Cherry-pick of #1643: Capacity prediction based on physical memory

[jkaniuk]

Cherry-pick of #1643
with dependency on #1550

/cc aleksandra-malinowska
/cc MaciekPytel
/cc losipiuk

/hold

[losipiuk]

/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: losipiuk

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• cluster-autoscaler/OWNERS [losipiuk]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[jkaniuk]

Tested by:

• spinning up cluster with empty autoscaling node-pool
• adding pod with capacity 1Mi bigger than computed according to new allocatable calculations minus 220Mi (DeamonSets)
• confirming that will not spin up a node
• removing pod
• adding pod with capacity computed according to new allocatable calculation minus 220Mi (DeamonSets)
• confirming that will spin up a node
• pod will not fit as real allocatable is slightly less (1-2MB)
• waiting for node to be taken down as unneeded
• confirming that no new node is added to that nodepool because of that pod (so cache is working)
• deleting node pool
• confirming that cluster-autoscaler handles node pool removal
• adding new node pool with the same name
• confirming that new node once again is spin up for given pod and new node pool (so cache is cleared)

Separately, capacity and allocatable calculations were checked against many machine-types and GKE versions 1.11 and 1.12.
Calculations for 1.13 should be the same as for 1.12.

[jkaniuk]

/hold cancel