If a Pod (service instance) goes down — what does Kubernetes do?

Kubernetes notices the failure → kubelet (on node) + Controller manager + ReplicaSet continuously monitor Pod health
Kubernetes kills the dead Pod completely (terminates it)
Kubernetes creates a brand-new Pod instance → With a new Pod UID and possibly a new IP address (It is NOT restarting the old Pod process. It is creating a fresh Pod)
Kubernetes ReplicaSet ensures “desired replicas” count Example: You said: "I want 3 replicas" → If 1 Pod crashes → ReplicaSet sees there are only 2 Pods running → immediately creates a 3rd Pod
The new Pod automatically registers in Service Endpoints → The Kubernetes Service will automatically route traffic to healthy Pods only (dead Pods are removed from Endpoints list)

Summary

Action	What happens?
Pod crashes	Kubernetes deletes it
New Pod?	Yes — freshly created (NOT restart same Pod)
Same IP?	Usually NO → new Pod gets a new IP
Traffic affected?	Minimal → Service routes only to healthy Pods
Manual action needed?	NO — Kubernetes auto-recovers

Clean interview answer

“If a Pod crashes in Kubernetes, it does not restart the old process. Kubernetes ReplicaSet immediately creates a brand-new Pod instance to replace it, ensuring the desired replica count is maintained. The new Pod may have a new IP, but the Kubernetes Service automatically updates its Endpoints list to route traffic only to healthy Pods — providing self-healing and high availability.”

Advanced note (optional but gold in interviews)

You can also configure liveness probes and readiness probes → So that Kubernetes knows when to restart unhealthy Pods and when to stop sending traffic to Pods that are not ready

Bottom line

Kubernetes creates a fresh Pod — NOT restart same Pod process

Liveness Probe vs Readiness Probe

Probe Type	What does it check?	Action if fails
Liveness Probe	“Is the application still alive (running)?”	Kill + Restart the Pod
Readiness Probe	“Is the application ready to receive traffic?”	Remove Pod from Service endpoints (Stop sending traffic)

Why do we need these probes?

Apps sometimes hang / get stuck (process is running but not responsive)
Apps sometimes boot slowly (not ready to serve traffic yet after restart)
Apps may go into temporary errors (like DB connection lost)

We need automated ways to:

Restart dead apps (liveness)
Stop sending traffic to unready apps (readiness)

Example to make it clear

Let’s say you have a Spring Boot app:

It takes 30 seconds to start up
It sometimes hangs after running 5 hours (thread deadlock)

Without probes → Kubernetes thinks Pod is fine as long as process runs But in reality, it may be unresponsive or not ready

Example config

livenessProbe:
  httpGet:
    path: /health
    port: 8080
  initialDelaySeconds: 60
  periodSeconds: 10

readinessProbe:
  httpGet:
    path: /ready
    port: 8080
  initialDelaySeconds: 30
  periodSeconds: 5

What does this mean?

Liveness Probe → Kubernetes will call /health every 10s → If /health fails (non-200) → Kill and restart Pod
Readiness Probe → Kubernetes will call /ready every 5s → If /ready fails → Remove Pod from Service endpoints (no traffic) → When /ready passes → Add Pod back

Common use cases

Use Case	Which probe?	Why?
App crashes or hangs	Liveness	Restart automatically
App starts slowly	Readiness	Don’t send traffic too early
App temporarily unhealthy (DB down)	Readiness	Stop traffic temporarily, don’t kill
App stuck in infinite loop	Liveness	Kill and restart

Clean interview answer

“Liveness probes check if an app is still alive; if it fails, Kubernetes kills and restarts the Pod. Readiness probes check if an app is ready to serve traffic; if it fails, Kubernetes stops routing traffic to that Pod but does not restart it. Liveness ensures self-healing, and readiness ensures smooth traffic handling during startup or transient failures.”

Bonus — How are probes implemented?

You can configure probes in 3 ways:

HTTP GET — hit an endpoint (common for web apps)
TCP Socket — check if a port is open
Exec command — run a command inside container

Golden rule (to remember easily)

Probe	LIVE or READY?	Kill? or Remove from traffic?
Liveness	“Am I alive?”	Kill Pod
Readiness	“Am I ready to serve?”	Remove Pod from traffic