January 2019
Intermediate to advanced
484 pages
11h 48m
English
Kubernetes only uses requests to schedule pods, so the summation of limits from all pods scheduled to the same node might exceed the node's capacity. We can have, for example, a node with 1 Gi memory, but the total limits from all pods on the node could be 1.1 Gi or more. This model allows Kubernetes to oversubscribe a node, hence leading to higher resource utilization. Nonetheless, the allocatable resources on a node is finite. If a pod without any resource limit exhausts all resources and causes an out of memory event on a node, how does Kubernetes ensure that other pods can still get their requested resources? Kubernetes approaches this problem through ranking pods by their QoS classes.
There are three ...
Read now
Unlock full access