Found the JIRA: https://issues.apache.org/jira/browse/BEAM-2506

TL;DR  TestPipeline.run() would be a noop and only in the @Rule teardown it
would run a single pipeline with all the tests in it.

There are two big issues I can think of:

 - Which test failed? Eugene comments on the JIRA that PAssert site capture
might make this acceptable
 - Some tests actually need to run a pipeline and interact with the job,
like PubsubReadIT and PubsubJsonIT. This is fine - they don't really need
TestPipeline at all. But they do need the automatic testing
PipelineOptions, so we need to build that path.

Kenn

On Fri, Jun 29, 2018 at 7:23 PM Reuven Lax <[email protected]> wrote:

> My run spent 5 hours and 40 minutes in the queue. I could easily see
> enough people trying to run these tests to cause days of waiting time.
>
>
>
> On Fri, Jun 29, 2018 at 3:57 PM Jason Kuster <[email protected]>
> wrote:
>
>> There's been a fairly long-standing discussion regarding merging
>> ValidatesRunner tests in the same class together into one pipeline; this
>> would give us more breathing room in terms of Dataflow job quota and may
>> allow us to run multiple ValidatesRunner suites at the same time.
>>
>> On Fri, Jun 29, 2018 at 3:49 PM Lukasz Cwik <[email protected]> wrote:
>>
>>> I believe it was increased so we could run batch and streaming tests in
>>> parallel but not enough to run multiple Dataflow VR runs in parallel.
>>>
>>> On Fri, Jun 29, 2018 at 3:43 PM Reuven Lax <[email protected]> wrote:
>>>
>>>> Didn't we recently have the quota increased?
>>>>
>>>> On Fri, Jun 29, 2018, 3:40 PM Lukasz Cwik <[email protected]> wrote:
>>>>
>>>>> Dataflow VR only runs one at a time due to Dataflow job quota capacity.
>>>>>
>>>>> On Fri, Jun 29, 2018 at 3:39 PM Andrew Pilloud <[email protected]>
>>>>> wrote:
>>>>>
>>>>>> Looks like it is only running 1 instance of Dataflow ValidatesRunner at
>>>>>> a time, and the job takes 2+ hours. The queue is a little backed up.
>>>>>> https://builds.apache.org/job/beam_PostCommit_Java_ValidatesRunner_Dataflow_Gradle_PR/
>>>>>>
>>>>>> Andrew
>>>>>>
>>>>>> On Fri, Jun 29, 2018 at 3:36 PM Lukasz Cwik <[email protected]> wrote:
>>>>>>
>>>>>>> I think the jobs aren't being scheduled or the status is failing to
>>>>>>> be sent back to Github.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On Fri, Jun 29, 2018 at 3:22 PM Reuven Lax <[email protected]> wrote:
>>>>>>>
>>>>>>>> Well it's now moving on 2.5 hours, and the tests still haven't even
>>>>>>>> started to execute. Something seems very wrong.
>>>>>>>>
>>>>>>>> On Fri, Jun 29, 2018 at 3:14 PM Andrew Pilloud <[email protected]>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> I don't know what is going on, but I saw the same high latency in
>>>>>>>>> jobs right after filtered pre-commit was turned on for the first 
>>>>>>>>> time. (I
>>>>>>>>> don't have anything to suggest it is related, just a memorable time it
>>>>>>>>> happened.) I'm also noticing the Jenkins UI is a bit laggy.
>>>>>>>>>
>>>>>>>>> Andrew
>>>>>>>>>
>>>>>>>>> On Fri, Jun 29, 2018 at 3:09 PM Reuven Lax <[email protected]>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>> In addition to they delay in the trigger phrases, I'm seeing
>>>>>>>>>> multi-hour (!) delays in running tests.
>>>>>>>>>>
>>>>>>>>>> For example: https://github.com/apache/beam/pull/5545
>>>>>>>>>>
>>>>>>>>>> I triggered the Dataflow ValidatesRunner tests. The check showed
>>>>>>>>>> up as yellow with no "Details" link, which generally means it had 
>>>>>>>>>> not yet
>>>>>>>>>> been scheduled on Jenkins. It stayed like this for two hours, never 
>>>>>>>>>> even
>>>>>>>>>> starting to run. Apparently there were a ton of idle Jenkins 
>>>>>>>>>> executors at
>>>>>>>>>> the time, so it's not that all our Jenkins executors were busy. As 
>>>>>>>>>> of now,
>>>>>>>>>> it still has not started running.
>>>>>>>>>>
>>>>>>>>>> Does anybody have any idea what's going on here?
>>>>>>>>>>
>>>>>>>>>> Reuven
>>>>>>>>>>
>>>>>>>>>
>>
>> --
>> -------
>> Jason Kuster
>> Apache Beam / Google Cloud Dataflow
>>
>> See something? Say something. go/jasonkuster-feedback
>> <https://goto.google.com/jasonkuster-feedback>
>>
>

Reply via email to