Skip to content

Executes Terraform configuration as job/pod inside a Kubernetes cluster.

License

Notifications You must be signed in to change notification settings

RaphaelVogel/terraformer

 
 

Repository files navigation

Terraformer

CI Build status

Terraformer is a tool that can execute Terraform commands (apply, destroy and validate) and can be run as a Pod inside a Kubernetes cluster. The Terraform configuration and state files (main.tf, variables.tf, terraform.tfvars and terraform.tfstate) are stored as ConfigMaps and Secrets in the Kubernetes cluster and will be retrieved and updated by Terraformer.

For talking to the API server, Terraformer will either use the service account of its Pod or the provided kubeconfig (via the --kubeconfig flag). Names of the ConfigMaps and Secrets have to be provided via command line flags. The namespace of the objects can be specified via --namespace or will be defaulted to the Pod's namespace.

For more details and example Kubernetes manifests see the example directory.

⚠️ Running Terraformer as a Job

Usually, terraformer apply|destroy|validate runs within a single Pod. Please note, that running Terraformer as a Job is not recommended. The Job object will start a new Pod if the first Pod fails or is deleted (for example due to a Node hardware failure or a reboot). Thus, you may end up in a situation with two running Terraformer Pods at the same time which can fail with conflicts.

State file watcher + update worker

While Terraform itself is running, Terraformer watches the state file for changes and updates the state ConfigMap as soon as a change occurs. Internally, this is achieved by leveraging standard Kubernetes controller mechanisms: the file watcher inserts an update key into a queue for each file write event, and a separate worker goroutine reads from the queue and updates the state ConfigMap for every key.

After Terraform exits, Terraformer tries to update the state ConfigMap one last time, and retries the operation with the exponential backoff until it succeeds or times out.

Signal handling

Apart from dealing with Terraform configuration and state, Terraformer also handles Pod lifecycle event, i.e. shutdown signals. It will try to relay SIGINT and SIGTERM to the running Terraform process in order to stop the ongoing infrastructure operations on Pod deletion.

Termination log

If the Terraform command execution fails (e.g. because of invalid credentials or similar), Terraformer will copy Terraform's logs to $base_dir/terraform-termination-log (where $base_dir is the directory specified in the --base-dir command line flag (defaults to /)). Controllers deploying Terraformer Pods for infrastructure management can set spec.containers[].terminationMessagePath in the Pod specification to this file path to have the kubelet populate the .status.containerStatuses[].lastState.terminated.message field (see this doc for reference). It can then be used to detect error codes from the Terraform logs without fetching all Pods logs, that also include info messages from Terraformer.

The Pod status will then look similar to this:

status:
  containerStatuses:
  - name: terraformer
    ready: false
    restartCount: 0
    started: false
    state:
      terminated:
        exitCode: 1
        finishedAt: "2021-04-26T10:28:40Z"
        message: "\nError: error configuring Terraform AWS Provider: error validating
          provider credentials: error calling sts:GetCallerIdentity: SignatureDoesNotMatch:
          The request signature we calculated does not match the signature you provided.
          Check your AWS Secret Access Key and signing method. Consult the service
          documentation for details.\n\tstatus code: 403, request id: 49163947-9edf-4c44-ae7d-b013b119442a\n\n\n"
        reason: Error
        startedAt: "2021-04-26T10:28:32Z"

How to run it locally

The Makefile specifies targets for running and developing Terraformer locally:

make run

make run (with an optional variable COMMAND=[command]) will run the given terraformer command (or apply by default) locally on your machine using go run. You will have to point Terraformer to a running cluster, where the config and state is stored, via the KUBECONFIG environment variable. This can be any Kubernetes cluster, either a local one (setup using minikube, kind or the local nodeless garden) or even a remote one.

$ make run COMMAND=apply
# running `go run ./cmd/terraformer apply`
...

make start

make start is similar to make run, but instead of directly running terraformer on your machine it starts a docker development container with all the prerequisites installed (like go, terraform and the terraform aws provider plugin) and starts terraform via go run in it. This allows running terraformer in an environment that is very close to the environment that it is executed in, when running it in a Kubernetes Pod.

Just as with make run, you will have to set your KUBECONFIG to point to a Kubernetes cluster used for hosting config and state. If you are using kind, you might want to use the example cluster config to create a cluster with certificates, that will be trusted from inside the docker VM (on Mac or Windows).

When running this for the first time, the development container will have to be built, which might take a minute. After that, the image will be reused and only rebuilt if you change something in the files relevant for the image build.

make start-dev-container

The mentioned development container can also be started directly by executing running make start-dev-container. This will open a shell in this container, where you can execute all commands for development and testing like make, go, ginkgo and so on.

How to test it

Terraformer currently comes with three suites of tests: a unit test, a binary e2e and a Pod e2e test suite.

Unit Tests and Binary E2E tests

The unit tests and binary e2e tests can be executed by running make test. Most tests are executed against a local control plane by leveraging the envtest package. Therefore, the kube-apiserver and etcd binaries are fetched to bin/kubebuilder and will be started by the tests.

You can also run the tests against a different cluster for debugging or other purposes by setting the USE_EXISTING_CLUSTER and KUBECONFIG environment variable:

KUBECONFIG=$HOME/.kube/configs/local/garden.yaml USE_EXISTING_CLUSTER=true make test

Pod E2E tests

The Pod E2E tests can be executed via make test-e2e. This will run e2e tests against a terraformer Pod with the AWS plugin installed. It uses an existing cluster (given by the KUBECONFIG env var) and deploys a terraformer apply Pod, that will create some lightweight resource (ec2 keypair) on AWS. The test validates, that the resource was created on AWS using the AWS go-sdk and that the state ConfigMap has been updated accordingly. After that, the test deploys a terraformer destroy Pod and validates again the changes on AWS and the state ConfigMap.

In order to execute this test suite, you will need an existing Kubernetes cluster, where the terraformer Pod will be deployed (pointed to by the KUBECONFIG env var). This can be any Cluster (either local, e.g. via kind) or a remote one running in the Cloud. Additionally, you will need a set of AWS credentials with which the test resources will be created (stored under .kube-secrets/aws/{access_key_id,secret_access_key}.secret).

$ make test-e2e
# Executing pod e2e test with terraformer image eu.gcr.io/gardener-project/gardener/terraformer-aws:v2.1.0-dev-b705a1b4b9bfd47a106998892f48ced0dc8caa56
# If the image for this tag is not built/pushed yet, you have to do so first.
# Or you can use a specific image tag by setting the IMAGE_TAG variable
# like this: `make test-e2e IMAGE_TAG=v2.0.0`
=== RUN   TestTerraformer
Running Suite: Terraformer Pod E2E Suite
========================================
...

Docker Images

Terraformer images are built with every pipeline run and pushed to a public GCR repository. The list of existing images and tags can be found in europe-docker.pkg.dev/gardener-project/public/gardener.

Image variants

This repo features different container image variants, all and a few different provider-specific variants. Each image variant includes the terraformer binary itself, plus terraform and some terraform provider plugins.

The image variants are specified under build, each directory represents one variant. Each variant defines which version of terraform (TF_VERSION file) and which terraform provider plugins (with their respective versions, see (terraform-bundle.hcl file)) should be packaged into the image as well.

Historically, terraformer images included provider plugins for all Gardener provider extensions that were using terraformer for managing a Shoot cluster's infrastructure. This image is equivalent to the all variant. Packaging all provider plugins makes terraformer's images quite large and thus unnecessarily increases image pull time, network traffic and cost. With the different image variants, Gardener provider extensions can now deploy terraformer images with only the needed plugins inside. Also, the different extensions don't have to agree on a common terraform version, but are able to choose the terraform version which they want to use in their provider-specific image.

The all image variant is tagged as europe-docker.pkg.dev/gardener-project/public/gardener/terraformer, while the provider-specific image variants are tagged as europe-docker.pkg.dev/gardener-project/public/gardener/terraformer-{aws,gcp,...}.

Building images locally

You can use make docker-images to build all image variants locally.

Alternatively, make docker-image PROVIDER={all,aws,gcp,...} can be used to build only one specific image variant.

Find out more & Get in touch!

Terraformer was presented in the Gardener Community Call on Nov, 13th 2020.
Watch the recording to learn the story behind terraformer v2 and how it is used in our provider extensions!

Feedback and contributions are always welcome!

Reach out to us:

Please find further resources about our project here:

About

Executes Terraform configuration as job/pod inside a Kubernetes cluster.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Go 90.0%
  • Makefile 4.3%
  • Shell 2.9%
  • HCL 1.5%
  • Dockerfile 1.3%