On Mon, Nov 29, 2010 at 9:53 AM, Simon Laws <[email protected]> wrote: > On Mon, Nov 29, 2010 at 7:44 AM, ant elder <[email protected]> wrote: >> On Mon, Nov 29, 2010 at 1:43 AM, Luciano Resende <[email protected]> >> wrote: >>> On Sun, Nov 28, 2010 at 2:30 PM, ant elder <[email protected]> wrote: >>>> I've spent a bit of time on this now and talked to the Hudson build >>>> people, still don't yet have a very perfect approach. >>>> >>>> The problems are that the Tuscany build takes ages and so often times >>>> out or it quite often fails due to some transient issue, perhaps the >>>> running so long is making it more susceptible to those transient >>>> issues. As there is just one monolithic build run any problem causes >>>> it all to fail so we don't get the snapshot jars or distributions >>>> published, and also may not get to see anything thats been broken by a >>>> change. >>>> >>>> One of the reasons it takes so long is that there is a problem with >>>> Hudson and/or the ASF Hudson set up which makes the archiving between >>>> the Hudson salves and master really slow, eg it can take many hours to >>>> archive the tuscany distributions as they're so big, but even some of >>>> the other big jars can be quite slow too. >>>> >>>> There is a way to disable the archiving which makes it a lot faster >>>> but then it doesn't keep the last successful build (which we use for >>>> the nightly distro downloads on the website) or publish the snapshots >>>> after a successful build. With the Hudson archiving disabled the build >>>> could do a deploy during the build run (ie mvn deploy instead of mvn >>>> install) but then if there is a failure somewhere it could be >>>> deploying something incomplete or broken. >>>> >>>> It seems like there should be some way to get it working better by >>>> splitting it up into multiple Hudson jobs which depend on each other, >>>> but i haven't yet found an approach that works very well or isn't >>>> really complicated or slow. I'll keep playing around but if anyone >>>> else has any suggestions then please do chime in. >>>> >>>> ...ant >>>> >>> >>> I still think that we SHOULD NOT deploy SNAPSHOT if the build didn't >>> pass the tests. >>> >>> As for the archive, I don't think we need them anymore, and we can >>> link the nightly build directly to the "published distribution" from >>> the snapshot repository [1]. >>> >> >> How can we do those two things without enabling Hudson archiving which >> will mean the build will nearly always fail? >> >> ...ant >> > > There has been recent discussion on the archive issue [1] but no > solution yet. My fastest local full build takes 30 minutes. On Hudson > it regularly takes many hours. The majority of this extra time seems > to be wasted on the archive task. This is an issue for Hudson in > general. > > [1] http://mail-archives.apache.org/mod_mbox/www-builds/201011.mbox/browser > > Simon > > -- > Apache Tuscany committer: tuscany.apache.org > Co-author of a book about Tuscany and SCA: tuscanyinaction.com >
I've been struggling to make useful suggestions here but I've been mulling this over today and on balance I'm personally more interested in seeing Hudson run the complete build (including tests) successfully than having Hudson publish artifacts. If required we can always publish the artifacts manually. If we went this route I'm assuming, based on what Ant posted previously, that we could turn off archiving and let the full build run, Just my 2c Simon -- Apache Tuscany committer: tuscany.apache.org Co-author of a book about Tuscany and SCA: tuscanyinaction.com
