(ARCHIVED) Bazel Kubernetes Rules

This repository is no longer maintained.

Prow	Bazel CI

• k8s_defaults
• k8s_object
• k8s_objects

This repository contains rules for interacting with Kubernetes configurations / clusters.

Add the following to your WORKSPACE file to add the necessary external dependencies:

• Info for rules_docker

load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")

# https://github.com/bazelbuild/rules_docker/#setup
# http_archive("io_bazel_rules_docker", ...)

http_archive(
    name = "io_bazel_rules_k8s",
    strip_prefix = "rules_k8s-0.5",
    urls = ["https://github.com/bazelbuild/rules_k8s/archive/v0.5.tar.gz"],
    sha256 = "773aa45f2421a66c8aa651b8cecb8ea51db91799a405bd7b913d77052ac7261a",
)

load("@io_bazel_rules_k8s//k8s:k8s.bzl", "k8s_repositories")

k8s_repositories()

load("@io_bazel_rules_k8s//k8s:k8s_go_deps.bzl", k8s_go_deps = "deps")

k8s_go_deps()

As is somewhat standard for Bazel, the expectation is that the kubectl toolchain is preconfigured to authenticate with any clusters you might interact with.

For more information on how to configure kubectl authentication, see the Kubernetes documentation.

NOTE: we are currently experimenting with toolchain features in these rules so there will be changes upcoming to how this configuration is performed

For Google Container Engine (GKE), the gcloud CLI provides a simple command for setting up authentication:

gcloud container clusters get-credentials <CLUSTER NAME>

NOTE: we are currently experimenting with toolchain features in these rules so there will be changes upcoming to how this configuration is performed

New: Starting https://github.com/bazelbuild/rules_k8s/commit/ff2cbf09ae1f0a9c7ebdfc1fa337044158a7f57b

These rules can either use a pre-installed kubectl tool (default) or build the kubectl tool from sources.

The kubectl tool is used when executing the run action from bazel.

The kubectl tool is configured via a toolchain rule. Read more about the kubectl toolchain here.

If GKE is used, also the gcloud sdk needs to be installed.

NOTE: we are currently experimenting with toolchain features in these rules so there will be changes upcoming to how this configuration is performed

load("@io_bazel_rules_k8s//k8s:object.bzl", "k8s_object")

k8s_object(
  name = "dev",
  kind = "deployment",

  # A template of a Kubernetes Deployment object yaml.
  template = ":deployment.yaml",

  # An optional collection of docker_build images to publish
  # when this target is bazel run.  The digest of the published
  # image is substituted as a part of the resolution process.
  images = {
    "gcr.io/rules_k8s/server:dev": "//server:image"
  },
)

In your WORKSPACE you can set up aliases for a more readable short-hand:

load("@io_bazel_rules_k8s//k8s:k8s.bzl", "k8s_defaults")

k8s_defaults(
  # This becomes the name of the @repository and the rule
  # you will import in your BUILD files.
  name = "k8s_deploy",
  kind = "deployment",
  # This is the name of the cluster as it appears in:
  #   kubectl config view --minify -o=jsonpath='{.contexts[0].context.cluster}'
  cluster = "my-gke-cluster",
)

Then in place of the above, you can use the following in your BUILD file:

load("@k8s_deploy//:defaults.bzl", "k8s_deploy")

k8s_deploy(
  name = "dev",
  template = ":deployment.yaml",
  images = {
    "gcr.io/rules_k8s/server:dev": "//server:image"
  },
)

Note that in load("@k8s_deploy//:defaults.bzl", "k8s_deploy") both k8s_deploy's are references to the name parameter passed to k8s_defaults. If you change name = "k8s_deploy" to something else, you will need to change the load statement in both places.

It is common practice in the Kubernetes world to have multiple objects that comprise an application. There are two main ways that we support interacting with these kinds of objects.

The first is to simply use a template file that contains your N objects delimited with ---, and omitting kind="...".

The second is through the use of k8s_objects, which aggregates N k8s_object rules:

# Note the plurality of "objects" here.
load("@io_bazel_rules_k8s//k8s:objects.bzl", "k8s_objects")

k8s_objects(
   name = "deployments",
   objects = [
      ":foo-deployment",
      ":bar-deployment",
      ":baz-deployment",
   ]
)

k8s_objects(
   name = "services",
   objects = [
      ":foo-service",
      ":bar-service",
      ":baz-service",
   ]
)

# These rules can be nested
k8s_objects(
   name = "everything",
   objects = [
      ":deployments",
      ":services",
      ":configmaps",
      ":ingress",
   ]
)

This can be useful when you want to be able to stand up a full environment, which includes resources that are expensive to recreate (e.g. LoadBalancer), but still want to be able to quickly iterate on parts of your application.

A common practice to avoid clobbering other users is to do your development against an isolated environment. Two practices are fairly common-place.

1. Individual development clusters
2. Development "namespaces"

To support these scenarios, the rules support using "stamping" variables to customize these arguments to k8s_defaults or k8s_object.

For per-developer clusters, you might use:

k8s_defaults(
  name = "k8s_dev_deploy",
  kind = "deployment",
  cluster = "gke_dev-proj_us-central5-z_{BUILD_USER}",
)

For per-developer namespaces, you might use:

k8s_defaults(
  name = "k8s_dev_deploy",
  kind = "deployment",
  cluster = "shared-cluster",
  namespace = "{BUILD_USER}",
)

You can customize the stamp variables that are available at a repository level by leveraging --workspace_status_command. One pattern for this is to check in the following:

$ cat .bazelrc
build --workspace_status_command="bash ./print-workspace-status.sh"

$ cat print-workspace-status.sh
cat <<EOF
VAR1 value1
# This can be overriden by users if they "export VAR2_OVERRIDE"
VAR2 ${VAR2_OVERRIDE:-default-value2}
EOF

For more information on "stamping", you can see also the rules_docker documentation on stamping here.

Another ugly problem remains, which is that image references are still shared across developers, and while our resolution to digests avoids races, we may not want them trampling on the same tag, or on production tags if shared templates are being used.

Moreover, developers may not have access to push to the images referenced in a particular template, or the development cluster to which they are deploying may not be able to pull them (e.g. clusters in different GCP projects).

To resolve this, we enable developers to "chroot" the image references, publishing them instead to that reference under another repository.

Consider the following, where developers use GCP projects named company-{BUILD_USER}:

k8s_defaults(
  name = "k8s_dev_deploy",
  kind = "deployment",
  cluster = "gke_company-{BUILD_USER}_us-central5-z_da-cluster",
  image_chroot = "us.gcr.io/company-{BUILD_USER}/dev",
)

In this example, the k8s_dev_deploy rules will target the developer's cluster in their project, and images will all be published under the image_chroot.

For example, if the BUILD file contains:

k8s_deploy(
  name = "dev",
  template = ":deployment.yaml",
  images = {
    "gcr.io/rules_k8s/server:dev": "//server:image"
  },
)

Then the references to gcr.io/rules_k8s/server:dev will be replaced with one to: us.gcr.io/company-{BUILD_USER}/dev/gcr.io/rules_k8s/server@sha256:....

Sometimes, you need to replace additional runtime parameters in the YAML file. While you can use expand_template for parameters known to the build system, you'll need a custom resolver if the parameter is determined at deploy time. A common example is Google Cloud Endpoints service versions, which are determined by the server.

You can pass a custom resolver executable as the resolver argument of all rules:

sh_binary(
  name = "my_script",
  ...
)

k8s_deploy(
  name = "dev"
  template = ":deployment.yaml",
  images = {
    "gcr.io/rules_k8s/server:dev": "//server:image"
  },
  resolver = "//my_script",
)

This script may need to invoke the default resolver (//k8s/go/cmd/resolver) with all its arguments. It may capture the default resolver's output and apply additional modifications to the YAML, printing the final YAML to stdout.

The k8s_object[s] rules expose a collection of actions. We will follow the :dev target from the example above.

Build builds all of the constituent elements, and makes the template available as {name}.yaml. If template is a generated input, it will be built. Likewise, any docker_build images referenced from the images={} attribute will be built.

bazel build :dev

Deploying with tags, especially in production, is a bad practice because they are mutable. If a tag changes, it can lead to inconsistent versions of your app running after auto-scaling or auto-healing events. Thankfully in v2 of the Docker Registry, digests were introduced. Deploying by digest provides cryptographic guarantees of consistency across the replicas of a deployment.

You can "resolve" your resource template by running:

bazel run :dev

The resolved template will be printed to STDOUT.

This command will publish any images = {} present in your rule, substituting those exact digests into the yaml template, and for other images resolving the tags to digests by reaching out to the appropriate registry. Any images that cannot be found or accessed are left unresolved.

This process only supports fully-qualified tag names. This means you must always specify tag and registry domain names (no implicit :latest).

Users can create an environment by running:

bazel run :dev.create

This deploys the resolved template, which includes publishing images.

Users can update (replace) their environment by running:

bazel run :dev.replace

Like .create this deploys the resolved template, which includes republishing images. This action is intended to be the workhorse of fast-iteration development (rebuilding / republishing / redeploying).

Users can "apply" a configuration by running:

bazel run :dev.apply

:dev.apply maps to kubectl apply, which will create or replace an existing configuration. For more information see the kubectl documentation.

This applies the resolved template, which includes republishing images. This action is intended to be the workhorse of fast-iteration development (rebuilding / republishing / redeploying).

Users can tear down their environment by running:

bazel run :dev.delete

It is notable that despite deleting the deployment, this will NOT delete any services currently load balancing over the deployment; this is intentional as creating load balancers can be slow.

Users can "describe" their environment by running:

bazel run :dev.describe

Users can "diff" a configuration by running:

bazel run :dev.diff

:dev.diff maps to kubectl diff, which will diff the live against the would-be applied version. For more information see the kubectl documentation.

This diffs the resolved template, but does not include republishing images.

k8s_object(name, kind, template)

A rule for interacting with Kubernetes objects.

  # Example
  k8s_object(
    name = "my_ingress",
    kind = "ingress",
# A template of a Kubernetes ingress object yaml.
template = ":ingress.yaml",

# An optional collection of docker_build images to publish
# when this target is bazel run.  The digest of the published
# image is substituted as a part of the resolution process.
substitutions = {
  "%{expand_template_variable}": "$(make_expanded_variable}",
})

k8s_objects(name, objects)

A rule for interacting with multiple Kubernetes objects.

k8s_defaults(name, kind)

A repository rule that allows users to alias k8s_object with default values.

To test rules_k8s, you can run the provided e2e tests locally on Linux by following these instructions.

Users find on stackoverflow, slack and Google Group mailing list.

Stackoverflow is a great place for developers to help each other.

Search through existing questions to see if someone else has had the same issue as you.

If you have a new question, please [ask] the stackoverflow community. Include rules_k8s in the title and add [bazel] and [kubernetes] tags.

The general bazel support options links to the official bazel-discuss Google group mailing list.

Slack and IRC are great places for developers to chat with each other.

There is a #bazel channel in the kubernetes slack. Visit the kubernetes community page to find the slack.k8s.io invitation link.

There is also a #bazel channel on Freenode IRC, although we have found the slack channel more engaging.

Here's a (non-exhaustive) list of companies that use rules_k8s in production. Don't see yours? You can add it in a PR!

• Etsy
• Jetstack
• Prow
• Tink
• Cookies