s3-operator

This Operator SDK based tool aims at managing S3 related resources (buckets, policies, ...) using a Kubernetes-centric approach. You can set Bucket or Policy custom resources, and let the operator create or update the corresponding bucket/policy on its configured S3 instance.

At a glance

Current S3 providers : Minio
Currently managed S3 resources : buckets, policies

Compatibility

This operator has been successfully tested with :

Kubernetes : 1.25, 1.26, 1.27 (v0.7.0), 1.28 (v0.7.0 and v0.8.0)
MinIO : 2023-05-27T05:56:19Z (up to v0.3.0 included), 2023-11-20T22-40-07Z (from v0.4.0 onwards)

Description

At its heart, the operator revolves around CRDs that match S3 resources :

buckets.s3.onyxia.sh
policies.s3.onyxia.sh
paths.s3.onyxia.sh
users.s3.onyxia.sh

The custom resources based on these CRDs are a somewhat simplified projection of the real S3 resources. From the operator's point of view :

A Bucket CR matches a S3 bucket, and only has a name, a quota (actually two, see Bucket example in Usage section below), and optionally, a set of paths
A Policy CR matches a "canned" policy (not a bucket policy, but a global one, that can be attached to a user), and has a name, and its actual content (IAM JSON)
A Path CR matches a set of paths inside of a policy. This is akin to the paths property of the Bucket CRD, except Path is not responsible for Bucket creation.
A S3User CR matches a user in the s3 server, and has a name, a set of policy and a set of group.

Each custom resource based on these CRDs on Kubernetes is to be matched with a resource on the S3 instance. If the CR and the corresponding S3 resource diverge, the operator will create or update the S3 resource to bring it back to.

Two important caveats :

It is one-way - if something happens on the S3 side directly (instead of going through the CRs), the operator has no way of reacting. At best, the next trigger will overwrite the S3 state with the declared state in the k8s custom resource.
Originally, the operator did not manage resource deletion. This has changed in release v0.8.0 (see #40), but it still isn't a focus, and the implementation is simple. For instance, bucket deletion will simply fail if bucket is not empty - no logic was added to opt-in a "forced" deletion of everything inside the bucket.

Installation

The S3 operator is provided either in source form through this repository, or already built as a Docker image available on Docker Hub.

Helm

With this Docker image, the recommended way of installing S3 Operator on your cluster is to use the Helm chart provided in the dedicated repository : https://github.com/InseeFrLab/helm-charts/tree/master/charts/s3-operator. Among other things, the chart takes care of managing a (Kubernetes) ServiceAccount for the operator to run as. The most basic way of using this chart would be :

helm repo add inseefrlab https://inseefrlab.github.io/helm-charts # or [helm repo update] if already available
helm install <name> s3-operator --values <yaml-file/url>  # see below for the parameters

Running from source

Alternatively, if you just wish to try out the operator without actually installing it, it is also possible to just clone this repository, and run the operator locally - outside of the Kubernetes cluster. This requires Go 1.19+, and prior installation of the CRDs located in config/crd/bases, typically with kubectl. After which, you can simply run :

git clone https://github.com/InseeFrLab/s3-operator.git # or use a tag/release
cd s3-operator
go run main.go --s3-endpoint-url *** --s3-access-key *** --s3-secret-key *** # see below for the parameters

To quote the Operator SDK README (also visible below), running the operator this way will automatically use the current context in your kubeconfig file (i.e. whatever cluster kubectl cluster-info shows). RBAC-wise, you need to be able to freely manipulate the custom resources associated to the operator (Bucket, Policy and Path) in every namespace - see also the generated ClusterRole manifest.

Kustomize

Finally, as this operator was generated through Operator SDK, it should be possible to use kustomize to bootstrap the operator as well. Though this method is untested by the maintainers of the project, the Operator SDK generated guidelines (see below) might help in making use of the various kustomize configuration files, possibly through the use of make.

Configuration

The operator exposes a few parameters, meant to be set as arguments, though it's possible to use environment variables for some of them. When an environment variable is available, it takes precedence over the flag.

The parameters are summarized in the table below :

Flag name	Default	Environment variable	Multiple values allowed	Description
`health-probe-bind-address`	`:8081`	-	no	The address the probe endpoint binds to. Comes from Operator SDK.
`leader-elect`	`false`	-	no	Enable leader election for controller manager. Enabling this will ensure there is only one active controller manager. Comes from Operator SDK.
`metrics-bind-address`	`:8080`	-	no	The address the metric endpoint binds to. Comes from Operator SDK.
`region`	`us-east-1`	-	no	The region to configure for the S3 client.
`s3-access-key`	-	`S3_ACCESS_KEY`	no	The access key used to interact with the S3 server.
`s3-ca-certificate-base64`	-	-	yes	(Optional) Base64 encoded, PEM format CA certificate, for https requests to the S3 server.
`s3-ca-certificate-bundle-path`	-	-	no	(Optional) Path to a CA certificates bundle file, for https requests to the S3 server.
`s3-endpoint-url`	`localhost:9000`	-	no	Hostname (or hostname:port) of the S3 server.
`s3-provider`	`minio`	-	no	S3 provider (possible values : `minio`, `mockedS3Provider`)
`s3-secret-key`	-	`S3_SECRET_KEY`	no	The secret key used to interact with the S3 server.
`useSsl`	true	-	no	Use of SSL/TLS to connect to the S3 server
`bucket-deletion`	false	-	no	Trigger bucket deletion on the S3 backend upon CR deletion. Will fail if bucket is not empty.
`policy-deletion`	false	-	no	Trigger policy deletion on the S3 backend upon CR deletion
`path-deletion`	false	-	no	Trigger path deletion on the S3 backend upon CR deletion. Limited to deleting the `.keep` files used by the operator.
`s3User-deletion`	false	-	no	Trigger S3User deletion on the S3 backend upon CR deletion.
`override-existing-secret`	false	-	no	Update secret linked to s3User if already exist, else noop

Minimal rights needed to work

The Operator need at least this rights:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "s3:CreateBucket",
                "s3:GetObject",
                "s3:ListAllMyBuckets",
                "s3:ListBucket",
                "s3:PutObject"
            ],
            "Resource": [
                "arn:aws:s3:::*"
            ]
        },
        {
            "Effect": "Allow",
            "Action": [
                "admin:CreatePolicy",
                "admin:GetBucketQuota",
                "admin:GetPolicy",
                "admin:ListPolicy",
                "admin:SetBucketQuota",
                "admin:CreateUser",
                "admin:ListUsers",
                "admin:DeleteUser",
                "admin:GetUser",
                "admin:AddUserToGroup",
                "admin:RemoveUserFromGroup",
                "admin:AttachUserOrGroupPolicy",
                "admin:ListUserPolicies"

            ],
            "Resource": [
                "arn:aws:s3:::*"
            ]
        }
    ]
}

Usage

The first step is to install the CRDs in your Kubernetes cluster. The Helm chart will do just that, but it is also possible to do it manually - the manifests are in the config/crd/bases folder.
With the CRDs available and the operator running, all that's left is to create some custom resources - you'll find some commented examples in the subsections below.
As soon as a custom resource is created, the operator will react, and create or update a S3 resource accordingly.
The same will happen if you modify a CR - the operator will adjust the S3 bucket or policy accordingly - with the notable exception that it will not delete paths for buckets.
Upon deleting a CR, the corresponding bucket or policy will be left as is, as mentioned in the Description section above

Bucket example

apiVersion: s3.onyxia.sh/v1alpha1
kind: Bucket
metadata:
  labels:
    app.kubernetes.io/name: bucket
    app.kubernetes.io/instance: bucket-sample
    app.kubernetes.io/part-of: s3-operator
    app.kubernetes.io/managed-by: kustomize
    app.kubernetes.io/created-by: s3-operator
  name: bucket-sample
spec:
  # Bucket name (on S3 server, as opposed to the name of the CR)
  name: dummy-bucket

  # Paths to create on the bucket 
  # As it is not possible to create empty paths on a S3 server, (limitation of either S3,
  # or at least Minio, the only currently implemented provider), this will actually create
  # a .keep file at the deepest folder in the path.
  paths:
    - a_path
    - another/deeper/path
  
  # Quota to set on the bucket, in bytes (so 1000000000 would be 1GB).
  # This is split over two different parameters, although there is only one actual quota
  #   - "default" is required, and is used as the baseline
  #   - "override" is optional, and as the name implies, takes precedence over "default"
  # Though clumsy, this pattern (for lack of a better word) allows to easily change the
  # default quota for every buckets without impacting the ones that might have received
  # a manual change. If this is not useful to you, you can safely skip using "override".
  quota:
    default: 10000000    
    # override: 20000000

Policy example

apiVersion: s3.onyxia.sh/v1alpha1
kind: Policy
metadata:
  labels:
    app.kubernetes.io/name: policy
    app.kubernetes.io/instance: policy-sample
    app.kubernetes.io/part-of: s3-operator
    app.kubernetes.io/managed-by: kustomize
    app.kubernetes.io/created-by: s3-operator
  name: policy-sample
spec:
  # Policy name (on S3 server, as opposed to the name of the CR)
  name: dummy-policy

  # Content of the policy, as a multiline string 
  # This should be IAM compliant JSON - follow the guidelines of the actual
  # S3 provider you're using, as sometimes only a subset is available.
  policyContent: >-
    {
      "Version": "2012-10-17",
      "Statement": [
      {
        "Effect": "Allow",
        "Action": [
          "s3:*"
        ],
        "Resource": [
          "arn:aws:s3:::dummy-bucket",
          "arn:aws:s3:::dummy-bucket/*"
        ]
      }
      ]
    }

Path example

apiVersion: s3.onyxia.sh/v1alpha1
kind: Path
metadata:
  labels:
    app.kubernetes.io/name: path
    app.kubernetes.io/instance: path-sample
    app.kubernetes.io/part-of: s3-operator
    app.kubernetes.io/managed-by: kustomize
    app.kubernetes.io/created-by: s3-operator
  name: path-sample
spec:
  # Bucket name (on S3 server, not a Bucket CR's metadata.name)
  bucketName: shared-bucket

  # Paths to create on the bucket 
  paths:
    - /home/alice
    - /home/bob

S3User example

apiVersion: s3.onyxia.sh/v1alpha1
kind: S3User
metadata:
  labels:
    app.kubernetes.io/name: user
    app.kubernetes.io/instance: user-sample
    app.kubernetes.io/part-of: s3-operator
    app.kubernetes.io/managed-by: kustomize
    app.kubernetes.io/created-by: s3-operator
  name: user-sample
spec:
  accessKey: user-sample
  policies:
    - policy-example1
    - policy-example2

Each S3user is linked to a kubernetes secret which have the same name that the S3User. The secret contains 2 keys: accessKey and secretKey.

Operator SDK generated guidelines

Click to fold / unfold

Getting Started

You’ll need a Kubernetes cluster to run against. You can use KIND to get a local cluster for testing, or run against a remote cluster. Note: Your controller will automatically use the current context in your kubeconfig file (i.e. whatever cluster kubectl cluster-info shows).

Running on the cluster

Install Instances of Custom Resources:

kubectl apply -f config/samples/

Build and push your image to the location specified by IMG:

make docker-build docker-push IMG=<some-registry>/s3-operator:tag

Deploy the controller to the cluster with the image specified by IMG:

make deploy IMG=<some-registry>/s3-operator:tag

Uninstall CRDs

To delete the CRDs from the cluster:

make uninstall

Undeploy controller

UnDeploy the controller to the cluster:

make undeploy

Contributing

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

Test It Out

Install the CRDs into the cluster:

make install

Run your controller (this will run in the foreground, so switch to a new terminal if you want to leave it running):

make run

NOTE: You can also run this in one step by running: make install run

Modifying the API definitions

If you are editing the API definitions, generate the manifests such as CRs or CRDs using:

make manifests

NOTE: Run make --help for more information on all potential make targets

More information can be found via the Kubebuilder Documentation

InseeFrLab / s3-operator