Ephemeral Runner Initialization Performance

[JohnYoungers]

Using ephemeral runners with gha-runner-scale-set, I'm seeing initialization timelines that look like the following:

MM:SS
00:00	Workflow queued
00:20	Controller scaling + Pod created
00:25	AWS EBS volumes provisioned and mounted
00:58	Connected to GitHub
01:01	Running job

For that initial ~20 seconds (the time from the workflow being queued to the controller recognizing it needs to scale), are there any configurations available to cutdown on that time?

Using the original controller, pods were generally created fairly quickly using the webhook server; my hope was with the new controller's setup (persistent connection/long polling?) it would be notified relatively quickly as well.

[JohnYoungers]

I think we're probably as performant as we can get outside of that initial scaling trigger; this is the timeline for a worker that uses no volumes and doesn't use docker:

MM:SS
00:00	Workflow queued
00:20	Controller scaling + Pod created
00:23	Connected to GitHub
00:26	Running job