with Terraform, Exoscale, Prometheus, Grafana
https://community.exoscale.com/documentation/
https://www.terraform.io/docs/index.html
https://www.exoscale.com/syslog/network-load-balancer-cli-tutorial/
https://gist.github.com/janoszen/7ced227c54d1c9e86a9c1cbd93a451f2
https://www.exoscale.com/syslog/autoscaling-with-grafana-and-prometheus/
https://github.com/exoscale/terraform-provider-exoscale/blob/v0.20.0/examples/instance-pool/main.tf
https://github.com/exoscale/terraform-provider-exoscale/blob/v0.20.0/examples/nlb/main.tf
https://github.com/prometheus/node_exporter
https://quay.io/repository/prometheus/node-exporter?tag=latest&tab=tags
https://github.com/janoszen/prometheus-sd-exoscale-instance-pools
https://prometheus.io/docs/prometheus/latest/installation/
https://prometheus.io/docs/prometheus/latest/getting_started/
https://fh-cloud-computing.github.io/exercises/3-containers/
http://docs.docker.oeynet.com/machine/drivers/exoscale/#environment-variables-and-default-values
https://github.com/janoszen/exoscale-grafana-autoscaler
http://httpd.apache.org/docs/2.0/programs/ab.html
https://quay.io/repository/janoszen/http-load-generator?tag=latest&tab=tags
https://fh-cloud-computing.github.io/exercises/5-grafana/
https://grafana.com/docs/grafana/latest/administration/provisioning/
https://grafana.com/tutorials/provision-dashboards-and-data-sources/#2
This repository contains a project work and its 3 stages or sprints which are built on each other: Sprint 1 Creating instance pool and Network Load Balancer (NLB) Sprint 2 Monitoring with Prometheus Sprint 3 Grafana and Autoscaling
https://www.exoscale.com/syslog/autoscaling-with-grafana-and-prometheus/
Similar: https://fh-cloud-computing.github.io/projectwork/
Register and log into Exoscale, go to IAM and create an API Key. The key and secret needs to be copy-pasted in variables.tf and its placeholders ${var.exoscale_key} and ${var.exoscale_secret}.
Before starting, let's check if Terraform is healthy with:
$ terraform plan
Once we’re happy with the plan, we can set it into action:
$ terraform apply
Look up the instance-pool ID in the Exoscale Console under Compute > Instance Pools and copy-paste it in the placeholder ${exoscale_instance_pool.cc-instance-pool.id}.
Update with:
$ terraform apply
Network Load Balancer
Copy the NLB-IP from Exoscale Console under Compute > Load Balancers and paste it in a browser window to check the following:
http://<NLB-IP>/health displays OK
http://<NLB-IP>/load displays Load generation done.
Prometheus
Copy the Monitoring-instance-IP from Exoscale Console under Compute > Instances and paste it in another browser window or tab to do and check the following:
http://<monitoring-instance-IP>:9090 displays Prometheus UI. Under Status > Targets are the instances of the instance pool
and the monitoring-instance itself displayed. Under Graph the metrics can be filtered and visualised in a graph such as:
Inserting queries like sum by (instance) (rate(node_cpu_seconds_total{mode!="idle"}[1m])) / sum by (instance) (rate(node_cpu_seconds_total[1m])) for each instance or (sum (rate(node_cpu_seconds_total{mode!=\"idle\", job=\"exoscale\"}[1m])) / sum (rate(node_cpu_seconds_total{job=\"exoscale\"}[1m]))) for all instances in the instance-pool in the metrics filter, the graph does not really change. To see the CPU usage increase and decrease, apply load by:
http://<NLB-IP>/load
Grafana and Autoscaling
http://<monitoring-instance-IP>:3000 displays Grafana UI. For first login, username admin with password admin can be used. Go to Configuration (gear wheel icon) > Data sources, select Prometheus and insert the monitoring-instance-IP for its placeholder 'localhost'. Same for the notification channels (webhooks) Scale up and Scale down under Alerting (bell icon) > Notification channels. Now, everything is ready for testing. Go to Dashboards (4 tiles icon) and open the dashboard Annotations & Alerts with its two panels High CPU usage and Low CPU usage.
Again, you can apply load by:
http://<NLB-IP>/load
Or run a bash script like:
for i in {1..n}
do
curl -X GET "http://<NLB-IP>/load"
echo "\n"
done
Or use Apache Benchmark Tool, e.g.:
ab -c 4 -n 1000 http://<NLB_IP>/load
Either ways, the average CPU usage of the instance-pool and the graphs increase. After alerting, new instances are created in the instance-pool one after another if the average CPU usage stays over 80%. When the load stops, the average CPU usage decrease. After alerting, an instance in the instance-pool is deleted one after another if the average CPU usage stays under 20%.
After investigating and testing, destroy with:
$ terraform destroy