = inst.vms_get_list()
>> self.assertTrue('kimchi-vm1' in vms)
>>
>> inst.vm_start('kimchi-vm1')
>> - rollback.prependDefer(inst.vm_stop, 'kimchi-vm1')
>> + rollback.prependDefer(self._rollback_wrapper, inst.vm_stop,
>> + 'kimchi-vm1')
>>
>> info = inst.vm_lookup('kimchi-vm1')
>> self.assertEquals('running', info['state'])
>> @@ -569,7 +572,8 @@ class ModelTests(unittest.TestCase):
>> 'kimchi-vm1', {'name': 'kimchi-vm2'})
>> inst.vm_update('kimchi-vm1', params)
>> self.assertEquals(info['uuid'], inst.vm_lookup(u'?e?-??')['uuid'])
>> - rollback.prependDefer(inst.vm_delete, u'?e?-??')
>> + rollback.prependDefer(self._rollback_wrapper, inst.vm_delete,
>> + u'?e?-??')
>>
>> @unittest.skipUnless(utils.running_as_root(), 'Must be run as root')
>> def test_network(self):
>> @@ -758,6 +762,17 @@ class ModelTests(unittest.TestCase):
>> inst.task_lookup(taskid)['message'])
>> self.assertEquals('failed', inst.task_lookup(taskid)['status'])
>>
>> + # This wrapper function is needed due to the new backend messaging in
>> + # vm model. vm_stop and vm_delete raise exception if vm is not found.
>> + # These functions are called after vm has been deleted if test finishes
>> + # correctly, then NofFoundError exception is raised and rollback breaks
>> + def _rollback_wrapper(self, func, vmname):
>> + try:
>> + func(vmname)
>> + except NotFoundError:
>> + # VM has been deleted already
>> + return
>> +
>> @unittest.skipUnless(utils.running_as_root(), 'Must be run as root')
>> def test_delete_running_vm(self):
>> inst = model.Model(objstore_loc=self.tmp_store)
>> @@ -769,10 +784,12 @@ class ModelTests(unittest.TestCase):
>>
>> params = {'name': u'ki-??hi--??', 'template': u'/templates/test'}
>> inst.vms_create(params)
>> - rollback.prependDefer(inst.vm_delete, u'ki-??hi--??')
>> + rollback.prependDefer(self._rollback_wrapper, inst.vm_delete,
>> + u'ki-??hi--??')
>>
>> inst.vm_start(u'ki-??hi--??')
>> - rollback.prependDefer(inst.vm_stop, u'ki-??hi--??')
>> + rollback.prependDefer(self._rollback_wrapper, inst.vm_stop,
>> + u'ki-??hi--??')
>>
>> inst.vm_delete(u'ki-??hi--??')
>>
>
>
>
> _______________________________________________
> Kimchi-devel mailing list
> Kimchi-devel at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/kimchi-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/kimchi-devel/attachments/20140226/1461f57a/attachment.html>
More information about the Kimchi-devel
mailing list