小编点评

## Summary of the content This document describes the logic used to get pods belonging to a specific deployment using labels. **Key points:** * Tags are not the only way to identify pods belonging to a deployment. * `replicaSets` are generated before `pods` and are used to determine the pods for a deployment. * When a pod is terminated, it is not automatically removed from the `pods` list. * A pod is considered to belong to a deployment if it is not controlled by a replicaSet of the deployment. * The code uses different `LabelSelector`s to find pods belonging to different deployment replicas. * If a pod is found to be in a terminating state, it is added to the `pods` list. ## Conclusion The code uses a combination of labels and `replicaSets` to identify pods belonging to a deployment. While tags are a valid approach, the code may not always use them, especially when dealing with rolling updates. This can lead to pods being missed or not being properly grouped.

正文

标签获取

我们获取那些pods属于某个deployment时最先想到的可能是通过标签获取，其实这个是不准确的。因为标签并不是唯一的，也就是说不同deployment其实是能有相同标签的。

replicaSets获取

deployment 的产生pod流程如下: deployment->replicaSets->pod。
deployment 先产生replicaSets, replicaSets再产生pod。所以我们可以根据以下步骤获取到pod。

1. 根据标签获取到replicaSets
2. 根据replicaSets ownerReferences字段进行过滤。就可以得到属于某个deployment的replicaset。
3. 根据标签获取到pods
4. 根据pods ownerReferences字段进行过滤。就可以得到属于某个replicaset的pods。

注意一点这里有个比较坑的地方，就是deployment在滚动更新中会产生多个replicaset, 如果没有设置revisionHistoryLimit保留历史的replicaSet的话。那么当新的pod起来了，老的pod在终止时， 此时老的replicaSets会被删除，但是
老的pod又是被老的replicaSets控制的，这个时候去查询deployment 下面的pods就会将这些终止态的pod忽略掉，这在某些系统其实问题很大，就比如我现在做的发布系统，某个发布必须要等老的都终止掉了才能执行下一个业务,那么这个时候其实是有问题的。目前我能想到的解决办法是通过标签筛选出来的pod如果是终止态的话,也归属于这个deployment。大家如果有更好的方法请在下面留言。

代码

func (r *Usecase) podsListByDeploymentV2(deployment *appv1.Deployment) ([]v1.Pod, error) {

	replicaSetSelector, err := metav1.LabelSelectorAsSelector(deployment.Spec.Selector)
	if err != nil {
		return nil, xerrors.WithStack(err)
	}

	replicaSetListOptions := metav1.ListOptions{LabelSelector: replicaSetSelector.String()}
	allReplicaSets, err := r.k8sClient.AppsV1().ReplicaSets(deployment.Namespace).List(context.TODO(), replicaSetListOptions)
	if err != nil {
		return nil, xerrors.WithStack(err)
	}
	ownedReplicaSetsUids := make(map[types.UID]struct{})
	for i := range allReplicaSets.Items {
		if !metav1.IsControlledBy(&allReplicaSets.Items[i], deployment) {
			continue
		}
		ownedReplicaSetsUids[allReplicaSets.Items[i].UID] = struct{}{}
	}

	podSelector, err := metav1.LabelSelectorAsSelector(deployment.Spec.Selector)
	if err != nil {
		return nil, xerrors.WithStack(err)
	}
	podListOptions := metav1.ListOptions{LabelSelector: podSelector.String()}
	allPods, err := r.k8sClient.CoreV1().Pods(deployment.Namespace).List(context.TODO(), podListOptions)
	if err != nil {
		return nil, xerrors.WithStack(err)
	}

	//replicaSetUID := replicaSet.UID
	rsPods := make([]v1.Pod, 0)
	for i, pod := range allPods.Items {
		controllerRef := metav1.GetControllerOf(&allPods.Items[i])
		if controllerRef != nil {
			if _, ok := ownedReplicaSetsUids[controllerRef.UID]; ok {
				rsPods = append(rsPods, pod)
				continue
			}
		}
		podStatus := r.k8sRepo.GetPodStatus(&pod)
		if podStatus.Status == repo.TerminatingStatus { // 终止态的也可以认为属于这个deployment
			rsPods = append(rsPods, pod)
			continue
		}
	}

	return rsPods, nil

}