Autoscaling¶

Starting state: You have deployed hello and received Hello World!. Keep the same cluster and default namespace.

Knative Serving provides automatic scaling, also known as autoscaling. This means that a Knative Service by default scales down to zero running pods when it is not in use.

List your Knative Service¶

Use the Knative (kn) CLI to view the URL where your Knative Service is hosted:

kn service list

Expected result

hello is listed with READY set to True and a URL. Use the reported URL rather than copying an example address.

Inspecting with kubectl instead

kubectl get ksvc also lists the Service. This tutorial still uses kn for requests and updates; the YAML examples are supporting reference, not a separate CLI-free route.

Access your Knative Service¶

Access your Knative Service by opening the previous URL in your browser or by running the command:

echo "Accessing URL $(kn service describe hello -o url)"
curl "$(kn service describe hello -o url)"

Expected output

Hello World!

Are you seeing curl: (6) Could not resolve host: hello.default.${LOADBALANCER_IP}.sslip.io?

In some cases your DNS server may be set up not to resolve *.sslip.io addresses. If you encounter this problem, it can be fixed by using a different nameserver to resolve these addresses.

The exact steps will differ according to your distribution. For example, with Ubuntu derived systems which use systemd-resolved, you can add the following entry to the /etc/systemd/resolved.conf:

[Resolve]
DNS=8.8.8.8
Domains=~sslip.io.

Then simply restart the service with sudo service systemd-resolved restart.

For MacOS users, you can add the DNS and domain using the network settings as explained here.

Observe autoscaling¶

Watch the pods and see how they scale to zero after traffic stops going to the URL:

kubectl get pod -l serving.knative.dev/service=hello -w

Note

It may take up to 2 minutes for your pods to scale down. Pinging your service again resets this timer.

Expected output

NAME                                     READY   STATUS
hello-world                              2/2     Running
hello-world                              2/2     Terminating
hello-world                              1/2     Terminating
hello-world                              0/2     Terminating

Scale up your Knative Service¶

Leave the watch running. In a second terminal using the same cluster and namespace, request the Service again:

curl "$(kn service describe hello -o url)"

You should receive Hello World!. In the first terminal, you can see a new pod running again:

Expected output

NAME                                     READY   STATUS
hello-world                              0/2     Pending
hello-world                              0/2     ContainerCreating
hello-world                              1/2     Running
hello-world                              2/2     Running

Pod names and intermediate status lines vary; the important observations are that no hello pod remains after idle time and a new pod appears when you request the Service again.

Exit the watch with Ctrl+C.

Continue to traffic splitting below to finish the chosen tutorial route. After the exercise, use these direct handoffs for your own Service:

Task: Set a concurrency target.
Explanation: How concurrency affects autoscaling.
Reference: Concurrency settings and scale bounds.

The Serving autoscaling overview routes to other scaling topics.

Previous: Deploy a Service · Next: Split traffic between Revisions