[Libvirt-cim] [PATCH] REDSAP: Fix double exception possibilty

[John Ferlan]

On 09/09/2013 07:12 AM, Viktor Mihajlovski wrote:
> On 09/06/2013 03:46 PM, John Ferlan wrote:
>> Test failed with the following:
>>
>> --------------------------------------------------------------------
>> KVMRedirectionSAP - 01_enum_KVMredSAP.py: FAIL
>> ERROR         - Failed to enumerate the class of KVM_KVMRedirectionSAP
>> ERROR         - Exception details: Got more than one record for:
>> test_kvmredsap_dom
>> ERROR         - Exception details: 'ElementName' Value Mismatch,
>> Expected 5988:-1, Got 5988:0
>> ERROR         - Exception: Failed to verify information for the
>> defined dom:test_kvmredsap_dom
>> --------------------------------------------------------------------
>>
>> There are two exceptions listed because the 'enum_redsap()' method was
>> perusing a list, finding a match, declaring success, and continuing to
>> peruse the list.  Then found another match declared an exception and
>> returned
>> with status = PASS.
>>
>> The caller just checked the status before continuing.
>>
>> This patch doesn't resolve the underlying cause, but does avoid the
>> call to
>> 'verify_redsap_values()' which will also fail...
>> ---
>>
>> NOTE: I was able to reproduce the initial exception if I run this test
>> very quickly two times in succession. I debated adding a 'sleep(5)' (or
>> similar) prior to the end of the test to ensure whatever teardown was
>> not occurring in a timely manner (by the networking code) could happen,
>> but I believe that wouldn't necessarily fix the problem - just make it
>> less likely to happen. The resource in question is a specific port in the
>> underlying VNC technology. That code may have built in 'safeguards' to
>> help
>> with throttling, e.g. reuse a recently used port thus make it
>> unreliable for
>> the test to rely on getting the "first" port. Since the test isn't
>> designed
>> to run twice in a row, I don't think it's a real issue...
>>
>>   .../libvirt-cim/cimtest/KVMRedirectionSAP/01_enum_KVMredSAP.py   | 9
>> ++++-----
>>   1 file changed, 4 insertions(+), 5 deletions(-)
>>
> The patch fixes the previously incorrect error handling, so I would
> suggest to push it anyway.
> The failure doesn't seem to happen because of a lingering socket but
> rather becasue the guest wasn't destroyed and/or undefined (returned
> port is 0). Did this failure ever reoccur *after* you have applied your
> change?
> 
> 

Yes - unfortunately. Although not very repeatable.  Sometimes running
twice in a row works and sometimes it doesn't.   Chasing after those
kind of timing problems is never easy.  When I worked my way through the
API's, I ended up in the VNC port allocation routines - I think
get_vnc_sessions() - but I was never quite sure exactly where the fault
was and chasing it seemed to be a too time consuming task.

John