Skip to content

Latest commit

 

History

History
44 lines (35 loc) · 1.65 KB

source code: controller - podgc controller.md

File metadata and controls

44 lines (35 loc) · 1.65 KB
Raw
tags: kubernetes, sourcecode, controller

kubernetes sourcecode: podgc controller

For a long running cluster, it will produce a lot of garbage pods. And the podgc controller is responsive to handling the garbage collection.

The controller runs gcc.gc every 20 seconds. And the gcc.gc deals with 3 kinds of garbages: terminated pods, orphaned pods, unscheduled terminating pods. Orphaned pod is pod that related node doesn't exist. Unscheduled terminating pod is pod that hasn't been scheduled and is terminating.

func (gcc *PodGCController) Run(stop <-chan struct{}) {
	go wait.Until(gcc.gc, gcCheckPeriod, stop)

	<-stop
}

func (gcc *PodGCController) gc() {
	pods, err := gcc.podLister.List(labels.Everything())
	nodes, err := gcc.nodeLister.List(labels.Everything())
	if gcc.terminatedPodThreshold > 0 {
		gcc.gcTerminated(pods)
	}
	gcc.gcOrphaned(pods, nodes)
	gcc.gcUnscheduledTerminating(pods)
}

For terminated pods, kubernetes has a default threshold that it won't delete them if total number of terminated pods is less than the threshold. And it will sort them by creationTimestamp (I don't know why not use deletionTimestamp) so that terminated pod that created earlier will be deleted earlier.

Attention, the default threshold 12500 is very large which means that it won't delete anyone until you have more than 12500 terminated pods. For small cluster you'd better use a small threshold. The --terminated-pod-gc-threshold option of controller manager can be used to set it.

references