Kubernetes development for ERDDAP (as of 2.18 and the 2022 IOOS Code Sprint)
Assuming you’ve performed the Docker installation of ERDDAP, start by installing Minikube.
We will be generating what can be called ‘manifest’ files for a Kubernetes pod that will contain the ERDDAP instance. The manifest will need to reference the docker-erddap image from a registry. In this case, the registry will reside locally. Follow this guide precisely. Be sure to grab the latest version of the kube-registry.yaml file, and edit as necessary to be compliant with the latest kubernetes APIs.
If you are using Linux, there are a couple of commands that will need to be executed to avoid having to run Docker as root:
sudo usermod -aG docker $USER && newgrp docker
sudo chmod 0777 /var/run/docker.sock
With Minikube installed:
minikube start
Once that is booted:
kubectl create -f ./kube-registry.yaml` # Had to change `DaemonSet` `apiVersion` to `apps/v1` as of 2022/04/28
kubectl get pods -A
kubectl port-forward --namespace kube-system kube-registry-xx-xxxxx 5001:5000 # Check pod ID. Check ports too
sudo ss -lptn 'sport = :5000' # Way to check ports, correct the above command as necessary, I went with port 5001
Make a new terminal. Associate Docker with Minikube and the registry, pull in the Docker ERDDAP image
eval $(minikube -p minikube docker-env)
docker pull axiom/docker-erddap
docker tag axiom/docker-erddap:latest localhost:5001/axiom/docker-erddap:latest
As we’ve been using Docker Compose, there’s an app called Kompose that should be installed. This will enable the generation of Kubernetes manifests to establish an ERDDAP Kubernetes pod. For the mounts to local files, we’ll want a type of hostPath:
kompose convert --volumes hostPath
If things went alright, two files should have been generated, and should be referenced in a file named kustomization.yaml:
resources:
- erddap-service.yaml
- erddap-deployment.yaml
Working from the project directory, you can change the main directory of minikube via, for example:
minikube mount /home/jpthesmithe/Desktop/GLOS_System_Map/erddap-code-sprint/erddap-gold-standard:/erddap-gold-standard
Correct the erddap-deployment.yaml file to read something like:
apiVersion: apps/v1
kind: Deployment
metadata:
annotations:
kompose.cmd: kompose convert --volumes hostPath
kompose.version: 1.26.0 (40646f47)
creationTimestamp: null
labels:
io.kompose.service: erddap
name: erddap
spec:
replicas: 1
selector:
matchLabels:
io.kompose.service: erddap
strategy:
type: Recreate
template:
metadata:
annotations:
kompose.cmd: kompose convert --volumes hostPath
kompose.version: 1.26.0 (40646f47)
creationTimestamp: null
labels:
io.kompose.service: erddap
spec:
containers:
- env:
- name: ERDDAP_MAX_MEMORY
value: 2G
- name: ERDDAP_MIN_MEMORY
value: 1G
image: axiom/docker-erddap:2.18
name: erddap-gold-standard
ports:
- containerPort: 8080
resources: {}
volumeMounts:
- mountPath: /usr/local/tomcat/bin/config.sh
name: erddap-hostpath0
- mountPath: /usr/local/tomcat/webapps/ROOT/robots.txt
name: erddap-hostpath1
- mountPath: /usr/local/tomcat/content/erddap/
name: erddap-hostpath2
- mountPath: /erddapData/
name: erddap-hostpath3
- mountPath: /datasets/
name: erddap-hostpath4
- mountPath: /usr/local/tomcat/temp/
name: erddap-hostpath5
restartPolicy: Always
volumes:
- hostPath:
path: /erddap-gold-standard/erddap/conf/config.sh
name: erddap-hostpath0
- hostPath:
path: /erddap-gold-standard/erddap/conf/robots.txt
name: erddap-hostpath1
- hostPath:
path: /erddap-gold-standard/erddap/content/
type: Directory
name: erddap-hostpath2
- hostPath:
path: /erddap-gold-standard/erddap/data/
type: Directory
name: erddap-hostpath3
- hostPath:
path: /erddap-gold-standard/datasets/
type: Directory
name: erddap-hostpath4
- hostPath:
path: /tmp/
type: Directory
name: erddap-hostpath5
status: {}
Make a new terminal. Then:
kubectl apply -k ./
kubectl get pods
kubectl port-forward erddap-7c94b648f8-ljxhs 8080:8080 # Correct to the appropriate pod name
localhost should have it. You may need to use kubectl describe pod [pod name] to diagnose problems
Future work
From here, with a successful local development environment, we can interact with the datasets.xml file via commands like kubectl cp (ref) or get into the container itself via kubectl exec (ref). In deploying this to a live Kubernetes cluster, arrangements should be made to replace the hostpaths with better volume types and transferring the files to those, hopefully persistent, volumes. Service accounts and load balancers are also to be considered.