docs/serving/samples/autoscale-go

Autoscale Sample

A demonstration of the autoscaling capabilities of an Knative Serving Revision.

Prerequisites

1. Install Knative Serving
2. Install docker

Setup

Build the autoscale app container and publish it to your registry of choice:

REPO="gcr.io/<your-project-here>"

# Build and publish the container, run from the root directory.
docker build \
  --build-arg SAMPLE=autoscale \
  --tag "${REPO}/sample/autoscale" \
  --file=sample/Dockerfile.golang .
docker push "${REPO}/sample/autoscale"

# Replace the image reference with our published image.
perl -pi -e "s@github.com/knative/docs/serving/samples/autoscale-go@${REPO}/sample/autoscale@g" sample/autoscale/sample.yaml

# Deploy the Knative Serving sample
kubectl apply -f sample/autoscale/sample.yaml

Export your ingress IP as SERVICE_IP.

# Put the ingress Host name into an environment variable.
export SERVICE_HOST=`kubectl get route autoscale-route -o jsonpath="{.status.domain}"`

# Put the ingress IP into an environment variable.
export SERVICE_IP=`kubectl get svc knative-ingressgateway -n istio-system -o jsonpath="{.status.loadBalancer.ingress[*].ip}"`

Request the largest prime less than 40,000,000 from the autoscale app. Note that it consumes about 1 cpu/sec.

time curl --header "Host:$SERVICE_HOST" http://${SERVICE_IP?}/primes/40000000

Running

Ramp up a bunch of traffic on the autoscale app (about 300 QPS).

kubectl delete namespace hey --ignore-not-found && kubectl create namespace hey
for i in `seq 2 2 60`; do
  kubectl -n hey run hey-$i --image josephburnett/hey --restart Never -- \
    -n 999999 -c $i -z 2m -host $SERVICE_HOST \
    "http://${SERVICE_IP?}/primes/40000000"
  sleep 1
done
watch kubectl get pods -n hey --show-all

Watch the Knative Serving deployment pod count increase.

watch kubectl get deploy

Look at the latency, requests/sec and success rate of each pod.

for i in `seq 4 4 120`; do kubectl -n hey logs hey-$i ; done | less

Cleanup

kubectl delete namespace hey
kubectl delete -f sample/autoscale/sample.yaml