Skip to content

Deployment installation Kubernetes

Jump to bottom
Bacluc edited this page Nov 2, 2024 · 11 revisions

This guide describes how eCamp v3 can be deployed on a Kubernetes cluster using helm.

Requirements

Some kubernetes cluster running kubernetes >= 1.19-0

Locally: an up-to-date version of kubectl, helm 3 and Git

Getting the cluster

For our instance https://dev.ecamp3.ch and the feature branch deployments, we created a kubernetes cluster on digitalocean.com, with the second-to-cheapest settings (two nodes instead of one). Once the cluster runs out of CPU or Memory, we simply increase it at digitalocean (adding more nodes for example).

Connecting to the cluster

Wherever you are hosting your kubernetes cluster, there should be some information on how to connect to it. On digitalocean, we can download the "config file" and place it in ~/.kube/config. Or if that file already exists and is in use, one can also save it in e.g. ~/.kube/ecamp3.yaml and execute export KUBECONFIG=~/.kube/ecamp3.yaml once in the terminal before using kubectl or helm.

You can test whether the connection to the cluster works using kubectl cluster-info or kubectl get nodes.

Probably, your hosting provider will give you access to a kubernetes dashboard. This is a graphical user interface for monitoring and managing the state of everything that is deployed on your cluster. The same information can be retrieved on the command line, using the kubectl tool. E.g. the pods in the default namespace can be listed using kubectl get pods. You can also use the helm tool to view information on all helm-managed releases, e.g. helm list shows all releases (in the default namespace).

SSL

We use Cloudflare for SSL termination, but you can set up cert-manager to use a let's encrypt certificate.

Ingress

In order to expose container ports on the internet, we need a so-called ingress controller in our cluster. The ingress controller will constantly monitor the containers (and their ingress definitions) in the cluster, and act as a reverse proxy to distribute network requests to the correct containers. For our purposes, an nginx ingress controller works fine. If you haven't set one up in the previous section follow the documentation. We used a one-click deployment from digitalocean to set this up for us.

Note: To upgrade the ingress-nginx installed in the cluster, either remove it and re-install it, or use the following command:

kubectl set image deployment/ingress-nginx-controller -n ingress-nginx controller=k8s.gcr.io/ingress-nginx/controller:v1.0.5@sha256:55a1fcda5b7657c372515fe402c3e39ad93aa59f6e4378e82acd99912fe6028d

cat << 'EOF' | tee -a nginx-values.yaml
controller:
  metrics:
    enabled: true
  podAnnotations:
    "prometheus.io/scrape": "true"
    "prometheus.io/port": "10254"
  config:
    log-format-escape-json: true
    log-format-upstream: {"timestamp":"$time_iso8601","requestID":"$req_id","proxyUpstreamName":"$proxy_upstream_name","proxyAlternativeUpstreamName":"$proxy_alternative_upstream_name","upstreamStatus":$upstream_status,"upstreamAddr":"$upstream_addr","httpRequest":{"requestMethod":"$request_method","requestUrl":"$host$request_uri","status":$status,"requestSize":"$request_length","responseSize":"$upstream_response_length","userAgent":"$http_user_agent","remoteIp":"$remote_addr","referer":"$http_referer","request_time_seconds":$upstream_response_time,"protocol":"$server_protocol"}}

EOF

helm upgrade --install ingress-nginx ingress-nginx \
             --namespace ingress-nginx \
             --create-namespace \
             --repo https://kubernetes.github.io/ingress-nginx \
             --values nginx-values.yaml
rm nginx-values.yaml

DNS

You can either set up the DNS records for your deployment manually, or you can install external-dns in the cluster, which has the ability to talk to many DNS providers via their APIs and automatically set up the necessary DNS records for the ingresses defined in the cluster. E.g. we use Cloudflare for DNS, so we installed external-dns as follows:

helm repo add bitnami https://charts.bitnami.com/bitnami
helm upgrade --install \
  -n external-dns --create-namespace \
  --set policy=sync \
  --set provider=cloudflare \
  --set-file cloudflare.apiToken=<file containing the API token for cloudflare> \
  --set txtOwnerId=<scope for dns entries created by helm chart> \
  ecamp3-external-dns bitnami/external-dns

If you want to use multiple clusters on the same Domain, you need to have a txtOwnerId per cluster. And its anyway good if bitnami/external-dns only deletes the entries it created.

Optional: Monitoring using Prometheus and Grafana

We use the 'Kubernetes Monitoring Stack' one-click app provided by DigitalOcean for monitoring the resource usage of the pods on the cluster. For installation instructions see Kubernetes Monitoring Stack.

To access the Grafana dashboards, you can forward the HTTP port to your localhost:

kubectl port-forward svc/kube-prometheus-stack-grafana 9000:80 -n kube-prometheus-stack

Then, access Grafana at localhost:9000 (yes, really at localhost!). Login is admin / prom-operator. Note: All our application pods are in the default namespace, the other namespaces are for system services.

Generating a JWT signing key pair

Run the following commands to generate a public/private key pair for signing JWT tokens. For optimal compatibility, these commands can also be run inside the php container: docker-compose run --entrypoint sh -v ${PWD}/:/app php

read -p 'Enter a passphrase that should be used for the key pair:' jwt_passphrase
echo -n "$jwt_passphrase" > jwt-passphrase.txt
echo "$jwt_passphrase" | openssl genpkey -out private.pem -pass stdin -aes256 -algorithm rsa -pkeyopt rsa_keygen_bits:4096
echo "$jwt_passphrase" | openssl pkey -in private.pem -passin stdin -out public.pem -pubout

The key pair is now stored in the files public.pem and private.pem, and the passphrase in jwt-passphrase.txt. You will need these three values in these files for the deployment, as described below.

Setting up the database

If you don't want the helm chart to automatically create your database, you need to provide one. With postgres v15 we came across some problems on DigitalOcean. To allow the specified user to migrate the schema, you need to grant the following permissions.

GRANT ALL PRIVILEGES ON DATABASE "$DATABASE" TO "$USER";
GRANT ALL PRIVILEGES ON SCHEMA public TO "$USER";

Connect with the user doadmin to your cluster to set the privileges. Make sure you are connected to $DATABASE when granting the privileges on SCHEMA public.

Connection Pooling

The small databases of digital ocean we use don't have enough connections to run 4 api instances at once. Thus we needed to use "Connection Pools" to manage the available connections for the many api instances (and their workers). You first need to create a connection pool in the digital ocean ui. We used a pool size of 11 and the pool mode of "Transaction". We also set the "User privileges override" setting. Then you need to use connection pools connection details instead of the connection details of the db directly.

Deploying ecamp3 to the cluster

ecamp3 comes with a helm chart, which can be thought of like a package that is installed using apt or npm or composer, but for installing software on a kubernetes cluster. The helm chart is based on the one coming with API platform, but was extended to include all the services that ecamp3 consists of.

First, you will have to get the chart to your computer. Since at this time we don't publish the chart to any helm repository, you will have to clone the GitHub repository which includes the chart, and go to the chart:

git clone https://github.com/ecamp/ecamp3.git
cd ecamp3/.helm/ecamp3

From there, perform the following steps to deploy ecamp3 on your cluster. The same command also works for upgrading an existing instance.

💡 If you are unsure, you can add the --dry-run and --debug arguments to test your deployment first

helm dependency update .
helm upgrade --install \
  --set imageTag=latest \
  --set sharedCookieDomain=.mydomain.com \
  --set api.domain=api-ecamp3.mydomain.com \
  --set frontend.domain=ecamp3.mydomain.com \
  --set print.domain=print-ecamp3.mydomain.com \
  --set mail.domain=mail-ecamp3.mydomain.com \
  --set postgresql.dropDBOnUninstall=false \
  --set php.dataMigrationsDir=dev-data \
  --set-file php.jwt.publicKey=public.pem \
  --set-file php.jwt.privateKey=private.pem \
  --set-file php.jwt.passphrase=jwt-passphrase.txt \
  --set deploymentTime=$(date -u +%s) \
  --set deployedVersion=$(git rev-parse --short HEAD) \
  --set recaptcha.siteKey=disabled \
  ecamp3 .

To re-deploy the same instance with just some adaptations to some --set values, you can also run e.g. helm upgrade --reuse-values --set deployedVersion=1.2.1 ecamp3 .

To deploy another copy of ecamp3, just change the deployment name (ecamp3 on the last line).

There are lots of other configuration values you can change. For a full list, refer to values.yaml.

Common errors and tasks

ImagePullBackoff

The image tag you selected with --set imageTag is not available on Docker Hub. To manually build and push the docker images:

docker build --target api_platform_php -t ecamp/ecamp3-api:$(git rev-parse HEAD) ../../api
docker push ecamp/ecamp3-api:$(git rev-parse HEAD)
docker build --target api_platform_caddy_prod -t ecamp/ecamp3-caddy:$(git rev-parse HEAD) ../../api
docker push ecamp/ecamp3-caddy:$(git rev-parse HEAD)
docker build -t ecamp/ecamp3-frontend:$(git rev-parse HEAD) --file ../../.docker-hub/frontend/Dockerfile ../..
docker push ecamp/ecamp3-frontend:$(git rev-parse HEAD)
docker build -t ecamp/ecamp3-print:$(git rev-parse HEAD) --file ../../.docker-hub/print/Dockerfile ../..
docker push ecamp/ecamp3-print:$(git rev-parse HEAD)

And then run the helm upgrade command again.

Restart a deployment

kubectl rollout restart deployment ecamp3-dev-frontend

Completely delete a helm release

helm delete ecamp3-pr1234

Locally reproduce an error which only happens on the deployment

See the dedicated documentation page on this topic.

Development

We have a GitHub Action that deploys our environments for us. But if we ever need to manually deploy, this is (more or less) the command we use:

helm upgrade --install \
  --set imageTag=$(git rev-parse HEAD) \
  --set sharedCookieDomain=.ecamp3.ch \
  --set api.domain=api-custom.ecamp3.ch \
  --set frontend.domain=custom.ecamp3.ch \
  --set print.domain=print-custom.ecamp3.ch \
  --set mail.domain=mail-custom.ecamp3.ch \
  --set postgresql.enabled=false \
  --set-file postgresql.url=postgres-url \
  --set-file postgresql.adminUrl=postgres-admin-url \
  --set postgresql.dropDBOnUninstall=false \
  --set php.dataMigrationsDir=dev-data \
  --set-file php.jwt.passphrase=jwt-passphrase \
  --set-file php.jwt.publicKey=public.pem \
  --set-file php.jwt.privateKey=private.pem \
  --set-file frontend.sentryDsn=frontend-sentry-dsn \
  --set-file print.sentryDsn=print-sentry-dsn \
  --set deploymentTime=$(date -u +%s) \
  --set deployedVersion=$(git rev-parse --short HEAD) \
  --set recaptcha.siteKey=disabled \
  ecamp3-mycustomrelease .

(Secret values aren't included here, but instead are placed in files like jwt-passphrase etc. and read using --set-file for this example.)

Clone this wiki locally