Cost-Efficient Continuous Integration | Hacker News Reader

Cost-Efficient Continuous Integration

taras.glek.net

59 pointscpeterso9 comments

A little sad to read they're thinking about replacing Buildbot. I spent quite a bit of time making a Buildbot-based CI system and was very impressed with its capabilities.

> "Bare metal virtualization with OpenStack for Mac, Windows, ARM boards is an open problem"

I guess, more specifically, is the programmatic virtualization of these systems. As you can't easily program into the cold-start process of Mac or Windows. Whereas e.g. with RedHat/CentOS you have a kickstart file.

I use buildbot but it does annoy me. Apart from things like not understanding ipv6 (Twisted's problem I gather), the main issue is that it is hard to actually get reproducible builds as unlike say Travis CI it does not rebuild an environment from scratch for each build.

That's not really buildbot's fault. You have a lot of control on how your project builds.

Our build system created a chroot each time it went to build - so that got us the clean environment we needed. Buildbot had no effect on that, it just ran the build script.

If you really wanted to you can have buildbot coldstart a machine from scratch - we coldstarted machines for testing, but it was definitely doable for building as well.

True, I probably will do. But I also forgot to mention its inability to clean up stray processes, ita habit of crashing, and all the other small annoyances...

Reproducible build is a very hard problem. Did you just say Travis can do reproducible build? If so, what's the proof of that? The science community has been struggled to do 100% reproducible build (I have been involved as undergraduate trying to design such building environment).

No I wouldnt say that at all of travis, but it starts from a clean (sort of) defined environment for each build. There was an interesting talk at Fosdem about Debian reproducible builds.

But that is as if someone export the vm into a vagrant box/ova file and uses such image as base to start off a new build, wouldn't it (plus using puppet/chef/ansible/saltstack/shell script to ensure all configurations are set).

I don't understand how it can be cost efficient running CI on Amazon AWS? Seems like it would be cheaper with a private cloud ...

Well it is a bursty workload, depending when people commit, so some spillage into public cloud probably makes sense, with base levels on private infrastructure. But they do say they intend to do most in house.

Reply on news.ycombinator.com