[Freeipa-devel] [PATCH] Workaround for trac N 5348

[Martin Basti]

On 10/08/2015 02:41 PM, Martin Kosek wrote:
> On 10/08/2015 02:06 PM, Martin Basti wrote:
>>
>> On 10/08/2015 11:18 AM, Jan Pazdziora wrote:
>>> On Thu, Oct 08, 2015 at 11:12:37AM +0200, Oleg Fayans wrote:
>>>>> When the ticket is addressed and these workarounds are no longer
>>>>> needed -- what is our process for finding these workarounds and
>>>>> reverting them, so that the tests test the real, expected behaviour?
>>>> As per discussion with Martin Basti, it was decided that this workaround
>>>> will only be applied to the current 4-2 branch, not to the upstream. In
>>> That sounds like a reasonable plan for this issue.
>>>
>>>> upstream the issue itself will (supposedly) be solved
>>> Except currently it's not, so (IIUIC) you will keep having
>>> nondeterministic failures in master.
>>>
>>> I was mostly interested in the general approach that we have to
>>> workarounds -- how do we track them, how do we make sure they don't
>>> stick in tests forever, even after the issue was already properly
>>> addressed.
>>>
>> I'm not sure if there is a formal process how to work with workarounds.
>>
>> Usually, we open ticket, push workaround there, and leave ticket opened until
>> the issue is fixed.
>> If there is no time to do proper fix and workaround works well we close ticket
>> and open new ticket in further milestones.
>>
>> I do not remember if we have a similar issue with test workaround in past.
> Can we anyhow utilize "wait_for_dns" knob we added for making DNS tests reliable?
No,

I already do that when records in CI test are created, there is polling.

The first part of oleg's workaround has nothing common with timing 
issue, only restart of named process will help

The second part, I do not know why there is 1sec delay needed, because 
presence of signed records was detected in step before, so DNSKEY record 
must be available, and probably this is caused by named internals, that 
DNSKEY record is available later than signed records of zone.
I don think so that extending testing for all types of DNSSEC records is 
worth it and 1sec is enough for bind to be ready.

Martin