Deploying to a Kubernetes Cluster

Kubernetes has become the most popular way to deploy, run and manage containers in production. Google Cloud Platform, Microsoft Azure and Amazon Web Services and many more companies provide managed Kubernetes environment.

The API Platform symfony variant contains a built-in Helm (the k8s package manager) chart to deploy in a wink on any of these platforms.

This guide is based on Helm 3.

If you want to deploy API Platform on a local Kubernetes cluster, check out our Minikube tutorial!

1. Create a Kubernetes cluster on your preferred Cloud provider or install Kubernetes locally on your server, for example with kubeadm
2. Install Helm 3 locally following their documentation
3. Be sure to be connected to the right Kubernetes cluster kubectl config view Details e.g. for Google Cloud running: gcloud config get-value core/project

Working-Dir: Your local installation of api-platform. Default /api-platform/

Change the name “test-api-platform” to your Google project ID (not the project name). Quickstart Google Cloud If you do not have gcloud yet, install it with these command.

curl https://sdk.cloud.google.com | bash

Versioning: The 0.1.0 is the version. This value should be the same as the attribute appVersion in Chart.yaml. Infos for Google Container pulling and pushing

docker build -t gcr.io/test-api-platform/php:0.1.0 -t gcr.io/test-api-platform/php:latest api --target frankenphp_prod
docker build -t gcr.io/test-api-platform/pwa:0.1.0 -t gcr.io/test-api-platform/pwa:latest pwa --target prod

Optional: If your pwa project use Static Site Generation (SSG) and you need to build it against the API running locally, you can build the pwa with the command below.

docker build -t gcr.io/test-api-platform/pwa:0.1.0 -t gcr.io/test-api-platform/pwa:latest pwa --target prod --network=host --add-host php=127.0.0.1

gcloud auth configure-docker
docker push gcr.io/test-api-platform/php
docker push gcr.io/test-api-platform/pwa

Optional: push the version images:

docker push gcr.io/test-api-platform/php:0.1.0
docker push gcr.io/test-api-platform/pwa:0.1.0

The result should look similar to these images.

helm version

If you are using version 2.x follow this guide to migrate Helm to v3

helm dependency update ./helm/api-platform

This will create a folder helm/api-platform/charts/ and add all dependencies there. Actual this is bitnami/PostgreSQL, a file postgresql-[VERSION].tgz is created.

helm lint ./helm/api-platform

helm upgrade main ./helm/api-platform --namespace=default --create-namespace --wait \
    --install \
    --set "php.image.repository=gcr.io/test-api-platform/php" \
    --set php.image.tag=latest \
    --set "pwa.image.repository=gcr.io/test-api-platform/pwa" \
    --set pwa.image.tag=latest \
    --set php.appSecret='!ChangeMe!' \
    --set postgresql.postgresqlPassword='!ChangeMe!' \
    --set postgresql.persistence.enabled=true \
    --set "corsAllowOrigin=^https?:\/\/[a-z]*\.mywebsite.com$"

The " are necessary for Windows. Use ^ on Windows instead of \ to split commands into multiple lines. You can add the parameter --dry-run to check upfront if anything is correct. Replace the values with the image parameters from the stage above. The parameter php.appSecret is the AppSecret from ./.env Fill the rest of the values with the correct settings. For available options see /helm/api-platform/values.yaml. If you want a test deploy you can set corsAllowOrigin=’*'

After a successful installation, there is a message at the end. You can copy these commands and execute them to set a port-forwarding and get access on your local machine to the deploy. See image below.

If you prefer to use a managed DBMS like Heroku Postgres or Google Cloud SQL (recommended):

helm upgrade api-platform ./helm/api-platform \
    # ...
    --set postgresql.enabled=false \
    --set postgresql.url=pgsql://username:password@host/database?serverVersion=13

Finally, build the pwa (client and admin) JavaScript apps and deploy them on a static site hosting service.

You can access the php container of the pod with the following command. In this example the symfony console is called.

CADDY_PHP_POD=$(kubectl --namespace=default get pods -l app.kubernetes.io/name=api-platform -o jsonpath="{.items[0].metadata.name}")
kubectl --namespace=default exec -it $CADDY_PHP_POD -c api-platform-php -- bin/console

If the pods do not run, and you get the following error from google Kubernetes engine logs, there is probably a problem with the system architecture. standard_init_linux.go:211: exec user process caused "exec format error Build the images with the same system architecture as the cluster runs. Example: Building with Mac M1 with arm64 leads to problems. Most cluster will run with x86_64. Solution: https://blog.jaimyn.dev/how-to-build-multi-architecture-docker-images-on-an-m1-mac

There are 2 main upgrade strategies.

Change the version in the attribute “appVersion” in Chart.yaml and tag the images with this version. You can upgrade with the same command from the installation and pass all parameters.

Infos about best practices for tagging images for Kubernetes You have to use the *.image.pullPolicy=Always see the last 3 parameters.

PHP_POD=$(kubectl --namespace=bar get pods -l app=php -o jsonpath="{.items[0].metadata.name}")
kubectl --namespace=bar exec -it $PHP_POD -- bin/console doctrine:schema:create
helm upgrade api-platform ./helm/api-platform --namespace=default \
    --set "php.image.repository=gcr.io/test-api-platform/php" \
    --set php.image.tag=latest \
    --set "pwa.image.repository=gcr.io/test-api-platform/pwa" \
    --set pwa.image.tag=latest \
    --set php.appSecret='!ChangeMe!' \
    --set postgresql.postgresqlPassword='!ChangeMe!' \
    --set postgresql.persistence.enabled=true \
    --set "corsAllowOrigin=^https?://[a-z\]*\.mywebsite.com$" \
    --set php.image.pullPolicy=Always \
    --set pwa.image.pullPolicy=Always

You can find a complete deploy command for GKE on the demo project:

Running Pods with the Messenger Component to consume queues requires additions to the Helm chart.

Start by creating a new template for the queue-worker-deployment. The deployment.yaml can be used as template, the caddy container and all unused ENV variables should be removed.

Add the following lines under containers to overwrite the command.

command:
{{ range .Values.queue_worker.command }}
    - {{ . | quote }}
{{ end }}
args:
{{ range .Values.queue_worker.commandArgs }}
    - {{ . | quote }}
{{ end }}

Here is an example on how to use it from your values.yaml:

command: ['bin/console']
commandArgs: ['messenger:consume', 'async', '--memory-limit=100M']

The readinessProbe and the livenessProbe can not use the default docker-healthcheck but should test if the command is running.

First, make sure to install the /bin/ps binary, otherwise the readinessProbe and livenessProbe will fail:

# api/Dockerfile

RUN apt-get update && apt-get install --no-install-recommends -y \
+	procps \
# ...

Then, update the probes:

readinessProbe:
  exec:
    command:
      [
        '/bin/sh',
        '-c',
        "/usr/bin/pgrep -af '^php.*bin/console.*messenger:consume'",
      ]
  initialDelaySeconds: 120
  periodSeconds: 3
livenessProbe:
  exec:
    command:
      [
        '/bin/sh',
        '-c',
        "/usr/bin/pgrep -af '^php.*bin/console.*messenger:consume'",
      ]
  initialDelaySeconds: 120
  periodSeconds: 3