| Re: [mdt-papyrus.dev] Hudson, Performances & Target Platforms |
|
Hi, The new build setup shows good results. However, the Papyrus-Master job still takes 25 minutes, whereas the complete Gerrit job takes 15 minutes. The only thing that Master does and Gerrit does not,
is packing the artifacts. This isn’t such an expensive operation, so I investigated a little bit on this issue. I found that the shared .m2 repository (Used – in theory – only by jobs that don’t depend on any nightly build) was 26 GB big, and contained a lot of Papyrus Nightly plug-ins. I’ve wiped this repository,
and I’d like to remind: If you are not using a job-private repository, do not use mvn clean install. If you depend on a nightly build, use a private repository Since the shared repository is not located on Hudson, there is no easy way to monitor/clean it, so please be careful when using it. I’ve added a job to clean it and removed 19 GB of Papyrus nightlies
(+ 7 GB of other projects). Now, a clean Papyrus-Master build is taking 15 minutes, which seems more consistent (Next builds should be faster) Also, if you’re using a private repository, please carefully check that your nightly dependencies are properly cleaned up (Either after each build, or by wiping the workspace every now and then) I’ve also reduced the Timeouts for Gerrit and the Test jobs, since they should now run faster – no need for a 3 hours timeout anymore Regards, De : mdt-papyrus.dev-bounces@xxxxxxxxxxx [mailto:mdt-papyrus.dev-bounces@xxxxxxxxxxx]
De la part de LETAVERNIER Camille Hi all, The contribution has been pushed to master. Please rebase your Gerrit patches on the latest master commit to benefit from the increased performances Regards, Camille
De :
mdt-papyrus.dev-bounces@xxxxxxxxxxx [mailto:mdt-papyrus.dev-bounces@xxxxxxxxxxx]
De la part de GERARD Sebastien 166342 Good job ! Thanks.
De :
mdt-papyrus.dev-bounces@xxxxxxxxxxx [mailto:mdt-papyrus.dev-bounces@xxxxxxxxxxx]
De la part de LETAVERNIER Camille Hi all, The recent experiments on Target Platforms [1] (And a few other optimizations, such as Eclipse Mirror discussed in Bug 492412 [2]) show promising results regarding performances, with full Gerrit builds taking less than
25 minutes (vs. 55 minutes currently), and down to 16 minutes when all remaining optimizations can be integrated. However, the cost of maintaining the target platforms is slightly higher than our current pom-based dependencies, mostly because we have some tooling support for Poms that we don’t have (yet) for Target Platforms. I’m
working on this, but this might still take a few days. I’d like to proceed with the current contribution as-is, so that general build performances can be improved (I’m sure all Gerrit contributors will welcome this patch :) ), while I keep working on the tooling
to avoid growing build maintenance. The basic Simrel-based updater is already contributed, so now it’s just a matter of going from ~15 clicks per milestone to ~3. There’s more information on the current tooling (What’s in, and what’s missing) in the bugzilla [1]. You can review the contribution here [3] [1] 492375: [Releng] Define a Target Platform to improve build performances https://bugs.eclipse.org/bugs/show_bug.cgi?id=492375 [2] 492412: Connection problems to external sources https://bugs.eclipse.org/bugs/show_bug.cgi?id=492412 [3] 70014: Bug 492375: [Releng] Define a Target Platform to improve build performances [I5d454d4a] https://git.eclipse.org/r/#/c/70014/ Regards, Camille | ||||||||||
