Hacker News Reader

Comment by shipit | Hacker News Reader

author here - we are using spot instances on GCP for executing the pipeline which at 10 hrs/day usage makes us $1 in profit/month :)

View on HN

I’d guess that’s for a fairly slow machine, or else the prices would be pretty high?

What sort of spec machine (or kubernetes pod) are you providing for that?

its a `2 vCPU, 7.5GB RAM` -- so far tested with open source projects including NDK and it stood up pretty well. Build times post close to what it would take on the local dev machine. Overall experience is that once you push your commit in about `2 mins + (local build time) mins` you'll get an email with the CI result.

Product wise we don't want the user to worry about machine resources as long as CI executes within reasonable time.

We can rent CPU/RAM in the cloud by the minute, but we can’t do that with the computer we’re developing on locally. I think this should naturally leads to faster builds being done in CI, more cost effectively, rather than CI being the slower builds.

Out of interest, have you modelled spot instance restarts into your costs? How about spot instance pricing going up?

How does Turing handle the instances being terminated? Does it just start again from the beginning?

We deploy the pipeline as a kubernetes job, once the build completes and posts logs/artifacts (apk) -- the job is archived and pods are evicted. So yes, starts from the beginning.

Just curious- is it built on Tekton? :) Have been tempted to play around with it.

ah no - just plain old Vue, Go, and k8s :)

might I recommend figuring something else out because... that sucks :)

hehe - actually I am an Android dev primarily and I thought that starting a project without CI sucks way more. So I built this for myself and was happy paying $7/mo for not having to spend weeks+$$$ on setting one up with current solutions.

Spot instances do take upto ~2 mins to come up so probably a turbo add-on that devs will feel happy paying for or please feel free to suggest.

Yeah, if there's no space for an extended period of time the service will be useless, with an endless backlog of jobs.

Reply on news.ycombinator.com