/eunomia

A GitOps Operator for Kubernetes

Primary LanguageGoApache License 2.0Apache-2.0

eunomia-logo-small.png

Eunomia - a GitOps Operator for Kubernetes

Join the chat at https://gitter.im/KohlsTechnology/eunomia Build Status Docker Repository on Quay Go Report Card codecov

Project Deprication and Archival

As of March 9, 2022 this project is no longer maintained. This repository is being archived(marked as read only).

Who is Eunomia

According to Wikipedia:

Eunomia (Greek: Εὐνομία) was a minor Greek goddess of law and legislation (her name can be translated as "good order", "governance according to good laws"), as well as the spring-time goddess of green pastures (eû means "well, good" in Greek, and νόμος, nómos, means "law", while pasturelands are called nomia).

How to Pronounce it

Eunomia is pronounced: yuˈnoʊ mi ə. The English phonetic respelling is yoo-noh-mee-uh.

You can listen to it here and here.

What is GitOps

GitOps is all about turning day 2 operations into code! Not just that, it means you start thinking about day 2 on day 1. This is a dream come true for any Operations team! GitOps leverages the strength of automation and combines it with the power of git based workflows. It is a natural evolution beyond infrastructure-as-code and builds on top of DevOps best practices.

Next Generation Change Management

Especially in large Enterprises, Change Management is usually a painful experience. GitOps allows to take a lot of that pain out and streamline the process itself. It does so by still providing what the process tries to accomplish (and thus still meet audit requirements), but does so in a way that is much faster, much more secure, and much more reliable.

Your changes now all of a sudden provide:

  • Version Control
  • Peer Reviews
  • Approvals
  • Audit Trail
  • Reproducibility
  • Consistency
  • Reliability

What's your backout plan for your change? How about simply moving back to the previous commit and getting EXACTLY what you had before?

Purpose

The Eunomia provides the ability to implement these git-based flows for any resources in Kubernetes. Eunomia does not care if you have a plain Kubernetes, a cloud based Kubernetes (like GKE), or a complete PaaS platform based on Kubernetes (like OpenShift). Eunomia also does not care how you want to structure your data, how many repos you want to use, or which templating engine is your favourite.

Eunomia can handle straight-up (static) yaml files with the complete definition or create dynamic ones based on your templating engine of choice. Eunomia already supports Helm Charts, OpenShift Templates, and Jinja2 Templates, but can easily be extended to support others.

These templates will be merged and processed with a set of environment-specific parameters to get a list of resource manifests. Then these manifest can be created/updated/deleted in Kubernetes.

Getting started

If you want to deploy Eunomia from scratch to a local Minikube cluster, begin with the Getting started document. It will guide your through the setup step-by-step.

Vision

While this controller can certainly be used to directly populate an individual namespace with a configuration stored in git, the vision is that a hierarchy of controllers will be used to populate multiple namespaces. Ideally this approach will be used to bring a newly created cluster to a desired configured state. Only the initial seeding CR (Custom Resource) should have cluster-level permissions. Any sub-CRs should have progressively less access assigned to their service accounts.

Here is a very simple example of how the hierarchy might look like:

Gitops-hierarchy.png

The main sections of the GitOpsConfig CRD (Custom Resource Definition) are described below.

Example

The configuration is described in the GitOpsConfig CRD, here is an example:

apiVersion: eunomia.kohls.io/v1alpha1
kind: GitOpsConfig
metadata:
  name: simple-test
spec:
  # Add fields here
  templateProcessorArgs: "-e cluster_name=my_cluster_name"
  templateSource:
    uri: https://github.com/KohlsTechnology/eunomia
    ref: master
    contextDir: simple/templates
    secretRef: template-gitconfig
  parameterSource:
    contextDir: simple/parameters
    secretRef: parameter-gitconfig
  triggers:
  - type: Change
  - type: Webhook
  - type: Periodic
    cron: "0 * * * *"
  serviceAccountRef:      "mysvcaccount"
  templateProcessorImage: "quay.io/kohlstechnology/eunomia-base:latest"
  resourceDeletionMode:   "Delete"
  resourceHandlingMode:   "Apply"

TemplateSource and ParameterSource and TemplateProcessorArgs

The TemplateSource and ParameterSource specify where the templates and the parameters are stored. The exact contents of these locations depend on the templating engine that has been selected.

The TemplateProcessorArgs can be used to pass arguments/flags to the template processor. They can be accessed by the template processor in an environment variable named TEMPLATE_PROCESSOR_ARGS.

The fields of this section are:

  templateProcessorArgs: "-e cluster_name=my_cluster_name"
  templateSource:
    uri: https://github.com/KohlsTechnology/eunomia
    ref: master
    contextDir: simple/templates
    httpProxy: <http proxy>
    httpsProxy: <https proxy>
    noProxy: <no Proxy>
    SecretRef: <gitconfig and credentials secret>

  parameterSource:
    uri: https://github.com/KohlsTechnology/eunomia
    ref: master
    contextDir: seed/parameters
    httpProxy: <http proxy>
    httpsProxy: <https proxy>
    noProxy: <no Proxy>
    SecretRef: <gitconfig and credentials secret>

These are the mandatory constraints and default behaviors of the fields:

field name mandatory default
uri yes N/A
ref no master
contextDir no .
httpProxy no
httpsProxy no
noProxy no
SecretRef no

If a secret is provided, then it is assumed that the connection to Git requires authentication. See the [Git Authentication](#git\ authentication) section below for more details.

If the uri is not specified in the parameterSource section, then it will default to the uri specified under templateSource.

Git Submodules

Some helm charts might require the configuration to be part of the chart itself (you can't read files from outside the chart). Loading files into a configmap is one example of this.

Separating the charts (templateSource) from the actual configuration (parameterSource) is a best practice. This allows you to separate your code (templates) from your configuration, which helps tremendously with change management.

One way to go about this is to use the config repo as a submodule and point to the master branch. During development you can of course point against another branch, just make sure you correct it in .gitmodules before the PR gets merged.

Add submodule to track master branch

git submodule add -b master <repo-url>

Checking out a repo with submodules

git clone <repo-url>
cd <repo>
git submodule init
git submodule update --recursive --remote

parameterSource processing

Eunomia uses the yq command to merge all yaml files in the specified folder. You have to be careful, if you have the same variable name in multiple files. Dictionaries will merge, lists will get overwritten.

Variable Hierarchy

You can provide a file hierarchy.lst, to allow a variable hierarchy. This will allow you to specify a default value and overwrite it on an environment level if necessary. This will greatly simplify your configuration and allows for deduplication of data, making your operational life a lot easier.

The contents of the file are simply relative path names, with the base being contextDir.

Example hierarchy.lst:

../defaults    #this is first ... lowest priority
../marketing   #this is second
../development #this is third ... highest priority

In this case it will load all yaml files from ../defaults, then merge it with everything in ../marketing, and lastly merges it with everything in ../development. You can also use the relative path ./, which means it'll also load the variables defined in contextDir directly (same folder that as hierarchy.lst). You can insert ./ in whatever order you want in the hierarchy.lst - it will determine its priority.

Upcoming features

Once issue #4 is resolved, you will be able to specify variable names to dynamically determine the correct folder. This will allow you to only have one hierarchy.lst. (Technically, it is actually already possible to use environment variables, but without #4, there are just none set that would be of any practical use in hierarchy.lst.)

Git Authentication

Specifying a SecretRef will automatically turn on git authentication. The secrets for the template and parameter repos will be mounted respectively in the /template-gitconfig and /parameter-gitconfig of the job pod. The referenced secrets must be available and how they are provisioned is beyond the scope of this operator. See the Vision paragraph on how to build a hierarchical structure, where the resources needed to run a given GitOpsConfig are configured by a predecessor GitpOpsConfig instance.

This secret will be linked from ~/ of the used running the pod. The secret must contain a .gitconfig file and may contain other files. The passed .gitconfig will be used during the git operations. It is advised to reference any additional files via the absolute path.

Username and password authentication

For username and password based authentication create the following .gitconfig:

[http]
    sslCAInfo = /template-gitconfig/ca.crt

[user]
    name = gitconfig

[credential]
    helper = store

When the credential helper is of type store, credentials are by default retrieved from the ~/.git-credentials file. This file should also be added to the secret and has the following format:

https://<username>:<password>@<git server fqdn>

Don't forget to provide the ca.crt file to the secret.

Certificate based authentication

For certificate based authentication, create the following .gitconfig:

[core]
    sshCommand = "ssh -i /template-gitconfig/mykey.rsa -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no"

and add the mykey.rsa file to the secret.

Job templates

For Eunomia to work properly there is a need for a specific Job or a CronJob.

A Job default template and a CronJob default template are built into the Dockerfile.

If you want to provide your own job templates, set the env variables JOB_TEMPLATE and CRONJOB_TEMPLATE. Their values should be set to paths, where appropriate yaml files can be found. The files themselves have to be accessible in the pod. To achieve this, you can for instance add ConfigMap data to a Volume.

Triggers

You can enable one or multiple triggers.

Name Description
Change This triggers every time the CR is changed, including when it is created.
Periodic Periodically apply the configuration. This can be used to either schedule changes for a specific time, use it for drift management to revert any changes, or as a safeguard in case webhooks were missed. It uses a cron-style expression.
Webhook This triggers when something on git changes. You have to configure the webhook yourself. For branches use just branch name in GitOpsConfig CR ref, but if you want webhook working for git tag, use refs/tags/[tag_name].

GitHub webhook configuration

To set up GitHub webhook follow this GitHub documentation. Create route on port 8080 to eunomia-service and use this route as GitHub webhook Payload URL with added webhook/ endpoint at the end.

Content type needs to be set to application/json.

Choose Just the push event to trigger webhook.

Template Engine

When it's time to apply a configuration, the GitOps controller runs a job pod. The image of the job pod can be specified in the templateProcessorImage field. This is the plugin mechanism to support multiple template engines. A base image is provided that can be inherited to simplify the process of adding support for a new templating engine. The base image provides the following workflow:

  1. gitClone.sh : This will clone the template and parameter repos. It is expected that there will be no need to customize this. Any required changes are most likely worthy of a pull-request upstream.

  2. discoverEnvironment.sh : This will create a set of environment variables that are specific to the target Kubernetes environment. Currently the following variables are supported:

    Name Description
    CA_BUNDLE Path to the platform-level CA bundle
    SERVICE_CA_BUNDLE Path to the service-level CA bundle
    NAMESPACE Current namespace

  3. hierarchy : processes all the parameter files and generates /tmp/eunomia_values_processed.yaml. It currently supports the following features:

    • Merging of all existing yaml files in the CLONED_PARAMETER_GIT_DIR location, into a single file for processing by the templating engine.
    • Substitution of variables with environment variables within the hierarchy.
    • Substitution of variables with environment variables within the final yaml file.

  4. processTemplate.sh : This file needs to be overwritten in order to support a different templating engine. The contract is the following:

    • Templates are available at the location specified by the variable: CLONED_TEMPLATE_GIT_DIR
    • Parameters are available at the location specified by the variable: CLONED_PARAMETER_GIT_DIR
    • After the template processing completes, the processed manifests should be stored at the location of this variable: MANIFEST_DIR

  5. resourceManager.sh : Processes the resources in MANIFEST_DIR. One or more files can be present, and all will be processed.

Currently the following templating engines are supported (follow the link to see examples of how new template processors can be added):

serviceAccountRef

This is the service account used by the job pod that will process the resources. The service account must be present in the same namespace as the one where the GitOpsConfig CR is and must have enough permission to manage the resources. It is out of scope of this controller how that service account is provisioned, although you can use a different GitOpsConfig CR to provision it (seeding CR).

In the Helm deployment, the ClusterRole eunomia-cluster-list provides list access to all resources, and will be provisioned if you set .Values.eunomia.operator.deployment.clusterViewer to true. This ClusterRole is intended to be used in a ClusterRoleBinding with "job runner" service accounts so they can find all of the resources that it owns. Without the ClusterRoleBinding, the jobs can still successfully run, however there will be error logs stating it can not find any cluster scoped resources.

Resource Handling Mode

This field specifies how resources should be handled, once the templates are processed. The following modes are currently supported:

  1. Apply, which is roughly equivalent to kubectl apply. Additionally, auto-detection of resources removed from git is performed, and they're deleted from the cluster. This is done by marking all the resources with a custom label, and removing resources for which the label was not touched by kubectl apply.
  2. Patch. Patch requires objects to already exists and will patch them. It's useful when customizing objects that are provided through other means.
  3. Create, equivalent to kubectl create. Template processors which take over the resource handling phase are not required to support this mode.
  4. Replace, equivalent to kubectl replace. Template processors which take over the resource handling phase are not required to support this mode.
  5. Delete, equivalent to kubectl delete. Template processors which take over the resource handling phase are not required to support this mode.
  6. None. In some cases there may be template processors or automation frameworks where the processing of templates and handling of generated resources are a single step. In that case, Eunomia can be configured to skip the built-in resource handling step.

Resource Deletion Mode

This field specifies how to handle resources when the GitOpsConfig object is deleted. Two options are available:

  1. Retain, resources previously created are left intact.
  2. Delete, resources are delete with the cascade option.
  3. None, resource deletion is not handled at all.

Installing Eunomia

Installing on Kubernetes Using Helm

See the helm README for details.

Installing on Kubernetes Using Operator Hub

See the getting started document for details.

Examples / Demos

We've created several examples for you to test out Eunomia. See EXAMPLES for details.

Monitoring

Monitoring with Prometheus

Prometheus is an open-source systems monitoring and alerting toolkit.

Prometheus collects metrics from monitored targets by scraping metrics HTTP endpoints.

  • configuring-prometheus

  • scrape_configs controls what resources Prometheus monitors.

  • kubernetes_sd_configs Kubernetes SD configurations allow retrieving scrape targets. Please see kubernetes_sd_configs for details.

  • Additionally, relabel_configs allow advanced modifications to any target and its labels before scraping.

By default, the metrics in Operator SDK are exposed on 0.0.0.0:8383/metrics

For more information, see Metrics in Operator SDK

Usage:

scrape_configs:
  - job_name: 'kubernetes-service-endpoints'
    kubernetes_sd_configs:
    - role: endpoints
    relabel_configs:
      - source_labels: [__meta_kubernetes_namespace]
        action: keep
        regex: test-eunomia-operator

You can find additional examples on their GitHub page.

Verify metrics port:

kubectl exec POD-NAME curl localhost:8383/metrics -n NAMESPACE

(e.g. kubectl exec eunomia-operator-5b9b664cfc-6rdrh curl localhost:8383/metrics -n test-eunomia-operator)

Kubernetes Events

Eunomia emits the following events in the namespace of the GitOpsConfig CR:

  • JobSuccessful - when a Job applying the CR finished successfully
  • JobFailed - when a Job applying the CR has finished with a failure (after all retries have failed)

Development

Please see our development documentation for details.

Troubleshooting

Please see our troubleshooting guide for details.

License

See LICENSE for details.

Code of Conduct

See CODE_OF_CONDUCT.md for details.