Added FreeBSD

[mtelvers]

Tested with opam-repo-ci-local.

[kit-ty-kate]

LGTM, pushed to live to see if it works.

[kit-ty-kate]

Could also add the "(experimental)" inscription that macos has in the web ui?

[kit-ty-kate]

Let's merge this tomorrow if the CI isn't caught in an infinite loop by then (looks good so far)

[kit-ty-kate]

It looks very stable. Thanks a lot @dustanddreams and @mtelvers!

There is a couple of minor visual issues but i’ll follow up on that elsewhere and it’s not related to opam-repo-ci so merging.

[kit-ty-kate]

cc @hannesm we’re now testing FreeBSD in opam-repository

[hannesm]

@kit-ty-kate great to hear. Congratulations.

[hannesm]

It may be useful to record not only FreeBSD, but as well the major.minor thereof (IIUC this is 13.2 in your current setup). Also, I don't quite understand "ocurrent" in depth, but are there mechanisms in place to upgrade the system packages eventually / regularly (i.e. gmake / git / ..)? And the OS itself (freebsd-update is a great utility for binary updates)?

[mtelvers]

The initial deployment was on 12.4; however, these have been upgraded to 13.2 this morning.

The machine(s) will be managed via Ansible scripts. These are under development. They use freebsd-update to update the OS and build the base images.

[tmcgilchrist]

ocurrent/obuilder#109

[hannesm]

Thanks for your work. I'm curious why FreeBSD is considered experimental, i.e. what is the process to be considered stable (is there anything I can help with)?

[kit-ty-kate]

The two reasons are:

• the way the CI machinery has been done for FreeBSD is all brand new code and behaviour (jails on FreeBSD vs. runc on Linux vs. ZFS shenanigans on macOS). We've already had bugs to fix (e.g. networking was not working as expected)
• opam-repository was never properly tested for FreeBSD so some packages might not be compatible and things might fail unexpectedly outside of the control of one's PR. So this is basically a way of telling the users "if you see failure on that system, it might be normal"

For example macOS is still tagged as experimental mainly for the first reason but also packages are less tested so reason 2 is also at play.
I personally would expect FreeBSD to do a lot better than macOS and to be untagged as experimental much sooner (a month or two maybe)

[hannesm]

Thanks for your reply @kit-ty-kate. If you see strange FreeBSD issues in CI, feel free to tag me. But certainly a availble: os != "freebsd" is fine for some packages as well (if nobody bothers / it is too hard to fix).

[tmcgilchrist]

Agreeing with the reasons that @kit-ty-kate has posted, 1-2 months while we work through the setup seems right.
We have 1 main server running FreeBSD 13.2 that is handling the load quite nicely but it is early days and we want to keep it out of the main loop for opam-repo. So marking it as experimental makes sense. There are some setup improvements we still need to make around the hardware to avoid killing as many NVME drives.

Right now we support 4.14 and 5.0 on FreeBSD 13.2 x86. When 5.1 comes out will would swap 5.0 for that and continue with the 4.14 LTS and latest 5.* version. Do you think it is worth providing more OCaml versions?

So far the performance of FreeBSD obuilder is roughly inline with Linux runc implementation, which is very nice. Unlike macOS and Windows 😞