structured build priorities

[syphar]

just jotting down some notes for now.

this month the queue was too full most of the time, and new builds come in more than we can handle.

But these can be of multiple categories.

• prio 0 (= normal builds) are fine most of the time, if I'm fast enough in depriotitizing workspaces that aren't yet
• prio 1 ( = depriotizied workspaces) also if often too long, but caches up
• prio 5 ( manual rebuilds from crates.io, but also new: deprioritized older releases when a newer was released)
• prio 20 (= automated rebuilds) don't matter for this story, we only queue them when the queue is empty enough.

things that could help:

• automated workspace discovery, best after the first full release, when we could group releases by repo and then find the prefix. Or mark the workspace in a field / table, with all crate names?
• clear text grouping / description for the priority sources/ groups.
• more intelligent deprio for older releases?

I'm also missing some metrics, not sure if we need to change the data or send more metrics to datadog.
idea:

• we have past build durations for a crate. we could check how stable these typically are, and then show the estimates in the queue? And perhaps even sum them up to tell you when your crate will be handled?