On 04/18/2016 12:39 PM, Michal Židek wrote:
On 04/18/2016 12:22 PM, Lukas Slebodnik wrote:
> On (18/04/16 11:14), Michal Židek wrote:
>> On 04/18/2016 10:39 AM, Lukas Slebodnik wrote:
>>> On (02/12/15 17:10), Michal Židek wrote:
>>>> I saw some integration tests failures recently,
>>>> and I think there is a race condition between the
>>>> enumeration refresh timeout and the sleeps
>>>> after some operations that wait for this timeout.
>>>> SSSD fails to populate changes from LDAP in time
>>>> and some asserts can fail because of this.
>>>> So far I saw 4 tests to fail like this, which
>>>> is already quite a lot.
>>>> The attached patch modifies the timeout values
>>>> and hopefully removes the issue.
>>> I think I found alternative solution for intermittent
>>> failures of ADD REMOVE test with enumeration.
>> Not sure if this is safer than my patch. I made the
>> timeouts bigger on purpose, so that we avoid
>> problems with machines under heavy load.
>> You decided to go the opposite direction by
>> making one of the timeouts shorter.
>> That being said, maybe your patch is better,
>> because if it does not fail even under heavy
>> load it will make the tests shorter.
>> Once the CI is up we can try 20 test
>> runs with your patch and if they do not
>> fail I will give you an ACK.
>> If they fail, we can still fallback to my patch.
> We will need to find another solution.
> Increasing timeout is not acceptable from long term perspective.
> We need to have faster tests and not slower.
Then lets hope your patch will work :D
Your patch did not fix the issue.
It still fails cca 1 out of 5 times because
in the enumeration tests.
Maybe the timeout magic needs more mana
or there is some other issue which we do not see.