Skip to content

Latest commit

 

History

History
102 lines (76 loc) · 5.59 KB

README.md

File metadata and controls

102 lines (76 loc) · 5.59 KB
Raw

tailscale-load-balancer-controller

A k8s load balancer service controller which allows you to expose kubernetes services to a tailscale network

Description

For every service of type LoadBalancer, the controller will launch a tailscale connected pod && forward traffic from that pod to the service. Additionally, the controller will populate the externalIPs field of the service with the assigned tailscale IP.

Getting Started

Requirements (hard codes that need to be broken out at somepoint):

  • a namespace with the name tailscale
  • a secret in that namespace with the name tailscale-token with a key of token that holds an tailscale auth key (preferably ephemeral)

Deployment Options:

  • make deploy will use kustomize to template the artifacts into your cluster
  • make template > your-spot-for-config.yaml will just template out the artifacts, allowing you to place them whereever fits into your deployment pipeline
  • helm chart, see below

Helm

Currently there is a helm chart, but its relatively unconfigurable as its just make template stored under templates/raw.yaml. In the future, having a full fledged helm chart could definetly be worth while, so as needs arise, sections may be broken out of raw.yaml into a normal, configurable helm template.

To Use:

helm repo add tailscale-load-balancer-controller https://pthomison.github.io/tailscale-load-balancer-controller
helm repo update
helm install tailscale-load-balancer-controller tailscale-load-balancer-controller/tailscale-load-balancer-controller

ToDo

  • stop using "latest" for the deployed LB pod
  • better helm chart
  • GH actions work, make sure image & chart publishing is working
  • Better "ip-updater" solution/loop
  • Configurable userspace vs kernel wireguard
  • Stop having trash commit messages on main
  • Scope down service account permissions
  • Separate service account for lb pods
  • Garbage Collection of lb pods
  • Easy way to toggle the use of dev images vs dockerhub
  • Testing
  • Configurable Namespace
  • Configurable secret/key
  • Using a NS scoped cache instead of the Uncached Client
  • Better logging
    • Reasonable log updates as the system operates
    • Use a logger instead of printing to stdout
  • Load testing
  • Requeue LB services for a duration && ensure load-balancers are plumbed correctly
  • Reqs/Limits for tlb pods

How it works

This project aims to follow the Kubernetes Operator pattern

It uses Controllers which provides a reconcile function responsible for synchronizing resources untile the desired state is reached on the cluster

More information can be found via the Kubebuilder Documentation

Learnings So Far

I'm still fairly new to the kubebuilder/controller-runtime scene, so as I've built out the projects I've learned some new stuff

  • Kubebuilder can be used to service any k8s API object (ie not just CRDs).

    • Somewhat the motivation for this project, I've been very curious how third party controllers integrate into the k8s ecosystem (ie ingress-controllers, cloud controllers, etc)
    • To wrap a kubebuilder project around a non-crd, do the normal kubebuilder init && then modify the controllers SetupWithManager function && add For(&k8sapi.object{}) example

  • By default, the manager will attempt to watch/list all cluster objects of a given Kind, even if the request is scoped to a namespace.

    • The cache can be namespaced, but afaict this only happens at the manager level so theres no built-in way to have different cache policies per object.
    • My current solution to this is to use an uncached client that I've attached to my reconciler example
    • A potentially better solution is to add another cache w/ NS restrictions, but tbd

Open Questions

  • Is there a more compact method of representing objects you want the controller to template into a cluster?

    • Currently have "Render" functions that return the go objects (ie corev1.Pod) with fields populated, but this is fairly clunky for adding/changing objects
    • Maybe some codegen tool that can take in yaml k8s manifests?

  • Is there a decent way to wrap my "Ensure" functions to handle k8s objects generically?

    • k8s interface?
    • generics?

  • When to use labels vs annotations?

    • I've generally understood that annotations are for computer generated items && labels were for human consumption, but some functions seem to work better with labels out of the box

  • Can I watch a single object or is watch exclusive to a List?

License

Copyright 2023.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.