I'd like to ask for thoughts on a compromise between CTR and RCT. I suggest that we connect the CI to JIRA and either run it automatically or add provide a mechanism for users to trigger the CI. CI should then report the results to the JIRA as a comment.
If CI comes back +1 on a committer's patch, then we can commit without further review. (Essentially automated RCT). If CI returns -1 for whatever reason, a committer can request a +1 from another committer to override and commit. I believe this would streamline the process and provide something similar to CTR but with some added protection of RCT. If we don't want to wait for CI integration, can we at least create a "checklist" documenting all conditions / tests a committer must follow before committing? On Mon, Sep 14, 2015 at 2:09 AM, Konstantin Boudnik <[email protected]> wrote: > I think I am even more relaxed with the initial requirements for the CI. > With > a few bugs that BIGTOP-1494 (yes, shame on me) the CI was able to catch > them. > And the fact that more than one pair of eye was reviewing the code and > doing > the testing didn't prevent the bugs from going in. > > I say CI where it is today should allow us to start doing CTR. I'd suggest > we > add a post commit job that will run the RAT check and at least compile all > the > code to quickly see if anything is wrong. > > What you have in the list is the idea situation, but perhaps we can start > adding this _after_ we are the CTR. What do you think? > > Cos > > On Mon, Sep 14, 2015 at 02:46PM, Evans Ye wrote: > > I notice the CTR discussion in ignite community which Cos mentioned > Bigtop > > there. > > I pretty much champion the CTR model since sometimes I fix a bug at > local, > > but just don't have the energy to put it on the table. And we're not > being > > able to adopt CTR because of our CI is not fully functional yet. > > Which, I'd like to know how comfortable you're if we have our CI being > able > > to do this. > > To be precisely, this gets me cross the line to be able to adopt CTR :) > > > > Bigtop CI features: > > * Nightly build for bigtop/slaves images (test bigtop toolchain) > > * Build each component on each OS using above nightly build images > > * Take Bigtop Puppet and run deploy test using multiple config template > > (non-HA, HA, kerberos, etc) > > * Deploy a cluster and run fully smoke tests > > * Deploy a cluster and run fully integration tests > > > > If you feel the same, then we can try to make this happened. > > > > Evans > > > > > > 2015-01-29 4:57 GMT+08:00 Roman Shaposhnik <[email protected]>: > > > > > On Mon, Jan 26, 2015 at 11:55 AM, Konstantin Boudnik <[email protected]> > > > wrote: > > > > It seems like a reasonable course of actions to me. As I have strated > > > this > > > > conversation, I think I am responsible to at least do the first stab > at > > > the > > > > workflow. Let me do it and send the URL to this thread. > > > > > > Sounds good! > > > > > > Thanks, > > > Roman. > > > >
