[Development] Qt CI reliability

Jędrzej Nowacki jedrzej.nowacki at qt.io
Tue May 3 17:36:49 CEST 2016


On Tuesday 03 of May 2016 14:54:50 Sean Harmer wrote:
> On Monday 02 May 2016 11:14:24 Jędrzej Nowacki wrote:
> > On Saturday 30 of April 2016 20:26:20 Sean Harmer wrote:
> (...)
> What would be a *very* useful feature would be if we can trigger a test
> build of a change on a particular configuration for such cases where we
> don't have ready access to a configuration locally.
We have this feature, but it is not exposed. Qt account integration is missing 
as well as some kind of limitation to protect CI resources. If you have a 
really urgent thing to test ping me on IRC. I can run a test run for you.

> > I will ask IT about network, it seems that network interface was
> > re-configured during CI run and DHCP assigned a different IP. It should
> > not
> > happen (TM)
> 
> Yes that sort of thing should be done in a specified window out of hours
> after disabling the CI master to be able to disseminate jobs to the nodes.
Well it was not about maintenance. A node tried to re-new it's IP lease and it 
got a new IP instead of an old one. It may be some kind of DHCP miss-
configuration or operating system failed to ask in time to re-new the IP lease. 
I do not know, how on earth this could happen. I need to access logs.

> > Rule of thumb is: if logs show broken compilation it means: real problem,
> > don't blame CI. There are three main reasons, that I'm aware of, that can
> > cause the problem (sorted according to the probability):
> > 1. One of changes being integrated broke the compilation
> 
> Fine and expected and, with timely failures, not an issue.
> 
> > 2. One of module dependencies broke source compatibility
> 
> This is very rarely an issue, at least for Qt 3D.
> 
> > 3. There was a untested template update (this reason will almost disappear
> > soon)
> 
> Do you mean VM template? If so then yes that's again something that should
> ideally be verified before deployment.
Eh, wrong phrasing. Templates are tested. The problem is that the current 
process is a bit racy. Updating template is a work that require time, 
especially for testing and deployment. Regressions may appear during that time 
window. Anyway the process is about to be changed soon. 

Cheers,
 Jędrek



More information about the Development mailing list