Meeting run by pietroalbini. Minutes written by pietroalbini. Attending: alexcrichton, kennytm, Mark-Simulacrum, pietroalbini, sgrif, shepmaster Start of the conversation
Rust 1.14.0 Cargo artifacts on the rust-lang-ci S3 bucket
Following on last meeting’s topics, pietroalbini investigated whether it would
be possible to redirect requests for Cargo 1.14.0 from the
bucket to our CDN. The answer is no, as the S3 support for redirects requires a
different endpoint than the one included in the manifest. The two main courses
of action were to do nothing (keeping the files in the
rust-lang-ci S3 bucket),
or to rewrite and resign the manifest of Rust 1.14.0 to point to the CDN.
Mark-Simulacrum was in favor of doing nothing, but pietroalbini pointed out keeping releases in two different places will make mirroring and caching efforts more complex, as mirror authors will have to special-case a separate domain just for a release. We then decided to rewrite and resign the manifest.
shepmaster proposed to add a test somewhere to ensure this doesn’t happen again, but nobody on the team had the time to do this. If someone is interested please ask in the infra channel.
Deduplicating CI configuration on GitHub Actions
One issue about GitHub Actions is it doesn’t allow to include shared pieces of configuration into workflows, forcing us to duplicate (for example) the steps needed to complete a build. pietroalbini investigated a few ways to work around the limitation, and the options he presented during the meeting are either writing our own configuration format and having a tool generate the GitHub Actions configuration from it, or duplicating the configuration manually and having a tool to ensure the manually duplicated things don’t diverge.
The rest of the team expressed concerns with generating the configuration, as using the generator adds even more complexity to our already complex CI. On the other hand pietroalbini didn’t like the tool to ensure the configuration doesn’t diverge, as that would make changes to the CI configuration harder for the people doing it. The agreement in the meeting is that pietroalbini will try to create other proof of concepts, hoping to find a better solution.
New server for perf
alexcrichton didn’t hear back from Hetzner about the new perf server yet.