Concurrency

The Downscaler is natively designed to leverage parallel processing for scaling workloads. This capability becomes particularly valuable in large clusters with numerous workloads that need to be scaled down at the same time.

Optimizing CPU Limit

To enable parallelism the Downscaler should be allowed to expand its CPU usage dynamically. This can be achieved by setting the CPU limit to a multiple of available cores.

For example 2000m enables using up to 2 VCPUs, 3000m up to 3 VCPUs, 4000m up to 4 VCPUs and so forth. The number of workloads that can be processed in parallel increases as the number of vCPUs that can be used increases.

The default configuration on the Helm Chart allows the Downscaler to use up to 2 full vCPUs:

resources:
  limits:
    cpu: 2000m
    memory: 900Mi
  requests:
    cpu: 200m
    memory: 300Mi

You should consider whether the default should be adjusted (higher or lower) based on your cluster size and workload to be scaled. Most use cases can be handled with a lower CPU Limit value.

When choosing the correct values, the following considerations should be taken into account:

1. Setting a high CPU limit can lead to usage spikes.
2. The CPU Limit doesn't guarantee that the Downscaler will use the full amount of CPU declared; the pod can expand its CPU usage as long as the resources are available on the node at any given time.

If you need to guarantee a certain amount of CPU resources you should modify the CPU request instead.