Add autoscaling service per product to prevent mixing billable and not billable services to run on same machines

[sanderegg]

in AWS-master for example we have 1 autoscaling service taking care of all the services that run autoscaled.

This leads to:

• billable service that require an entire EC2,
• non-billable service that require specific resources,
  --> to run on the same EC2 sometimes as autoscaling does not set a user_id or project_id or node_id on the machine it starts.

Workaround:

• 1 autoscaling service per product checking for different labels:
  • s4l product: io.simcore.s4l-scalable-service
  • tip product: io.simcore.tip-scalable-service
  • osparc product: io.simcore.osparc-scalable-service

For now I believe this is non-urgent, but we should keep it in mind once we fuse osparc with s4l for example.

[sanderegg]

This is increasingly a problem since people are using master/staging extensively with multiple products

[sanderegg]

Another option is to modify the way we start services so that they add a constraint to their node_id, and the autoscaling service would add that label on the node it creates therefore reducing the randomness of the docker swarm to choose on its own where to set a service.