# Kubernetes, Autoscaling, Resourcemanagement

Extraction Plugins can be run in a Kubernetes cluster. This can be the same cluster Hansken run on, or another external
cluster. For each plugin, a *pod* is created by Hansken. Each plugin will have 12 threads by default to process traces
separately within one pod.

## Autoscaling

Hansken will create a Horizontal Pod Autoscaler (*HPA*) for each pod. HPA's manage the number of replica's of a pod
based on metrics. The Extraction Plugins SDK provides two metrics to be set in the PluginInfo:

* Observed CPU utilization
* Observed memory usage

For more info on how to set these metrics follow these links:

* [Specifying system resources (Java)](../java/snippets.md#Specifying system resources)
* [Specifying system resources (Python)](../python/snippets.md#Specifying system resources)

If a pod reaches the CPU or memory usage provided with the PluginInfo, the HPA will increase the number of replicas for
that plugin. Scaling down is done automatically. The maximum number of replicas per pod is specified within Hansken
properties and can be adapted by an operator (needs restart).

## Finding the right settings

This depends on the kubernetes cluster settings, the nodes it runs on, and of course the plugin itself. Monitor the
number of replica's and resource metrics while an extraction is running and adapt accordingly.