[Feature] Prevent overcommitting memory

[alexellis]

Expected Behaviour

If a user has a 1GB node and sets memory limits of 100Mi for 10 containers, then that should prevent any new containers from being deployed i.e. the 11th and 12th which would overcommit the memory, and possibly cause an OOM somewhere.

Current Behaviour

You can continue to deploy containers beyond the memory available on the the host

Possible Solution

Read the available memory, minus a set amount of headroom for the system, and limit according to the total memory reserved by all deployed containers.

I.e.

1000MB total
System headroom = 500MB

Available = 1000-500 = 500MB

nodeinfo1 - 250MB
nodeinfo2 - 250MB

Deploying nodeinfo3 with a limit of 100MB will fail due to the sum of (nodeinfo1 - nodeinfo2 - headroom) == 0