Self-hosted EC2 Runners

[davinchia]

What

Use self-hosted ec2 runners for the main Airbyte build as discussed today. Will bring this up for evaluation after a week. Anecdotal testing shows builds generally finish in 9 - 10 mins vs 15 - 20 mins on the regular runner, even after accounting for the time taken to spin up the instance (about 1 min).

Removed this from the frontend tests since Cypress is running into an issue that seems to be related to how the tests are structured and I don't think we need to debug now.

The slash commands will continue running on the github runner.

How

• Use https://github.com/machulav/ec2-github-runner to start and stop an individual instance per job we ran as part of the build.
• This uses an 8 core EC2 machine (c5.2xlarge). This instance is cost/performance sweet spot from my testing. Cost is 0.34 cents an hour, which gives us 1000 USD / 0.34 USD * 60 mins ~= 177k mins/month of build time.

Recommended reading order

[davinchia]

A single approval is sufficient. Tagging everyone so people are aware of what is going on.

Not merging this in until our request to increase our AWS vCPU limit is approved.

[davinchia]

these are all resources living in the aws dev account.

[sherifnada]

only comment is that the checks are pretty muddled. (checkmark per start and stop) Can we just make starting the runner a step in the build?

[davinchia]

Our limit got increased to 160 vCPU, which means we might run into issues if too many builds are triggered at once. AWS says we can request another limit increase when this happens.

@sherifnada I looked into trying to clean this up; I don't think it's straightforward since the all steps of a job need to run on the same instance. The start/stop jobs are on Github's instance, while the build steps aren't. Will merge for now and look into this if other people find this annoying.