Repls are container-based. We control how much resources each repl can use and so we can control how it much it costs. We do swap out idle repls and wake them up on the first request that comes in (heroku style).
How do you start up these containers so quickly anyways? Do you keep some running hot, or delegate the startup to something like amazon's ECS? I've long wanted to run integration tests en masse in containers, but never figured out how to solve issues around long startup times.
Yes we do container pooling. We do all the container manaegment ourselves. Using services like ECS will generally be slower.
Is the wake up on first request code open source by any chance? Would love to reuse for GitLab.
Not yet, hoping for a big push oss push next quarter. You want to use it for your ci/cd platform?
OK, thanks for letting us know. We want to use it for our PaaS functionality, but the PaaS will also be used by our Web IDE so it is competitive.
Cool, I didn't know that Gitlab was going that direction. Really awesome how much surface area you're able to cover.
Thanks! You're not doing bad yourself either, keep up the good work!