Kubernetes ReplicaSet的selector字段作用及标签匹配冲突处理问题

Great question—this is a classic gotcha that catches many folks new to Kubernetes controller logic. Let's walk through exactly what happens with your configuration:

• First, remember that a ReplicaSet’s spec.selector is its "targeting system"—it tells the controller which Pods it’s supposed to manage. In your case, it’s looking for Pods labeled tier: front-end-2.
• When you apply this ReplicaSet, the controller first scans the cluster for existing Pods that match that selector. If there are none (which is almost certainly the case here), it will try to create new Pods to hit your replicas: 3 requirement.
• But here’s the problem: your Pod template creates Pods with tier: front-end—a label that doesn’t match the selector. So every time the ReplicaSet spins up a new Pod, that Pod is invisible to the ReplicaSet’s controller. The controller doesn’t count it towards the desired 3 replicas.
• This creates an infinite loop: the controller keeps checking, "Do I have 3 Pods with tier: front-end-2?" The answer is always no, so it keeps spawning more tier: front-end Pods. This will continue until your cluster runs out of resources (CPU, memory, etc.), because the ReplicaSet never reaches its target.

Will the Total Number of Both Labeled Pods Equal 3? #

Nope, not even close. Here’s why:

• The ReplicaSet doesn’t care about Pods with tier: front-end—it only tracks Pods that match its selector. Since those don’t exist, it will keep creating new tier: front-end Pods in a futile attempt to reach 3 replicas of tier: front-end-2 Pods.
• You’ll end up with way more than 3 Pods (all labeled tier: front-end), and the ReplicaSet will show a status like "0/3 replicas ready" indefinitely because it never hits its desired count.

Quick Fix #

To fix this, you just need to align the two labels: either update the ReplicaSet’s selector to match tier: front-end, or change the Pod template’s labels to tier: front-end-2. Once they match, the ReplicaSet will recognize the Pods it creates, stop spawning new ones once it hits 3 replicas, and manage those Pods properly (restarting them if they fail, etc.).

内容的提问来源于stack exchange，提问作者pkaramol