gcloud compute instances create nginx \
--image-family gci-stable \
--image-project google-containers \
--machine-type n1-standard-1
Login into the nginx VM
gcloud compute ssh nginx
Start the Kubernetes kubelet; which will manage our containers:
sudo systemctl enable kubelet
sudo systemctl start kubelet
Create a Pod manifest to run the nginx container under the Kubernetes kubelet.
cat > nginx.yaml << EOF
apiVersion: v1
kind: Pod
metadata:
name: nginx
spec:
hostNetwork: true
containers:
- name: nginx
image: nginx:1.10.1
EOF
Move the nginx pod manifest into place:
sudo mv nginx.yaml /etc/kubernetes/manifests/
Allow HTTP traffic
sudo iptables -A INPUT -p tcp --dport 80 -j ACCEPT
Exit out of the nginx VM and run the following commands to test the nginx instance
gcloud compute firewall-rules create nginx --allow tcp:80
NGINX_PUBLIC_ADDRESS=$(gcloud compute instances describe nginx \
--format 'value(networkInterfaces[0].accessConfigs[0].natIP)')
curl http://${NGINX_PUBLIC_ADDRESS}
cat > cloud-init << EOF
#cloud-config
write_files:
- path: etc/kubernetes/manifests/nginx.yaml
permissions: 0644
owner: root
content: |
apiVersion: v1
kind: Pod
metadata:
name: nginx
spec:
hostNetwork: true
containers:
- name: nginx
image: nginx:1.10.1
runcmd:
- systemctl daemon-reload
- systemctl start kubelet
- iptables -A INPUT -p tcp --dport 80 -j ACCEPT
- iptables-save
EOF
gcloud compute instances create nginx-cloud-init \
--image-family gci-stable \
--image-project google-containers \
--machine-type n1-standard-1 \
--metadata-from-file user-data=cloud-init
Testing the nginx-cloud-init VM:
NGINX_CI_PUBLIC_ADDRESS=$(gcloud compute instances describe nginx-cloud-init \
--format 'value(networkInterfaces[0].accessConfigs[0].natIP)')
$ curl http://${NGINX_CI_PUBLIC_ADDRESS}
Cloud-init is greate for automating the creation of a single VM, but what if you need say 100 VMs just like this one?
$ gcloud compute instance-templates create nginx-instance-template \
--image-family gci-stable \
--image-project google-containers \
--machine-type n1-standard-1 \
--metadata-from-file user-data=cloud-init
$ gcloud compute instance-groups managed create nginx-instance-group \
--base-instance-name nginx \
--size 3 \
--template nginx-instance-template
NAME LOCATION SCOPE BASE_INSTANCE_NAME SIZE TARGET_SIZE INSTANCE_TEMPLATE AUTOSCALED
nginx-instance-group us-west1-b zone nginx 0 3 nginx-instance-template no
$ gcloud compute instances list
NAME ZONE MACHINE_TYPE PREEMPTIBLE INTERNAL_IP EXTERNAL_IP STATUS
nginx-7dp9 us-west1-b n1-standard-1 10.240.0.2 104.199.112.124 RUNNING
nginx-sipu us-west1-b n1-standard-1 10.240.0.4 104.199.114.154 RUNNING
nginx-zs7v us-west1-b n1-standard-1 10.240.0.3 104.199.124.134 RUNNING
$ gcloud compute instance-groups managed resize nginx-instance-group --size 5
$ gcloud compute instances list
NAME ZONE MACHINE_TYPE PREEMPTIBLE INTERNAL_IP EXTERNAL_IP STATUS
nginx-7dp9 us-west1-b n1-standard-1 10.240.0.2 104.199.112.124 RUNNING
nginx-o68k us-west1-b n1-standard-1 10.240.0.6 104.199.114.46 RUNNING
nginx-sipu us-west1-b n1-standard-1 10.240.0.4 104.199.114.154 RUNNING
nginx-sz84 us-west1-b n1-standard-1 10.240.0.5 104.199.116.59 RUNNING
nginx-zs7v us-west1-b n1-standard-1 10.240.0.3 104.199.124.134 RUNNING
Scaling back down
$ gcloud compute instance-groups managed resize nginx-instance-group --size 1
$ gcloud compute instances list
NAME ZONE MACHINE_TYPE PREEMPTIBLE INTERNAL_IP EXTERNAL_IP STATUS
nginx-zs7v us-west1-b n1-standard-1 10.240.0.3 104.199.124.134 RUNNING
Start with the instance group size set to 1:
gcloud compute instance-groups managed resize nginx-instance-group --size 1
Set an autoscaling policy:
gcloud compute instance-groups managed set-autoscaling nginx-instance-group \
--max-num-replicas 20 \
--target-cpu-utilization 0.05 \
--cool-down-period 90
Notice the nginx-instance-group has autoscale enabled.
$ gcloud compute instance-groups managed list
NAME ZONE BASE_INSTANCE_NAME SIZE TARGET_SIZE INSTANCE_TEMPLATE AUTOSCALED
nginx-instance-group us-west1-b nginx 1 2 nginx-instance-template yes
The default min size of an instance group is 2. Notice how the instance group has been autoscaled to meet the target size of 2:
$ gcloud compute instances list
NAME ZONE MACHINE_TYPE PREEMPTIBLE INTERNAL_IP EXTERNAL_IP STATUS
nginx-o3xl us-west1-b n1-standard-1 10.240.0.2 104.199.114.154 RUNNING
nginx-zs7v us-west1-b n1-standard-1 10.240.0.3 104.199.124.134 RUNNING
Install a load tester:
go get -u github.com/rakyll/hey
Set the public IP address:
NGINX_PUBLIC_ADDRESS=$(gcloud compute instances describe nginx-zs7v \
--format 'value(networkInterfaces[0].accessConfigs[0].natIP)')
Test the connection:
$ curl http://${NGINX_PUBLIC_ADDRESS}
Send load to the nginx instance:
hey -n 2000 http://${NGINX_PUBLIC_ADDRESS}
Notice the number of VMs have scaled to 4 because of the CPU spike caused by the hey command:
gcloud compute instances list
NAME ZONE MACHINE_TYPE PREEMPTIBLE INTERNAL_IP EXTERNAL_IP STATUS
nginx-o3xl us-west1-b n1-standard-1 10.240.0.2 104.199.114.154 RUNNING
nginx-ot4b us-west1-b n1-standard-1 10.240.0.5 104.199.112.124 RUNNING
nginx-sf9m us-west1-b n1-standard-1 10.240.0.4 104.198.109.21 RUNNING
nginx-zs7v us-west1-b n1-standard-1 10.240.0.3 104.199.124.134 RUNNING