How to enhance autoscaling speed?

Hello everyone,

I always use the autoscaling (and before docker+machine) executor before for my pipelines.
But last week I tried to configure a dedicated VM with the “docker” executor.

My pipelines were 3 times faster for the same operation, even if my VM (hosted on Hyper-V machine in our locals) were less power than the AWS EC2.

I know there also a time for “machine management” aka deploy EC2 machines etc… on the autoscaler. But if I take this time out, the docker executor still 3 times faster on build operation etc…

Even if I choose powerful EC2 machines (r6.large), it doesn’t change anything about time taken to build our apps, seem like something is bottleneck.
I’m using S3 cache.

What does I misconfigured ? The manager is a t3.small, can he bottleneck all the pipeline even on the build operation ?

Suggest sharing more details, logs, versions in the feedback issue GitLab Runner Autoscaling - Feedback issue for the new runner autoscaling solution (#408131) · Issues · GitLab.org / GitLab · GitLab Might also provide insights from others.