k8s memory quota - request and limit

gistfile1.txt

Memory
	- Limit:
		○ On a ns: used to calculate available resource: avail = limit - used
		○ Sum(pod limit) must be less than the ns limit quota
		○ Special behavior: On a pod without ns: kill it if exceed
	- Request:
		○ On a ns: used to control how much can be requested
		○ Sum(pod request) must be less than the ns request quota
		○ Special behavior: When ns request exists, pod request missing, pod request falls back to pod limit
	

Memory
In any case, Pod request respect total available memory (total_physical - total_used)
	- When no request, no limit on ns
		○ When pod has no request, only limit: 
			§ No quota
		○ When pod has no limit, only request:
			§ No quota
		○ When pod has both request and limit
			§ No quota
			§ OOM killer can kick in if exceeds
		○ When pod has nothing
			§ No quota
	- When only limit on ns
		○ When pod has no request, only limit
			§ Pod limit must <= quota: limit_ns - sum(pod_limit)
			§ E.g., ns limit 100, pod limit is 60, end up with 1 pod
		○ When pod has no limit, only request
			§ Error, need a pod limit
		○ When pod has both request and limit
			§ Pod limit must <= quota: limit_ns - sum(pod_limit)
			§ Pod request is not used for quota
		○ When Pod has nothing
			§ Error, need a pod limit
	- When only request on ns
		○ When pod has no request, only limit
			§ pod request must <= quota: Request_ns - sum(pod_limit)
			§ Pod limit act as pod request
			§ E.g., ns request = 100, pod limit = 60, end up with 1 pod
		○ When pod has no limit, only request
			§ Pod request must <= quota: request_ns - sum(pod_request)
			§ Pod limit is not used for quota
			§ E.g., ns request= 100, pod request=40, we end up with 2 pods
		○ When pod has both request and limit
			Pod request must <= quota: request_ns - sum(pod_request)
			Pod limit is not used for quota
			E.g., ns request = 100, pod request= 40, pod limit = 70, we end up with 2 pods.
	- When both request and limit on ns
		○ When pod has no request, only limit
			You expect: Pod limit must <= quota: limit_ns - sum(pod_limit)
			Actually, because ns_request exists, pod request missing, pod request falls back to pod limit, and request quota is smaller than limit quota, so you will see: Pod request must <= quota: request_ns - sum(pod_limit)
			E.g., ns request = 100, ns limit = 200, pod limit = 60, consider pod limit, you expect 3 pods, but actually, we see 1 pod, because pod request falls back to 60, we reach request quota first.
		○ When pod has no limit, only request
			§ Report error
		○ When pod has both request and limit
			§ Both could work
			§ E.g.,
			NS: Request 100, limit 200
			Pod:  request 30, limit 100, 2
			Pod:  request 60, limit 100, 1