K8s autoscaler for pods that consume RabbitMQ

Autoscaling process (autoscale.sh):

• Loops through deployments defined in AUTOSCALING (env var), every INTERVAL (env var) seconds.
• Gets the messages queue on RabbitMQ for the current deployment's queue.
• Calculates the amount of desired pods and scales the deployment if required.
• Every event success or failure will log and notify slack if the SLACK_HOOK env var is set (logging intensity depends on LOGS env var).

This Pod runs in the kube-system namespace on k8s master nodes.

Requirements

• The virtual host in RabbitMQ for queues has to be /.
• Namespace(s), deployment(s) or queue(s) defined in AUTOSCALING env var can't have | or ; symbols in name(s).

Env vars

• INTERVAL: Seconds between checks in the pod autoscaling process described above (default 30s)
• RABBIT_HOST: RabbitMQ host (e.g. http://rabbitmq.example.com)
• RABBIT_USER: Username used for authentication with the RabbitMQ API (check deploy.yml, defaults to rabbit-pod-autoscaler secret, rabbit-user key)
• RABBIT_PASS: Password used for authentication with the RabbitMQ API (check deploy.yml, defaults to rabbit-pod-autoscaler secret, rabbit-pass key)
• AUTOSCALING: Contains min/max pods, messages handled per pod, deployment info and queue name in the following pattern:
  • single deployment to autoscale: <minPods>|<maxPods>|<mesgPerPod>|<k8s namespace>|<k8s deployment name>|<RabbitMQ queue name>
    • e.g. 3|10|5|development|example|example.queue
    • mesgPerPod represents the amount of RabbitMQ messages a Pod can process within the INTERVAL (env var). As an example, if a Pod needs 6s to process a message from RabbitMQ, the mesgPerPod value will be INTERVAL (30s) / process time (6s) = 5.
  • multiple deployments to autoscale: check example deploy.yml
• LOGS: Logging and Slack notifications intensity. Errors are logged and notified on every option
  • HIGH (default): Logs and notifies on every event
  • MEDIUM: Logs and notifies on min pods, avg pods ((min+max)/2) and max pods scale events
  • LOW: Logs and notifies on min pods and max pods scale events
• SLACK_HOOK: Slack incoming webhook for event notifications

Deployment

kubectl --context CONTEXT -n kube-system apply -f deploy.yml