apply scheduling design and practices

[Huang-Wei]

Matching incoming (Shared)LoadBalancer request to a real LoadBalancer is actually a scheduling/horizontal-auto-scaling problem.

• horizontal auto scaling
  • if LoadBalancer resources are out of capacity, create one dynamically
  • if getting error on creating a LoadBalancer service, usually you ran out of capacity/quota in your account, then need to invoke IaaS SDK to increase quota for your account - maybe safer to do it manually by operator
• scheduling predicates - filtering LB resources
  • ports conflict (implements Ports predicate #6)
  • protocol (TCP/UDP) matching?
  • service "requests/limits" (or simply use a "weight" term)
  • service "{anti-}affinity"
• scheduling priorities - prioritizing LB resources
  • lease requested (balanced) vs. most requested (packed)