Do not share compute resources between environments

[bogdanfazakas]

Let's find a way we can still run benchmark jobs algorithms

[alexcos20]

When C2D in node was designed, the architecture was the following:

• You have engines (docker, kubernetes, etc)
  • each engine has resources
    • each engine has one or more compute environments

It ended up having multiple engines, each one with only one environment.

Proposal: (even if there are breaking changes on config side)

• have multiple envs per engine
• resources are shared inside that engine between multiple environments, but they are exclusive. IE: if you have gpu1 on both envA and envB, and there is a running job on envA , starting a new job on envB with gpu1 as resource will break. (to make things simpler, when you call getComputeEnvironments, you will see gpu1 in use on both envs)