Well, this is a bit of a gray area inherent to distributed teams. What you do 
Christian is great, don't change that :). However expect to run into 
frustrating situations now and then.

In an ideal world all tests would pass, we would not depend on an internet 
connection and a build would not take more than 5 minutes. And yes, we would 
not have snapshot dependencies! I ran into both compile and test issues 
countless times because a dependent snapshot changed under me, and spent hours 
investigating stuff like that. Well, that's much better now. Sometimes after 
changing in core I run all the tests, making sure there are no adverse effects 
on other components. By the time my tests are done, there may be 2-3 newer 
commits on the trunk and have to start over. Doing things by the book, doesn't 
always scale and all in all things are not bad at all.

That said, I think we could do a better job with snapshots. What about taking 
turns on monitoring Hudson and giving a higher prio to fixing issues there? 
Another solution I toyed with a while ago is moving to a distributed build 
using GrigGain or something like that to hopefully cut the build time to a 
fraction. Anybody with some spare cycles?

Hadrian 


On Jan 19, 2010, at 4:31 PM, Christian Schneider wrote:

> Hi Claus,
> 
> I understand that it is not possible to run the unit tests for a snapshot 
> build at the moment. Of course they can only be switched on if they run 
> reliably.
> 
> I normally run the tests before a checkin (or before I submit a patch in case 
> of camel) to make sure that I do not break anything. If some tests fail or 
> worse fail only sometimes it is extremly difficult to decide if the failure 
> was because of my change. Any idea how to handle this?
> 
> Greetings
> 
> Christian
> 
> 
> Am 19.01.2010 09:13, schrieb Claus Ibsen:
>> 
>> No I think its great as it is.
>> Camel have 4600+ unit test and still counting. It takes like 3-4h on
>> the hudson server to complete a test and it can break somewhere in for
>> some strange reasons. We see this on the Fuse CI platforms where we
>> got 10+ platforms to run Camel every single day.
>> 
>> At Apache people want to latest code in SNAPSHOT which also can help
>> when they want to verify a fix committed fixed their issue.
>> If we only deploy if 100% green then SNAPSHOT could easily turn weeks
>> old or longer.
>> 
>> The current setup is great as it is IMHO.
>> 
>> If you want SNAPSHOT which have 100% green they you can use the
>> SNAPSHOTs from the fuse repos.
>> 
>> 
>>   
>>> Greetings
>>> 
>>> Christian
>>> 
>>> 
>>> Am 19.01.2010 03:10, schrieb Willem Jiang:
>>>     
>>>> I found this issue weeks ago, it looks like we just use the Apache Hundson
>>>> to deploy the camel kits now.
>>>> 
>>>> And Gert said we can create two Hudson jobs (one for deployment and other
>>>> for running test) for Camel.
>>>> 
>>>> It's time to get them started :)
>>>> 
>>>> Willem
>>>> 
>>>> chrislovecnm wrote:
>>>>       
>>>>> A quick question
>>>>> 
>>>>> Why are the tests skipped during the CI Hudson builds? I have filed three
>>>>> bugs today that deal directly with unit tests not building or failing.
>>>>> IMHO
>>>>> these may have been caught if the test where not skipped.  Or a hudson
>>>>> build
>>>>> could be setup that nightly runs the unit tests?
>>>>> 
>>>>> I am sure that there is a reason why the tests are off, but I was just
>>>>> curious.  Plus a bit frustrated that the build is broken :)
>>>>> 
>>>>> Chris
>>>>>         
>>>> 
>>>>       
>>> 
>>>     
>> 
>> 
>>   
> 

Reply via email to