On 23-11-15 13:57, Gordan Bobic wrote: > On 2015-11-22 23:16, Jacco Ligthart wrote: >> Hi All, >> >> As you might have noticed, Redhat released version 7.2 last week. I >> starting the preparation for the build of that. >> Sofar, I consolidated all trees we have (7.1, 7.1-updates, >> 7.1-updates-testing) and I have generated all new srpms from >> git.centos.org >> >> The buildsystem is now busy building these srpms. I guess it'll take a >> week or so just to build it all. > > Speaking of which, Bjarne, any news on the Koji exploration? We could > really do with moving to a build farm with more machines for mass > rebuilds > like this. > > On a semi-related note, I now have one of the 8-core (well, 4+4) Samsung > Chromebook 2s, and it IS in fact possible to run all 8 cores at the same > time, you just disable the big.LITTLE scheduler in the kernel (build time > configuration). The task scheduler is unaware of the assymetry, and the > little cores are approximately 3x slower than the big cores, so this > can cause performance issues, but this can be defeated for the build > server by simply running two build jobs at a time, one taskset to cores > 0-3, and the other taskset to cores 4-7. One build job will be the slow > one, the other the fact one, but they will both be progressing at the > same time, with the total compute throughput being increased. Effectively > it means treating the 8 core machine as two separate 4 core machines, > with one 3x faster than the other, but hey, it's still reduced overall > build time. :-)
Today the Odroid XU4 arrived, I think it has the same CPU. Let's see if I can tweak the thing similarly :) > >> I want to build all new rpms twice. so the plan is to build the git >> generated stuff now, and later download srpms from centos (when they >> have done their release). these will be build for the "real" RSEL 7.2 >> release. > > Can you elaborate on this? Are you saying you are building everything > from git, rather than src rpms? Is that just to get a head start on > finding problematic packages? Yes, but also because we found in the past that the order in which such a large set of updates is build matters. If you get that wrong, you might end up with building new applications before the new libraries that they depend on. Basically this is the issue we had with 'evolution-data-server' on RSEL6. (said differently: sometimes the requires section in a srpm is not specific enough) Because I don't know the "right" order I just build all new packages twice. Once against the current distribution and the second time against current + all newly build stuff. To get a head start I now do the first iteration from srpms build from their git. > >> To free some disk space, I've removed the 7-alpha and 7-alpha-updates >> trees from http://ftp.redsleeve.org/pub/el7-devel/el7/ >> I'm reasonably sure that that won't be missed by anyone. > > Good idea. As I said in the earlier post, I really need to have a look > into refactoring the structure of the repositories, and update the > redsleeve-release packages to match if required. > >> @Gordan, in the consolidation step I found that the srpm for >> redsleeve-indexhtml-7-9.el7.0.noarch.rpm is missing. Do you have a copy >> of that somewhere? > > Good catch. I will look for it, but I have a suspicion it may be on a > machine that is currently powered off and I won't be near it until next > weekend. If all else fails I'll build a new one. > > Gordan _______________________________________________ users mailing list [email protected] https://lists.redsleeve.org/mailman/listinfo/users
