release-manager

README

Release Manager

GitOps release manager for kubernetes configuration repositories.

This project is used as an internal project at Lunar and it therefore contains some assumptions on our setup. This includes environment naming (dev, staging, prod). Further it is build around assumptions made by our OSS project shuttle, and id's for releases are a combination of branch name, git-sha from source repo, and git-sha from shuttle plan repo. Our initial intent is not to support this as an open source project.

We will however, have it public available for reference. This might change over time.

Design

The release-manager consist of 4 different "microservices" with each having a specific responsibility in the pipeline. The applications are basically utilities for moving files around a Git repository.

A simplified overview of all the components involved in the flow can be seen below.

To read more about the what each service is responsible for see Components. In addition to the services we utilize Jenkins as a CI server and flux as a release operator running inside each cluster.

Interactions

As seen on the illustration the main interaction point for developers are their service repository, doing code changes and with the hamctl CLI.

Below are descriptions of the common commands used in day to day activities.

Promote

The promotion flows, is a convetion based release process. It can be invoked by hamctl as follows:

hamctl promote --service example --env dev

The convention follows the following flow: master -> dev -> staging -> prod As seen in the example above, the example service will be promoted from the lastest available artifact from master to the dev environment.

Another example, is a promotion of an artifact running in, e.g. staging, to the production environment. This can be achieved with the following command:

hamctl promote --service example --env prod

The above locates what is running in the staging environment, and takes the necessary steps to run the same artifact in prod.

Release

The release flow, is a more liberal release process. There is no conventions in how artifacts move between environments. This makes it suitable for releasing hotfix-branches to production or feature-branches to a specific environment for testing before merging into master.

The release flow currently consist of two approaches, either the release of the lastest artifact from a given branch, or a specific artifact id.

Example of a release of a feature branch to the dev environment:

hamctl release --service example --branch "feature/new_feature" --env dev

Example of a release of a specific artifact id to the staging environment:

hamctl release --service example --artifact dev-0017d995e3-67e9d69164 --env staging

Status

Status is a convience flow to display currently released artifact to the three different environments; dev, staging,prod.

$ hamctl status --service example

dev:
  Tag: master-1c1508405e-67e9d69164
  Author: Kasper Nissen
  Committer: Peter Petersen
  Message: empty-commit-to-test-flow
  Date: 2019-04-01 11:14:26 +0200 CEST
  Link: https://jenkins.example.lunar.app/job/github/job/example-service/job/master/132/display/redirect
  Vulnerabilities: 0 high, 0 medium, 0 low

staging:
  Tag: master-1c1508405e-67e9d69164
  Author: Kasper Nissen
  Committer: Peter Petersen
  Message: empty-commit-to-test-flow
  Date: 2019-04-01 11:14:26 +0200 CEST
  Link: https://jenkins.example.lunar.app/job/github/job/example-service/job/master/132/display/redirect
  Vulnerabilities: 0 high, 0 medium, 0 low

prod:
  Tag: master-8fgh08405e-67e9d69164
  Author: John John
  Committer: Hans Hansen
  Message: some-commit
  Date: 2019-04-01 11:14:26 +0200 CEST
  Link: https://jenkins.example.lunar.app/job/github/job/example-service/job/master/132/display/redirect
  Vulnerabilities: 0 high, 0 medium, 0 low

Policies

It is possible to configure policies for releases with hamctl's policy command and globally with flags on the server.

You can list, apply and delete policies for a specific service like below.

hamctl policy --service <service> list
hamctl policy --service <service> apply <policy>
hamctl policy --service <service> delete <policy-id> [<policy-id>]

See below for details on how to apply specific policies.

Some policies cannot be applied simultaniously as they semantically does not support each other. An example is an auto-release policy releasing a branch not compatible with a branch-restriction policy. These cases are validated when applying either of them.

Auto-release artifacts from branches to environments

An auto-release policy instructs the release manager to deploy new artifacts from a specific branch into an environment.

Multiple policies can be applied for the same branch to different environments, e.g. release master artifacts to dev and staging.

This is an example of applying an auto-release policy for the product service for the master branch and dev environment.

hamctl policy --service example apply auto-release --branch master --env dev

Branch restriction on environments

A branch-restriction policy instructs the release manager to only allow artifacts from specific branches to be released to an environment. The --branch-regex flag defines a regular expression that is matched against the branch name on every release.

As an example, the following command applies a branch-restriction policy for the example service that only allows the master branch to be released to the prod environment.

hamctl policy --service example apply branch-restriction --env prod --branch-regex '^master$'

Another example is to allow only master or hostfix/* branches in prod like this.

hamctl policy --service example apply branch-restriction --env prod --branch-regex '^(master|hotfix\/.+)$'

It is not possible to create an auto-release policy for a non-matching branch to an environment that is protected by a branch-restriction policy.

Be aware that the regular expression should be as strict as possible otherwise you might get unexpected results. A branch regex like master will also allow branch names like refactor-master-worker, so make sure to mark the start ^ and end $ of the string.

The server can also enforce branch restrictions on all managed services by setting the policy-branch-restrictions flag. It takes a comma seprated list of <environment>=<branchRegex> values.

server start --policy-branch-restrictions 'production=^master$,staging=^staging$'

They will be visible with hamctl policy list but cannot by removed with hamctl. It is also not possible to overwrite them with custom policies, e.g. changing branch of a globally restricted environment.

Releases and policies

Release files are structured as shown below. In the root are folders for each environment, e.g. dev, prod. These folders contain a releases directory with kubernetes resource definitions of each namespace and their running applications. If an artifact contains a Flux Kustomization (apiVersion: kustomize.toolkit.fluxcd.io/v1beta1 and kind: Kustomization) custom resource the release manager moves it into the clusters directory tree. This is tailored to support Flux2.

A policies directory holds all recorded release policies. These are stored as JSON files for each service.

.
├── policies
│   └── <service>.json
├── <environments>
├── dev
│   └── releases
│       ├── <namespaces>
│       └── dev
│           └── <service>
│               ├── artifact.json
│               ├── 01-configmap.yaml
│               ├── 02-db-configmap.yaml
│               ├── 40-deployment.yaml
│               └── 50-service.yaml
└── clusters
    ├── <environments>
    └── dev
        ├── <namespaces>
        └── dev
            └── <service>.yaml

When running kubectl apply files are applied to the cluster alphabetically so the following convention should be used by configuration generators.

00 CRDs
01-09 configmaps
10-19 secrets
20-29 volumes
30-39 rbac
40-49 deployments/daemonsets
50-59 service
60-69 ingress

Resources starting with 00_ will skip resource validation in the Lunar shuttle plans. CustomResourceDefintions requires custom schemas which are usually not available so they should always start with 00_.

Components

The release-manager consists of four applications.

Application	Description
artifact	a simple tool for generating an artifact.json blob with information from the CI pipeline
daemon	a daemon reporting events about cluster component status back to the release-manager server
hamctl	a CLI client for interacting with the release-manager server
server	the API-server where clients (hamtcl) connects to, and daemon reports events to. It further implements different flows, e.g., promote a release, release an artifact

Artifact

artifact is used to generate, what we refer to as artifacts. These are a json file containing relevant information from the CI flow. The id's of the artifacts, are composed of <branch-name>-<application-git-sha>-<shuttle-plan-git-sha>.

We use shuttle plans to centralize our CI pipeline definitions, which is why we include the version of the plan in the artifact id.

Here is a redacted example of a generated artifact specification.

{
  "id": "dev-0017d995e3-67e9d69164",
  "application": {
    "sha": "0017d995e32e3d1998395d971b969bcf682d2085",
    "message": "fix something",
    "name": "example-service",
    ...
    "url": "https://github.com/lunarway/example-service/commits/0017d995e32e3d1998395d971b969bcf682d2085",
    "provider": "GitHub"
  },
  "ci": {
    "jobUrl": "https://jenkins.example.lunar.app/job/github/job/example-service/job/dev/84/display/redirect",
    "start": "2019-03-29T13:47:15.259380775+01:00",
    "end": "2019-03-29T13:49:57.686299407+01:00"
  },
  "shuttle": {
    "plan": {
      "message": "Support-new-feature",
      "url": "git://git@github.com:lunarway/lw-shuttle-go-plan.git"
    }
  },
  "stages": [
    {
      "id": "build",
      "name": "Build",
      "data": {
        "dockerVersion": "18.09.3",
        "image": "quay.io/lunarway/example",
        "tag": "dev-0017d995e3-67e9d69164"
      }
    }
  ]
}

hamctl

hamctl is a CLI for interacting with the release-manager server. Examples of commands are hamctl release or hamctl status.

See Interactions for more examples.

It uses a token-based authentication model for interacting with the server. This can either be provided as a command-line argument --http-auth-token or using the environment variable HAMCTL_AUTH_TOKEN.

Completions

Shell completions are available with the command completion. The following commands will add completions to the current shell in either bash or zsh.

source <(hamctl completion bash)
source <(hamctl completion zsh)

For a more detailed installation instruction see the help output.

hamctl completion --help

daemon

The daemon is an agent running in each of the kubernetes clusters and reports state changes in the environment back to the release-manager. daemon needs access to the kubernetes API server, and can be configured using a ServiceAccount.

daemon uses a token-based authentication model for interacting with the release-manager. This token can be set using the command-line argument --auth-token or the ENV variable: DAEMON_AUTH_TOKEN

Server

The server is responsible for all the operations related to releasing new versions. hamctl and artifact communicates with it over HTTP to initiate releases, register new artifacts etc.

In its simplest form it is responsible for moving files around a Git repository based on the commands it receives, eg. release artifact.

Notifications

When releasing applications the server will notify different upstream services along with outputting an identifiable log useful for log aggregation statistics.

info  command/start.go:145  Release [dev]: verification (master-e8da185c2c-06249f1a78) by Bjørn Sørensen, author Bjørn Sørensen

A Slack message is pushed to a #releases-<env> Slack channel.

Grafana is annotated with release metadata and tag deployment useful for plotting on graphs to see when new releases are rolled out.

If the artifact provider is GitHub and a GitHub API token is provided (--github-api-token) the application source repository is tagged with <env> on the released Git SHA.

Tracing support

The server collects Jaeger spans. This is enabled by default and reported as service release-manager. The jaeger configuration can be customized with available environment variables.

For local development a jaeger all-in-one instance can be created with Docker running make jaeger. The Jaeger UI will be available on localhost:16686.

To disable collection set JAEGER_DISABLED=true.

Storage

Multiple entities are stored in different storage layers to allow the release manager to work.

Artifacts

Artifacts are stored in an AWS S3 bucket. They are stored as zip files with keys from the service name and artifact id.

.
├── <service>
│  └── <artifact-id>
└── example
    └── master-sha1234-plan1234

Policies

Policies are stored in the Git repository along with all releases. Each service policy is a JSON file in the policies/<service>.json path.

{
  "service": "example",
  "autoReleases": [
    {
      "id": "auto-release-master-dev",
      "branch": "master",
      "environment": "dev"
    }
  ]
}

Installation

All the applications are cross compiled to Linux and MacOS and available in the Releases page. The server and daemon are also available as Docker images at quay.io/lunarway/release-manager and quay.io/lunarway/release-daemon.

Usually you will release the server and daemon with kubernetes Deployment resources and distribute the hamctl CLI to developers. artifact should be distributed to the Jenkins CI server and used in the pipelines.

Access to the config repository

The release manager server needs read/write permissions to the config repo.

To create a secret that the release manager can consume: (expects that the filename is identity)

kubectl create secret generic release-manager-git-deploy --from-file=identity=key

This secret should be mounted to /etc/release-manager/ssh

Development

The Makefile exposes targets for building, testing and deploying the release manager and its CLIs. See it for details.

The most common operations are build and tests.

$ make build
go build -o dist/hamctl ./cmd/hamctl
go build -o dist/server ./cmd/server
go build -o dist/artifact ./cmd/artifact

$ make build_server
go build -o dist/server ./cmd/server

$ make test
go test -v ./...

e2e setup

To help development it is possible to use the e2e setup.

This setup is based a kubernetes cluster managed by kind. The following resources is setup up

Name	Description
source-git-repo	A local git repository in e2e-test/source-git-repo that is used as the config repository.
fluxd	The fluxd service inside the k8s cluster, which is connected to the source-git-repo. It is polling the repo for changes every 5s, so it triggers as soon as a commit is done in source-git-repo, like a webhook from github normally would. Additionally fluxd is setup to --connect to release-daemon
release-daemon	A locally built binary of the release-daemon, but running inside the k8s cluster. The binary is mounted from local e2e-test/binaries for quick rebuild, so the pod can just be restarted while developing. This is done using the rebuild or watch actions.
release-server	A locally built binary of the release-daemon, that is running in the same manner as the release-daemon
rabbitmq	A simply setup rabbitmq server for the release-manager

To use the e2e setup there are the following actions supported:

Action	Command	Description
Start e2e setup	make e2e-setup	Start and initiate kind and e2e setup
Rebuild manager	make e2e-rebuild-local-manager	Rebuild the manager and restart pod in e2e cluster
Rebuild daemon	make e2e-rebuild-local-daemon	Like "Rebuild manager" but for the daemon
Watch manager	make e2e-rebuild-local-manager	Watch source code changes and rebuild the manager and restart pod in e2e cluster. Requires nodemon
Watch daemon	make e2e-rebuild-local-daemon	Like "Watch manager" but for the daemon
Do dummy release	make e2e-do-release	Do a release in git repo to trigger fluxd change
Do another dummy release	make e2e-do-another-release	Do another kind of release in git repo to trigger fluxd change
Stop e2e setup	make e2e-teardown	Stop and cleanup the e2e setup

Releasing

This project is configured with goreleaser and releases all 4 applications at once. Push a new tag to the main branch and Travis will publish a new release and create the changelog.