Migrating From Py-Kube-Downscaler

In most cases it's very simple to migrate and step 3 and 4 can be skipped.

Step 1:

Uninstall the py-kube-downscaler

helm uninstall py-kube-downscaler

Step 2:

Adjust any old values.yaml files to the new structure.

Step 3:

Make sure all non-official automations support the breaking changes.

You can skip this if you didn't have external automations for the downscaler.

Step 4:

Make sure all timestamps are RFC3339. To make this easier you could also just run the downscaler and see if it gives an error.

Since only some edge cases of the ISO 8601 timestamp format are not compatible you can likely skip this step. Should the downscaler error because of this you can still change them afterwards.

Step 5:

Now you're ready to install the Downscaler. For that you can follow our Installation Guide.

Features

Incompatibility

Some values are now incompatible with each other. This makes it easier to debug unwanted behavior from misconfiguration.

See the breaking changes entry for more information.

Duration Units

Instead of just integer seconds you can now also use duration strings with units. This makes it more intuitive to configure.

Uniform Timestamps

The downscaler now uses RFC3339 timestamps exclusively. This means that you will not have to juggle with multiple timestamp formats.

See the breaking changes entry for more information.

Scopes

The scaling values are now ordered into scopes. This makes it easier to understand where the values are taken from.

See the breaking changes entry for more information.

Explicit Include

Added the --explicit-include cli argument.

This is a simple way to exclude all workloads by default and only add those you want to have scaled.

Comfort Spaces

Lists now allow for spaces in-between the entries.

This allows for lists to be more readable since the entries are separated.

Reverse Relative Timespans

Reverse Timespans can now overlap into the "next day" (20:00-08:00)

This is a faster and less verbose way of matching the opposite of an up/downtime.

Actual Exclusion

Excluding a workload now will ignore scaling the workload, instead of forcing it into uptime. This means that what happens when excluding a workload is more intuitive.

See the breaking changes entry for more information.

Workload Errors

Configuration errors on workloads are now shown as error events on the workload. This makes it so you don't have to have access to the downscalers console just to see whats wrong with your configuration.

Auto Exclude ScaledObject Scaled Workloads

The Workloads scaled by scaledobjects are automatically excluded to avoid conflicts during scaling.

note

This currently only happens when scaled object scaling is enabled via --include-resources

Leader Election

Leader election is now available making it possible to have multiple replicas running redundantly without causing conflicts.

More Workload Types

The downscaler can now also scale:

• prometheus.v1.monitoring.coreos.com

Multiple Timespans in Force* Time

Force uptime and force downtime now allow multiple timespans as their value instead of just one.

Timespans in Exclude

The Exclude value now allows timespans as a value, this allows for more controlled time-based exclusion.

Breaking Changes

From Features

This is a collection of breaking changes caused by a feature:

Feature	Breaking Changes	Severity
Incompatibility	Will now error out and skip scaling if misconfigured	low - might cause some workloads not to be scaled
Actual Exclusion	Excluded workloads will no longer be scaled up	low - if you relied on this quirk
Scopes	Scaling value order is now consistent	medium - could result in other values setting scaling
Uniform Timestamps	RFC3339 timestamps are now used exclusively	medium - some workloads might need to be updated

New Structure of values.yaml

Some fields of the values.yaml file have been restructured.

The following fields would have to be adjusted in an old values.yaml file from the py-kube-downscaler:

• constrainedDownscaler is now not needed anymore
• configMapName is now under configMap.name
• extraConfig is now under configMap.extraConfig