Amazon EC2 machine provisioning is terribly slow

When provisioning an EC2 instance, it takes more than 10mins just to provision a single t2.micro instance. I understand that it means:

  • starting an instance,
  • installing docker,
  • pull rancher-agent,
  • running and registering the agent on the rancher server.

Is there something i can do to improve the provisioning speed without having to create pre-configured AMIs manually? Where is the Java code doing this in Cattle?

I tested again with an m3.large instance and it could provision the machine in 4 minutes, which is fine. Also, as my rancher server is running on my local machine, the machine provisioning is limited by the upload speed of my internet connection.

You can close the subject, speed should be okay when i’ll migrate our rancher-server on a cloud instance.

FYI the work is all in docker-machine, we’re pretty much just calling it other than one docker run to start the agent at the end.

t2s get a slowly filling bucket of CPU time credits and get zero time when they are out until more credits flow in. So the install process is probably exhausting the credits and then everything slows to a crawl from there until it is complete and has time to build up credits again.

Thanks for the insight, that’s what i thought. I noticed the similarity between the docker-machine cloud driver requirements and the rancher UI requirements.