Re: CI builds on Windows
On 9.6.16 6:21 , Tommaso Teofili wrote: I'm for option 2 too! As this is the predominant opinion here I went ahead and created https://issues.apache.org/jira/browse/OAK-4456 to track this. Let's follow up there. Michael
Re: CI builds on Windows
On 9.6.16 6:02 , Davide Giannella wrote: On 08/06/2016 08:33, Michael Dürig wrote: 2. Try to enable it again and involve with the Apache infra team to see what options we have to speed things up I'm for option 2. As Windows boxes are so slow, we may want to start enabling only the unit testing and see how it goes. If we see we have room, enable ITs. I would definitely not run pedatic on an already busy box. IMO it should be the other way around: we need appropriate infrastructure to run our tests. So let's run them as we require them to run and see whether that instance can cope with it. Michael Cheers Davide
Re: CI builds on Windows
I'm for option 2 too! Regards, Tommaso Il giorno gio 9 giu 2016 alle ore 18:03 Davide Giannellaha scritto: > On 08/06/2016 08:33, Michael Dürig wrote: > > 2. Try to enable it again and involve with the Apache infra team to > > see what options we have to speed things up > I'm for option 2. As Windows boxes are so slow, we may want to start > enabling only the unit testing and see how it goes. If we see we have > room, enable ITs. I would definitely not run pedatic on an already busy > box. > > Cheers > Davide > > >
Re: CI builds on Windows
On 08/06/2016 08:33, Michael Dürig wrote: > 2. Try to enable it again and involve with the Apache infra team to > see what options we have to speed things up I'm for option 2. As Windows boxes are so slow, we may want to start enabling only the unit testing and see how it goes. If we see we have room, enable ITs. I would definitely not run pedatic on an already busy box. Cheers Davide
Re: CI builds on Windows
1. Do nothing, we don't need Windows coverage. I believe we do need it. Not only because people might actually use, but also because failures on Windows can be symptoms of problems that could occur on other platforms as well. Such as: - incorrect assumptions about system timers ("System.currentTimeMillis() surely will have changed during this call" -- dangerous with systems getting faster and faster) or - "being able to delete the file is sufficient" -- no, it's not, the file handle might still be open... Agreed. 2. Try to enable it again and involve with the Apache infra team to see what options we have to speed things up. Is that a big issue if we decouple the CI builds from the Linux ones? Not if there is an easy way to copy the configuration of the existing build matrix. However, my expectations for this working is quite low. But you are right, let's try first and complain later ;-) Michael 3. Move (some of) the CI builds to different systems (which might not be publicly available). ...as a last resort... Best regards, Julian
Re: CI builds on Windows
On 2016-06-08 09:33, Michael Dürig wrote: Hi, We currently don't have any CI coverage on Windows platforms. The build matrix on the Apache Jenkins instance has the respective components disabled [1]. Last time I checked the Windows slaves where so overloaded that enabling Windows builds would bump the build time from 40min to 9h. How should we proceed here: 1. Do nothing, we don't need Windows coverage. I believe we do need it. Not only because people might actually use, but also because failures on Windows can be symptoms of problems that could occur on other platforms as well. Such as: - incorrect assumptions about system timers ("System.currentTimeMillis() surely will have changed during this call" -- dangerous with systems getting faster and faster) or - "being able to delete the file is sufficient" -- no, it's not, the file handle might still be open... 2. Try to enable it again and involve with the Apache infra team to see what options we have to speed things up. Is that a big issue if we decouple the CI builds from the Linux ones? 3. Move (some of) the CI builds to different systems (which might not be publicly available). ...as a last resort... Best regards, Julian
Re: CI builds on Windows
I'm definitely for option 2 as well, as I think basic Windows coverage is more than nice to have. In order to reduce the time, maybe we don't need all the tests to run on windows? Daniel On 6/8/16, 10:02 AM, "Alex Parvulescu"wrote: >Hi, > >I would opt for #2 as well, we need to have at least some basic coverage >on >Win. > >alex > >On Wed, Jun 8, 2016 at 9:33 AM, Michael Dürig wrote: > >> >> Hi, >> >> We currently don't have any CI coverage on Windows platforms. The build >> matrix on the Apache Jenkins instance has the respective components >> disabled [1]. Last time I checked the Windows slaves where so overloaded >> that enabling Windows builds would bump the build time from 40min to 9h. >> >> How should we proceed here: >> >> 1. Do nothing, we don't need Windows coverage. >> >> 2. Try to enable it again and involve with the Apache infra team to see >> what options we have to speed things up. >> >> 3. Move (some of) the CI builds to different systems (which might not be >> publicly available). >> >> >> My preferred way would be to 2. and move on to 3. pretty quickly if this >> doesn't lead anywhere. >> >> Michael >> >> https://builds.apache.org/job/Apache%20Jackrabbit%20Oak%20matrix/ >>
Re: CI builds on Windows
Hi, I would opt for #2 as well, we need to have at least some basic coverage on Win. alex On Wed, Jun 8, 2016 at 9:33 AM, Michael Dürigwrote: > > Hi, > > We currently don't have any CI coverage on Windows platforms. The build > matrix on the Apache Jenkins instance has the respective components > disabled [1]. Last time I checked the Windows slaves where so overloaded > that enabling Windows builds would bump the build time from 40min to 9h. > > How should we proceed here: > > 1. Do nothing, we don't need Windows coverage. > > 2. Try to enable it again and involve with the Apache infra team to see > what options we have to speed things up. > > 3. Move (some of) the CI builds to different systems (which might not be > publicly available). > > > My preferred way would be to 2. and move on to 3. pretty quickly if this > doesn't lead anywhere. > > Michael > > https://builds.apache.org/job/Apache%20Jackrabbit%20Oak%20matrix/ >