Hello all, In my ongoing crusade to provide reliably green CI results for Tahoe, I'm now looking at Buildbot.
I don't think anyone would argue that Buildbot has done anything other than provide many, many years of valuable service to the project. Yet, these days, there are many alternatives which can satisfy many of Tahoe's needs. Tahoe is already using TravisCI for various Linux- and macOS-based testing and CircleCI for still more Linux-based testing and Appveyor provides Windows-based testing. These three services are managed and provided free of charge to the project (they all offer limited free service to open source projects). They also offer almost complete configuration and control through in-repository configuration files which confers on all Tahoe contributors to potential to be CI maintainers as well. Contrasting with this, Buildbot gives us more control over configurations than any of these SaaS CI providers and puts more (all) of the management overhead on us as well. Currently, we spread that management cost out by having approximately one person who can manage the master and then approximately one person to manage each of the slaves. Every one of these people (I'm one of them) is a single point of failure for at least one of the CI jobs. In practice, failures come up somewhat often and end up being painful to resolve because there is no organization to manage all of these individual maintainers. The current state of Buildbot is not great. 4 of of 11 builders are red and have been red for a while. These are just a distraction for ongoing development. When "buildbot is red" then failures on buildbot become noise that gets ignored by developers - ie, a small number of consistent failures on buildbot negate almost all of the value of buildbot for the project. Another point is that all of the Linux configurations tested on buildbot are currently covered by CircleCI and CircleCI gives us an easy way to cover additional Linuxes. TravisCI covers macOS so the Buildbot macOS builders may not be necessary either. The only platform Buildbot covers that is not easily available from another CI system we're using now is OpenBSD - and the OpenBSD builders have been broken for a long time so while they exist, they aren't actually providing much (if any) value to the project. So ... I want to raise the idea that maintaining buildbot may not be a net positive to the project at this point and hear what others think about this. Jean-Paul
_______________________________________________ tahoe-dev mailing list tahoe-dev@tahoe-lafs.org https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev
