In this lab, you follow a Pod through its lifecycle, run several containers in one Pod, organize Pods
with labels and separate them with namespaces.
Step 1: Follow the Lifecycle of a Pod
Create a directory for the files of this lab and change into it:
mkdir -p ~/labs/2.3
cd ~/labs/2.3
Create the file countdown.yaml in ~/labs/2.3. The Pod counts down from 5 and then ends:
apiVersion:v1kind:Podmetadata:name:countdownspec:restartPolicy:Nevercontainers:- name:countdownimage:busybox:1.37command:["sh","-c","for i in 5 4 3 2 1; do echo $i; sleep 2; done"]
Create the Pod:
kubectl apply -f countdown.yaml
pod/countdown created
Watch the Pod right away. --watch keeps the command running and prints a new line for every change:
kubectl get pod countdown --watch
NAME READY STATUS RESTARTS AGE
countdown 0/1 ContainerCreating 0 0s
countdown 1/1 Running 0 4s
countdown 0/1 Completed 0 16s
Once the status is Completed (after about 15 seconds), stop the watch with Ctrl+C and show what the container
printed:
kubectl logs countdown
5
4
3
2
1
Explanation
Every Pod goes through phases. The column STATUS shows them in more detail:
Phase
Status
Meaning
Pending
ContainerCreating
The Pod has a node, which downloads the image and starts the container
Running
Running
The container runs
Succeeded
Completed
The container ended successfully
command replaces the default command of the image. restartPolicy: Never tells Kubernetes not to start the
container again after it ended.
Step 2: Restart the Container Automatically
Change the restartPolicy in countdown.yaml from Never to Always:
spec:restartPolicy:Always
The restartPolicy of an existing Pod can’t be changed. Delete the Pod:
With restartPolicy: Always, Kubernetes starts the container again every time it ends, even after a successful
run. RESTARTS counts how often that happened. Despite its name, CrashLoopBackOff doesn’t mean that the container
crashed: the container keeps ending, so Kubernetes waits a little longer before every new restart, up to 5 minutes.
Always is the default and the right policy for servers like nginx, which should run all the time. Never fits
tasks that end. The countdown keeps restarting for the rest of this lab.
Step 3: Run Several Containers in One Pod
Create the file web.yaml in ~/labs/2.3 with a Pod that runs three containers:
apiVersion:v1kind:Podmetadata:name:webspec:initContainers:- name:waitimage:busybox:1.37command:["sh","-c","echo Waiting for the database...; sleep 10"]containers:- name:nginximage:nginx:1.29- name:checkerimage:busybox:1.37command:["sh","-c","trap 'exit 0' TERM; while true; do wget -q -O /dev/null http://localhost && echo nginx is up; sleep 5; done"]
Create the Pod:
kubectl apply -f web.yaml
pod/web created
Watch it right away:
kubectl get pod web --watch
NAME READY STATUS RESTARTS AGE
web 0/2 Init:0/1 0 0s
web 0/2 PodInitializing 0 11s
web 2/2 Running 0 18s
The init container needs about 10 seconds, so Init:0/1 stays for a while. Once READY shows 2/2 (after about 20
seconds), stop the watch with Ctrl+C. Show the output of the container checker:
kubectl logs web -c checker
nginx is up
nginx is up
...
Explanation
The Pod starts its containers in two stages:
The init container wait runs first and has to end successfully. Here it only sleeps for 10 seconds; in a real
app, it would wait until the database is reachable. Meanwhile, the status is Init:0/1: 0 of 1 init containers
have finished.
Then the containers nginx and checker start together. While they are being created, the status is
PodInitializing. READY 2/2 means both of them run.
All containers of a Pod share the same network. That’s why checker reaches nginx at localhost. A helper
container like checker that runs next to the main container is called a sidecar.
trap 'exit 0' TERM lets checker end within a few seconds when Kubernetes stops the Pod. Without it, deleting
the Pod takes 30 seconds: Kubernetes waits that long for a container to end before it kills it.
Without -c, kubectl logs shows the first container, nginx. Use -c to choose another one.
Step 4: Organize Pods with Labels
Add labels to the metadata of both manifests. In countdown.yaml:
NAME READY STATUS RESTARTS AGE LABELS
countdown 0/1 CrashLoopBackOff 3 (13s ago) 90s app=countdown,env=dev
web 2/2 Running 0 30s app=web,env=dev
countdown keeps restarting, so its status may also be Running or Completed. Your RESTARTS and AGE differ.
Select the Pods with the label app=web:
kubectl get pods -l app=web
NAME READY STATUS RESTARTS AGE
web 2/2 Running 0 30s
Select the Pods with the label env=dev:
kubectl get pods -l env=dev
NAME READY STATUS RESTARTS AGE
countdown 0/1 CrashLoopBackOff 3 (13s ago) 90s
web 2/2 Running 0 30s
Explanation
Labels are key-value pairs in the metadata. You choose the keys and values yourself. configured means that
Kubernetes updated the existing Pods. Labels can be changed while a Pod runs.
-l takes a selector and returns all Pods with matching labels, regardless of their names. Kubernetes uses
selectors in the same way to find the Pods that belong together.
Step 5: Separate Pods with Namespaces
Create a namespace:
kubectl create namespace playground
namespace/playground created
Create a second Pod web in it:
kubectl apply -f web.yaml -n playground
pod/web created
List the Pods in the new namespace:
kubectl get pods -n playground
NAME READY STATUS RESTARTS AGE
web 0/2 Init:0/1 0 0s
And in the namespace default:
kubectl get pods
NAME READY STATUS RESTARTS AGE
countdown 0/1 CrashLoopBackOff 3 (13s ago) 90s
web 2/2 Running 0 30s
Delete the namespace. This takes about 10 seconds:
kubectl delete namespace playground
namespace "playground" deleted
Check that its Pod is gone:
kubectl get pods -n playground
No resources found in playground namespace.
Explanation
A Namespace divides a cluster into separate areas. Names only have to be unique within a namespace,
so both Pods can be called web. Without -n, kubectl works in the namespace default.
Deleting a namespace deletes everything in it.
Optional Tasks
Finished early? Try these tasks. The following labs don’t depend on them. Task 3 lists the new commands it needs.
Open the solution if you get stuck.
Task 1: Second Chance
Set restartPolicy: OnFailure in countdown.yaml. Does the countdown restart now? What changes when
its command ends with exit 1?
You know all the commands you need. Remember to delete the Pod before you apply the changed file.
Solution
Change the restartPolicy to OnFailure, then delete the Pod:
kubectl delete -f ~/labs/2.3/countdown.yaml
pod "countdown" deleted from default namespace
Create it again:
kubectl apply -f ~/labs/2.3/countdown.yaml
pod/countdown created
Watch it:
kubectl get pod countdown --watch
NAME READY STATUS RESTARTS AGE
countdown 0/1 ContainerCreating 0 0s
countdown 1/1 Running 0 1s
countdown 0/1 Completed 0 11s
The countdown isn’t restarted. Now make its command fail with exit 1 at the end:
command:["sh","-c","for i in 5 4 3 2 1; do echo $i; sleep 2; done; exit 1"]
OnFailure restarts a container only when it fails, i.e. when it ends with an exit code other than 0. It’s the right
policy for tasks that should be retried until they succeed.
Task 2: Stuck at the Gate
Create a Pod gate with an init container that fails. Does the main container start? A container fails when its
command ends with an exit code other than 0, e.g. exit 1.
Solution
Create the file gate.yaml in ~/labs/2.3:
apiVersion:v1kind:Podmetadata:name:gatespec:initContainers:- name:checkimage:busybox:1.37command:["sh","-c","echo Database not reachable; exit 1"]containers:- name:nginximage:nginx:1.29
Create the Pod:
kubectl apply -f ~/labs/2.3/gate.yaml
pod/gate created
Watch it:
kubectl get pod gate --watch
NAME READY STATUS RESTARTS AGE
gate 0/1 Init:0/1 0 0s
gate 0/1 Init:Error 0 5s
gate 0/1 Init:CrashLoopBackOff 1 (2s ago) 7s
gate 0/1 Init:Error 2 (15s ago) 20s
...
The container nginx never starts: a Pod only starts its containers after all init containers ended successfully.
Kubernetes keeps retrying the init container because of the default restartPolicy: Always. With Never, the Pod
would fail instead. The output of the init container shows why it fails:
kubectl logs gate -c check
Database not reachable
Task 3: Label Maker
Show the labels app and env as columns. Add the label tier=frontend to the Pod web, select the Pod by this
label and remove the label again.
Command
Effect
kubectl get pods -L <key>,<key>
Shows the labels as columns
kubectl label pod <name> <key>=<value>
Adds a label to a Pod
kubectl label pod <name> <key>-
Removes a label from a Pod
Solution
Show the labels as columns:
kubectl get pods -L app,env
NAME READY STATUS RESTARTS AGE APP ENV
countdown 0/1 CrashLoopBackOff 6 (3m41s ago) 10m countdown dev
web 2/2 Running 0 8m47s web dev
Add the label:
kubectl label pod web tier=frontend
pod/web labeled
Select the Pod by the new label:
kubectl get pods -l tier=frontend
NAME READY STATUS RESTARTS AGE
web 2/2 Running 0 8m47s
Remove the label again:
kubectl label pod web tier-
pod/web unlabeled
kubectl label changes the Pod directly, not your manifest. The next kubectl apply -f web.yaml doesn’t remove a
label that you added this way.
Clean Up
Delete all Pods of this lab, since the following labs don’t need them:
kubectl delete -f ~/labs/2.3/
pod "countdown" deleted from default namespace
pod "web" deleted from default namespace
If you did Task 2, the Pod gate is deleted as well.
Complete files
~/labs/2.3/countdown.yaml:
apiVersion:v1kind:Podmetadata:name:countdownlabels:app:countdownenv:devspec:restartPolicy:Alwayscontainers:- name:countdownimage:busybox:1.37command:["sh","-c","for i in 5 4 3 2 1; do echo $i; sleep 2; done"]
~/labs/2.3/web.yaml:
apiVersion:v1kind:Podmetadata:name:weblabels:app:webenv:devspec:initContainers:- name:waitimage:busybox:1.37command:["sh","-c","echo Waiting for the database...; sleep 10"]containers:- name:nginximage:nginx:1.29- name:checkerimage:busybox:1.37command:["sh","-c","trap 'exit 0' TERM; while true; do wget -q -O /dev/null http://localhost && echo nginx is up; sleep 5; done"]