On 01/27/2016 10:02 PM, Remi Bergsma wrote: > Hi all, > > We should keep the simple approach that was used until now: one LGTM based on > code review and one LGTM based on integration tests (that’s not the same as > 2xLGTM). > > If we care about master stability, every change has to be tested for > regression. Period. Things may look OK, but still break something else in an > unexpected way. IMHO, giving up on it would be a shame. > > In a perfect world, CI would automate this for us and run it against every > PR. Current impediments I see for proper CI on every PR: > - there is no community hardware to run integration tests on (there was > supposed to be something) > - no support from Apache Infra: need Github write access (also to hook 3rd > party CI). It’s nowhere close. > - the community has very few people that are able to run the integration > tests on their own laptop/hardware and post the results on PRs > > > We tried to address points 1 and 2 and I propose to give up on them. > > If all devs would stop developing for a couple of days and make sure they > have a box that can run KVM, then clone the code we use [1] it should be > pretty straight forward to run the integration tests. We choose to make it > all virtual and used nested-virtualisation, but that is optional. A simple > Intel Nuc or similar will do. >
Yes, that's great code. I still have to master it, but the README should get us there. I still need to create one which also spins up a Ceph cluster. Probably a good thing to do now ;) Wido > People can then test the PRs that they find interesting and post results, > after which it can be merged. Distribute it. Share the load. > > This is what Wilder, Miguel, Boris, Daan an I have been doing for months. > Simply running the tests. Hundreds of times. That’s why we can run a 100% > Mission Critical Cloud close to master branch at Schuberg Philis. > > Regards, > Remi > > [1] https://github.com/schubergphilis/MCT-shared/ > > > > > > > On 27/01/16 21:33, "Sebastien Goasguen" <run...@gmail.com> wrote: > >> >>> On Jan 27, 2016, at 9:25 PM, Wido den Hollander <w...@widodh.nl> wrote: >>> >>> >>> >>> On 01/27/2016 09:18 PM, Sebastien Goasguen wrote: >>>> Folks, >>>> >>>> How about we freeze our repo entirely until we get proper CI in place. >>>> >>>> Seems to me all the hard work from Remi and co could be lost if we start >>>> committing again. >>>> >>>> Now Travis is not running, Jenkins fails all the time and nobody cares… >>>> >>>> So how about we figure out CI now ? and not do anything else. >>>> >>> >>> I think forces have to be combined to make this work. >>> >>> Questions which come to mind: Who runs Jenkins? Do we need a additional >>> slave? >>> >>> I haven't figured out the Integration tests completely personally. >>> >> >> In an ideal case, PR should trigger tests totally distributed on everyone’s >> own hardware. Then tests would report back on the PR. >> Only when all are green can we merge. >> >> there is an issue with creating triggers in github on our own, but I think >> that’s what we should aspire to. >> >> for instance, how can pcextreme automate its testing and report back on each >> PR ? >> >>> Wido >>> >>>> -Sebastien >>>> >>