On Mon, 20 Jun 2011 09:35:00 -0500, Zach Pfeffer <zach.pfef...@linaro.org> wrote: > On 20 June 2011 08:29, James Westby <james.wes...@canonical.com> wrote: > > On Mon, 20 Jun 2011 12:10:08 +1200, Michael Hudson-Doyle > > <michael.hud...@linaro.org> wrote: > >> Something I don't completely have a plan for in my head at the moment is > >> connecting the dots, or preserving identity between componenets (as I > >> think Zygmunt put it). We have a build, which will trigger a test run, > >> which will result in a test bundle being uploaded to the dashboard. I > >> can see how to link the build to the test run and (mostly -- I'd like to > >> talk about the details of this) see how to link the test run to the > >> bundle, but I don't quite see how to have a link on the build page that > >> goes straight to the results. I guess we can always have a > >> $job_url/results link that does a redirect to the appropriate page in > >> the dashboard... > > > > I think it's vital for LAVA to connect a test run and the results from > > it, and I'm sure it will, but the details of that are something for you > > to decide. > > > > As a consumer of LAVA though I think it's also vital that there is some > > way to link to a particular test run for things like > > android-build.linaro.org.
Clearly. > > One way to do this would be to provide a test id back as the response to > > the API calls that schedule tests. That test id can then be used to > > contstruct URLs or make API calls. Yes, that sounds fine. > > If you don't want test ids to be the basis of APIs etc. then we could > > also make use of something that worked based on an externally defined > > identifier (e.g. the android-build.linaro.org URL.) If we could make API > > calls specifying our identifier, and maybe construct URLs too then we > > can also use that to display results or whatever on > > android-build.linaro.org. > > Lets cut out one level of indirection and use the build URL as the id. Is this expressing a preference or just trying to be helpful by getting some decisions made? Because in general I would rather those doing the implementation being the ones to make the implementation decisions. And requirements are best expressed using the "as a ... i want to ... so that i can ..." template. (apologies if this sounds snarky) Is being able to externally specify the job/test identifier a requirement? > > In terms of implementation I think that the latter would be easier given > > where android-build.linaro.org is now, but I don't think the former > > would be that much harder for us to do. > > > > As for android-build.linaro.org building other things I don't think > > that's really an issue for LAVA, as I think that we should push from > > android-build.linaro.org to LAVA, rather than having LAVA > > pull. android-build.linaro.org can then decide based on the build type > > what info to submit, including the type of artifacts, and LAVA can have > > a simpler job to map from say "android" to the set of tests to run for > > that. > > Yeah, I agree that LAVA shouldn't pull anything. The build system > should be requesting the tests at a high level, i.e. please run the > "Android Benchmark and Test load on build x." Cool. > The build should also probably wait until the test is done and get a > pass fail, this should help set us up for Gerrit integration in the > near future. I guess the naive implementation of this would leave the EC2 instance running while the test runs which would be wasteful... but that's something for Paul to figure out. > I'll pull a BP together. Cool. I don't *think* this is covered by an existing blueprint... Cheers, mwh _______________________________________________ linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev
