KDynaRoll

A Kubernetes Controller for dynamically managing application behavior through labels.

Features

• Assigns labels to pods based on percentage distributions.
• Supports multiple labels per pod.
• Automatically maintains label distribution as the number of pods changes.
• Exposes controller metrics for Prometheus-compatible scrapers.

Applications

KDynaRoll was originally created to address a very particular use case, combining pod labels and feature flags (for use with tools like Unleash, Flagsmith, Harness, etc).

This is a slightly more generic version of the original use case and it hasn't been stress tested as much, hence the beta status.

I'd very much like to hear about other cases and requirements you may have. Open an issue I'll check it out!

Here are some other applications for KDynaRoll:

• A/B testing or canary deployments:
• Resource segmentation: dynamically assign pods to resource groups based on specific needs or quotas, then schedule them differently based on these labels.
• Dynamically change internal or external ingress access to services
• Fault Tolerance Testing: assign failure-injection labels to simulate real-world failure scenarios
• Cost Optimization: use labels to represent different classes of service (e.g., spot-instances, reserved-instances), helping allocate cheaper resources to less critical pods.
• Security/Compliance Segmentation: In multi-tenant or compliance-focused environments, apply different policies through labels.

The Kubernetes Downward API can be used to expose pod labels to applications.

Using a docker image

Images are available on dockerhub under biraignacio/kdynaroll.

docker pull biraignacio/kdynaroll

It's heavily rate limited so consider building from source and and deploying to your local docker repository.

Building

To build from source, you need a Rust compiler with at least edition 2021. After building the binary, create and publish a Docker image to your repository.

cargo build --release

# specify the repository and tag
docker build -t <my.repository>:<port>/kdynaroll:<tag> target/release/ -f Dockerfile

docker push <my.repository>:<port>/kdynaroll:<tag>

Getting started

Installation

Install KDynaRoll using the provided helm chart. Configure KDynaRoll via environment variables, defined in values.yaml.

kdynaroll:
  env:
    logLevel: "INFO" # log level
    metricsBindIpAddr: "0.0.0.0" # address the metrics server will listen to
    metricsBindPort: 5085 # metrics server port

You should also specify the docker repository and tag for pulling the KDynaRoll image. (See build instructions above)

image:
  repository: my.docker.registry:5000/kdynaroll
  tag: v0.1.0

A new Custom Resource Definition (CRD) named kdynarolls.shiftflow.ml will then be created.

Specifying targets labels to be applied

After installation, create a KDynaRoll resource to specify the target applications where the labels will be applied. Labels don't need to be predefined.

Label name is, well the name of the label. A label can have multiple values, distributed across pods using the defined percentages.

KDynaRoll supports multiple namespaces, applications, and labels. Each label's values must have percentages between 0 and 100, with their total adding up to 100%.

A KDynaRoll definition has the following structure:

apiVersion: shiftflow.ml/v1
kind: Kdynaroll
metadata:
  name: kdynaroll-controller-config
  namespace: <your namespace>
spec:
  podLabeler:
    namespaces: # list of namespaces
    - namespace: <namespace>
      apps: # list of apps
      - app: <app name>
        labels: # list of labels to be applied
        - labelName: <label name>
          labelValueDistribution: # list of possible values for this label
          - value: <label value 1>
            percentage: <percentage for this value as a floating point > 0 and <= 100.0>

Make sure the name of the configuration (in metadata.name) is kdynaroll-controller-config. For the moment, that's a requirement.

metadata:
  name: kdynaroll-controller-config

KDynaRoll currently applies only to applications with the label app. Ensure your target applications include this label.

See loadbalancer for an example CRD config file.

FAQ

What is the best way to make coffee?

It's a matter of personal preference. What works for you is the best method.

Why Rust?

A: Personally, I prefer French press coffee.