Best Practices, Common Issues and Gotchas

Best Practices

Run Escalator with --drymode enabled when initially deploying it to a production environment to get an idea of how it will react before letting it taint/untaint and terminate instances.
Initially run Escalator with safe or moderate threshold values when deploying it to production for the first time, tune the values accordingly once Escalator is running as desired in your environment.
Run it with a high hard_delete_grace_period timeout to prevent Escalator terminating nodes that are still running workloads.
Run Escalator with a low scan interval, for example 30 or 60 seconds. This will ensure Escalator is responsive enough during a spike in load or pods.
Run Escalator with leader election enabled, in a HA deployment (that is, >1 replica). Turn this behaviour on with --leader-elect, see the Command line options docs for more info. You can also inspect the leader events with kubectl describe lease <lease-name>, where the default Lease name is escalator-leader-elect.

Ensure scale in protection is enabled for the cloud provider node group. This will prevent the cloud provider from terminating instances in the rare case where there are more instances than the desired node group size.
It is recommended to match Escalator min_nodes and max_nodes to the value in the cloud provider. This will prevent weird cases where Escalator will try to scale down but will be blocked by the cloud provider.
Escalator only supports one cloud provider per deployment. You will need to run multiple different deployments of Escalator inside the cluster to use more than one cloud provider.
Escalator only supports one cloud provider region per deploy. As with running multiple cloud providers, multiple different deployments of Escalator will need to be used.
Escalator only supports running the same type of instance (e.g. instances with different CPU and memory configurations) in a node group in the cloud provider. If you would like to use different types of instances and still have have Escalator manage them, you will need to place each instance type in it's own node group in the cloud provider.