Developing Applications on Multi-tenant Clusters with GitOps
In this tutorial, Stefan Prodan discusses how to manage deployments with GitOps using Flux with Kustomize on a multi-tenant cluster. It also includes steps on how to implement Flagger to automate progressive delivery strategies like canaries, and A/B deployments through namespaces.
To maximize resources and minimize costs when developing applications on Kubernetes, you may opt to use a single multi-tenant cluster split into namespaces. The use of namespaces allows your teams to develop applications and maybe even QA them on a single cluster. Fewer running clusters can help prevent costs from spiralling out of control and more seriously, can reduce security breaches that may occur with multiple unattended clusters running throughout your organization.
In this tutorial, Stefan Prodan (@stefanprodan) describes how to manage deployments with GitOps using Flux and Kustomize on a multi-tenant cluster split into namespaces. But before we dive into the details, let’s define some of the terms and tools that we’ll be using in this tutorial:
- Flux - Flux is a GitOps operator for continuous delivery that automatically ensures that the state of a cluster matches the config in git. It uses an operator in the cluster to trigger deployments inside Kubernetes. means you don't need a separate CD tool. Flux monitors all relevant image repositories, detects new images, triggers deployments and updates the desired running configuration based on that (and a configurable policy).
- Namespaces and Multi-tenant clusters - Kubernetes challenges the way we have traditionally thought about development environments. Kubernetes clusters have a built-in feature that allows you to share a cluster among different environments and between different projects on separate teams using namespaces. A namespace is a way to divide the resources of one cluster amongst teams.
- Kustomize - lets you customize raw, template-free YAML files and use them for multiple purposes, while at the same time leaving the original YAML untouched and usable as is.
- Flagger - an open source tool that automates progressive delivery strategies like canary, A/B and other more complex deployments.
fluxcd-multi-tenancy repository serves as a starting point for a multi-tenant cluster managed with Git, Flux and Kustomize.
I'm assuming that a multi-tenant cluster is shared by multiple teams. The cluster wide operations are performed by the cluster administrators while the namespace scoped operations are performed by various teams each with their own Git repository. This means a team member, who is not a cluster admin, can't create namespaces, custom resources definitions or change something in another team’s namespace.
#1. Create two git repositories: one for cluster admins and another for teams.
Create two git repositories:
- Clone the fluxcd-multi-tenancy repository for the cluster admins. This will be referred to as
- Clone the fluxcd-multi-tenancy-team1 repository for the dev team1. This repo will be referred to it as
|Team||Namespace||Git Repository||Flux RBAC|
|ADMIN||all||Cluster wide e.g. namespaces, CRDs, Flux controllers|
|DEV-TEAM1||team1||Namespace scoped e.g. deployments, custom resources|
|DEV-TEAM2||team2||Namespace scoped e.g. ingress, services, network policies|
Cluster Admin repository structure:
base folder holds the deployment spec used for installing Flux in the
flux-system namespace and in the teams namespaces. All Flux instances share the same Memcached server deployed at install time in
.flux.yaml, configure Flux to run the Kustomize build on the cluster dir and to deploy the generated manifests:
Repository Structure for development team1:
workloads folder contains the desired state of the
team1 namespace and the
flux-patch.yaml contains the Flux annotations that define how the container images should be updated.
.flux.yaml configure Flux to run the Kustomize build, apply the container update policies and to deploy the generated manifests:
#2 Install the cluster admin Flux agent
In the dev-cluster repo, change the git URL to point to your fork:
vim ./install/flux-patch.yaml --firstname.lastname@example.org:org/dev-cluster
Install the cluster-wide Flux with kubectl kustomize:
kubectl apply -k ./install/
Get the public SSH key with:
fluxctl --k8s-fwd-ns=flux-system identity
Add the public key to the
github.com:org/dev-cluster repository deploy keys with write access.
The cluster-wide Flux does the following:
- creates the cluster objects from
cluster/commondirectory (CRDs, cluster roles, etc)
- creates the
team1namespace and deploys a Flux instance with restricted access to that namespace
#3 Install a Flux agent per team
Change the dev team1 git URL:
vim ./cluster/team1/flux-patch.yaml --email@example.com:org/dev-team1
After committing your changes, the system Flux configures the team1's Flux to sync with
Get the public SSH key for team1 with:
fluxctl --k8s-fwd-ns=team1 identity
Add the public key to the
github.com:org/dev-team1 deploy keys with write/access. The team1's Flux applies the manifests from
org/dev-team1 repository only in the
team1 namespace, which is enforced with RBAC and role bindings.
If team1 needs to deploy a controller that depends on a CRD or a cluster role, they'll have to open a PR in the
org/dev-cluster repository and add those cluster wide objects in the
The team1's Flux instance can be customized with different options than the system Flux using the
k8s-allow-namespace restricts Flux discovery mechanism to a single namespace.
#4 Install Flagger
Flagger is a progressive delivery Kubernetes operator that can be used to automate Canary, A/B testing and Blue/Green deployments.
You can deploy Flagger by including its manifests in the
Commit the changes to git and wait for system Flux to install Flagger and Prometheus:
fluxctl --k8s-fwd-ns=flux-system sync kubectl -n flagger-system get po NAME READY STATUS flagger-64c6945d5b-4zgvh 1/1 Running flagger-prometheus-6f6b558b7c-22kw5 1/1 Running
A team member can now push canary objects to
org/dev-team1 repository and Flagger will automate the deployment process. Flagger can notify your teams when a canary deployment has been initialized, when a new revision has been detected and if the canary analysis failed or succeeded.
Enable Slack notifications by editing the
#5 Configure the pod security policies per team
With pod security policies a cluster admin can define a set of conditions that a pod must run with in order to be accepted into the system. For example you can forbid a team from creating privileged containers or use the host network.
Edit the team1 pod security policy
Set privileged, hostIPC, hostNetwork and hostPID to false and commit the change to git. From this moment on, team1 will not be able to run containers with an elevated security context under the default service account.
If a team member adds a privileged container definition in the
org/dev-team1 repository, Kubernetes will deny it:
kubectl -n team1 describe replicasets podinfo-5d7d9fc9d5 Error creating: pods "podinfo-5d7d9fc9d5-" is forbidden: unable to validate against any pod security policy: [spec.containers.securityContext.privileged: Invalid value: true: Privileged containers are not allowed]
#6 Enforce custom policies per team
Gatekeeper is a validating webhook that enforces CRD-based policies executed by Open Policy Agent.
You can deploy Gatekeeper by including its manifests in the
Inside the gatekeeper dir there is a constraint template that instructs OPA to reject Kubernetes deployments if no container resources are specified.
Enable the constraint for team1 by editing the
Commit the changes to git and wait for system Flux to install Gatekeeper and apply the constraints:
fluxctl --k8s-fwd-ns=flux-system sync kubectl -n gatekeeper-system get po
If a team member adds a deployment without CPU or memory resources in the
org/dev-team1 repository, Gatekeeper will deny it:
kubectl -n flux-system logs deploy/flux admission webhook "validation.gatekeeper.sh" denied the request: [denied by containerresources] container <podinfo> has no memory requests [denied by containerresources] container <sidecar> has no memory limits
#7 Add a new team, namespace and repository
If you want to add another team to the cluster, first create a git repository as
Run the create team script:
./scripts/create-team.sh team2 team2 created at cluster/team2/ team2 added to cluster/kustomization.yaml
Change the git URL in `cluster/team2` dir:
vim ./cluster/team2/flux-patch.yaml --firstname.lastname@example.org:org/dev-team
Push the changes to the master branch of `org/dev-cluster` and sync with the cluster:
fluxctl --k8s-fwd-ns=flux-system sync
Get the team2 public SSH key with:
fluxctl --k8s-fwd-ns=team2 identity
Add the public key to the
github.com:org/dev-team2 repository deploy keys with write access. Team2's Flux applies the manifests from the
org/dev-team2 repository to only the
#8 Isolate tenants
With this setup, Flux prevents a team member from altering cluster level objects or other team's workloads.
To harden tenant isolation, cluster admin should consider using:
- Resource quotas (limit the compute resources that can be requested by a team)
- Network policies (restrict cross namespace traffic)
- Pod security policies (prevent running privileged containers or host network and filesystem usage)
- Open Policy Agent admission controller (enforce custom policies on Kubernetes objects)