Collabora Online for Kubernetes

In order for Collaborative Editing and copy/paste to function correctly on kubernetes, it is vital to ensure that all users editing the same document and all the clipboard request end up being served by the same pod. Using the WOPI protocol, the https URL includes a unique identifier (WOPISrc) for use with this document. Thus load balancing can be done by using WOPISrc – ensuring that all URLs that contain the same WOPISrc are sent to the same pod.

Deploying Collabora Online in Kubernetes

  1. Install helm

  2. Setting up Kubernetes Ingress Controller

    1. Nginx:

    Install Nginx Ingress Controller

    1. HAProxy:

    Install HAProxy Ingress Controller

    Note

    Openshift uses minimized version of HAproxy called Router that doesn’t support all functionality of HAProxy but for COOL we need advance annotations Therefore it is recommended deploy HAproxy Kubernetes Ingress in collabora namespace

  3. Create an my_values.yaml (if your setup differs e.g. take an look in then values.yaml ./collabora-online/values.yaml) of the helmchart

    1. HAproxy:

    replicaCount: 3

ingress:
   enabled: true
   className: "haproxy"
   annotations:
      haproxy.org/timeout-tunnel: "3600s"
      haproxy.org/backend-config-snippet: |
         balance url_param WOPISrc check_post
         hash-type consistent
   hosts:
      - host: chart-example.local
        paths:
        - path: /
          pathType: ImplementationSpecific

image:
   tag: "latest"

autoscaling:
   enabled: false

collabora:
   aliasgroups:
      - host: "https://example.integrator.com:443"
   extra_params: --o:ssl.enable=false --o:ssl.termination=true
   # for production enviroment we recommend appending `extra_params` with `--o:num_prespawn_children=4`. It defines number of child processes to keep started in advance and waiting for new clients

resources:
   limits:
      cpu: "1800m"
      memory: "2000Mi"
   requests:
      cpu: "1800m"
      memory: "2000Mi"

# for production enviroment we recommend the following resource values
# resources:
   # limits:
      # cpu: "8000m"
      # memory: "8000Mi"
   # requests:
      # cpu: "4000m"
      # memory: "6000Mi"

    1. Nginx:

    replicaCount: 3

ingress:
   enabled: true
   className: "nginx"
   annotations:
      nginx.ingress.kubernetes.io/upstream-hash-by: "$arg_WOPISrc"
      nginx.ingress.kubernetes.io/proxy-body-size: "0"
      nginx.ingress.kubernetes.io/proxy-read-timeout: "600"
      nginx.ingress.kubernetes.io/proxy-send-timeout: "600"
   hosts:
      - host: chart-example.local
        paths:
        - path: /
          pathType: ImplementationSpecific

image:
   tag: "latest"

autoscaling:
   enabled: false

collabora:
   aliasgroups:
      - host: "https://example.integrator.com:443"
   extra_params: --o:ssl.enable=false --o:ssl.termination=true
   # for production enviroment we recommend appending `extra_params` with `--o:num_prespawn_children=4`. It defines number of child processes to keep started in advance and waiting for new clients

resources:
   limits:
      cpu: "1800m"
      memory: "2000Mi"
   requests:
      cpu: "1800m"
      memory: "2000Mi"

# for production enviroment we recommend the following resource values
# resources:
   # limits:
      # cpu: "8000m"
      # memory: "8000Mi"
   # requests:
      # cpu: "4000m"
      # memory: "6000Mi"

    Note

    • Horizontal Pod Autoscaling(HPA) is disabled for now. Because after scaling it breaks the collaborative editing and copy/paste. Therefore please set replicaCount as per your needs

    • If you have multiple host and aliases setup set aliasgroups in my_values.yaml:

    collabora:
   - host: "<protocol>://<host-name>:<port>"
     # if there are no aliases you can ignore the below line
     aliases: ["<protocol>://<its-first-alias>:<port>, <protocol>://<its-second-alias>:<port>"]
   # more host and aliases list is possible

    • Specify server_name when the hostname is not reachable directly for example behind reverse-proxy

    collabora:
   server_name: <hostname>:<port>

    • For production enviroment we recommended following resource values. We recommend appending extra_params with –o:num_prespawn_children=4. It defines number of child processes to keep started in advance and waiting for new clients

    resources:
   limits:
      cpu: "8000m"
      memory: "8000Mi"
   requests:
      cpu: "4000m"
      memory: "6000Mi"

    • In Openshift , it is recommended to use HAproxy deployment instead of default router. And add className in ingress block so that Openshift uses HAProxy Ingress Controller instead of Router:

    ingress:
   className: "haproxy"

  4. Install helm-chart using below command, it should deploy the collabora-online

    helm repo add collabora https://collaboraonline.github.io/online/
helm install --create-namespace --namespace collabora collabora-online collabora/collabora-online -f my_values.yaml

  5. Follow only if you are using NodePort service type in HAProxy and/or using minikube to setup, otherwise skip

    1. Each container port is mapped to a NodePort port via the Service object. To find those ports

      kubectl get svc --namespace=haproxy-controller

      Example output:

      |----------------|---------|--------------|------------|------------------------------------------|
|NAME            |TYPE     |CLUSTER-IP    |EXTERNAL-IP |PORT(S)                                   |
|----------------|---------|--------------|------------|------------------------------------------|
|haproxy-ingress |NodePort |10.108.214.98 |<none>      |80:30536/TCP,443:31821/TCP,1024:30480/TCP |
|----------------|---------|--------------|------------|------------------------------------------|

      In this instance, the following ports were mapped:

      • Container port 80 to NodePort 30536

      • Container port 443 to NodePort 31821

      • Container port 1024 to NodePort 30480

  6. Additional step if deploying on minikube for testing:

    1. Get minikube ip:

      minikube ip

      Example output:

      192.168.0.106

    2. Add hostname to /etc/hosts

      192.168.0.106   chart-example.local

    3. To check if everything is setup correctly you can run:

      curl -I -H 'Host: chart-example.local' 'http://192.168.0.106:30536/'

      It should return a similar output as below:

      HTTP/1.1 200 OK
last-modified: Tue, 18 May 2021 10:46:29
user-agent: COOLWSD WOPI Agent 6.4.8
content-length: 2
content-type: text/plain

Kubernetes cluster monitoring

  1. Install kube-prometheus-stack, a collection of Grafana dashboards, and Prometheus rules combined with documentation and scripts to provide easy to operate end-to-end Kubernetes cluster monitoring with Prometheus using the Prometheus Operator.

  2. Enable prometheus service monitor, rules and grafana in your my_values.yaml

    prometheus:
   servicemonitor:
      enabled: true
      labels:
         release: "kube-prometheus-stack"
   rules:
      enabled: true # will deploy alert rules
      additionalLabels:
         release: "kube-prometheus-stack"
grafana:
   dashboards:
      enabled: true # will deploy default dashboards

    Note

    Use kube-prometheus-stack as release name when installing kube-prometheus-stack helm chart because we have passed release=kube-prometheus-stack label in our my_values.yaml. For Grafana Dashboards you may need to enable scan in correct namespaces (or ALL), enabled by sidecar.dashboards.searchNamespace in Helmchart of grafana (which is part of PrometheusOperator, so grafana.sidecar.dashboards.searchNamespace)

Dynamic/Remote configuration in kubernetes

For big setups, you may not want to restart every pod to modify WOPI hosts, therefore it is possible to setup an additional webserver to serve a ConfigMap for using Remote/Dynamic Configuration

collabora:
   env:
      - name: remoteconfigurl
        value: https://dynconfig.public.example.com/config/config.json

 dynamicConfig:
   enabled: true

   ingress:
      enabled: true
      annotations:
      "cert-manager.io/issuer": letsencrypt-zprod
      hosts:
      - host: "dynconfig.public.example.com"
      tls:
      - secretName: "collabora-online-dynconfig-tls"
         hosts:
            - "dynconfig.public.example.com"

   configuration:
      kind: "configuration"
      storage:
         wopi:
         alias_groups:
            groups:
            - host: "https://domain1\\.xyz\\.abc\\.com/"
               allow: true
            - host: "https://domain2\\.pqr\\.def\\.com/"
               allow: true
               aliases:
                  - "https://domain2\\.ghi\\.leno\\.de/"

Note

In current state of COOL remoteconfigurl for Remote/Dynamic Configuration should be HTTPS.

Useful commands to check what is happening

Where is this pods, are they ready?

kubectl -n collabora get pod

example output :

NAME                                READY   STATUS    RESTARTS   AGE
collabora-online-5fb4869564-dnzmk   1/1     Running   0          28h
collabora-online-5fb4869564-fb4cf   1/1     Running   0          28h
collabora-online-5fb4869564-wbrv2   1/1     Running   0          28h

What is the outside host that multiple coolwsd servers actually answering?

kubectl get ingress -n collabora

example output :

|-----------|------------------|--------------------------|------------------------|-------|
| NAMESPACE |       NAME       |           HOSTS          |         ADDRESS        | PORTS |
|-----------|------------------|--------------------------|------------------------|-------|
| collabora | collabora-online |chart-example.local       |                        |  80   |
|-----------|------------------|--------------------------|------------------------|-------|

To uninstall the helm chart

helm uninstall collabora-online -n collabora