IBM Connections Component Pack (Pink) 6.0.0.7 Install script

IBM released their Component Pack (formerly known as Pink) in version 6.0.0.7 for Connections. It contains awesome technology based on modern tools like Docker and Kubernetes. Unfortunately, it's hard to install and doesn't care about restrictions on custom storage providers like nfs. This script should make it consistent and create some documentation, so that we can full automate this using Ansible later.

Prerequisites

• Download Component Pack in IBMs Fix Central by selecting IBM Connections on Linux, search for Text Component Pack and select the latest version (6.0.0.8 at time of writing). Pre-Selected here
• Since IBM doesn't have a Docker registry nor Helm repo, they only provide a package with exported packed images. See this documentation
• Manually deploy Component Pack archives to this registry
• A kubernetes cluster (I'm using Rancher's rke) with all Prerequisites for Component Pack
• Cloned copy of this repo to a work/or jumphost that can access all required machines

Installation process

Fix broken pvc

IBM outsourced their pvcs to connections-persistent-storage-nfs-0.1.0.tgz. Think about before you install this package! It's not only unflexible to have everything mounted in /pv-connections. It will balso break modern dynamic provisioners like nfs-client-provisioner since IBM uses incompatible labels.

Since we noticed that all apps automatically allocate their pvcs anyway, we skipped the pvc archive. For customizer and elasticsearch, this doesn't seem to apply. So I reversed the required pvcs from those archive and commented out all problematic labels. You need to apply them:

kubectl apply -f pvc/customizer-pvc.yml
kubectl apply -f pvc/elasticsearch-pvc.yml

Set evironment variables

Many valules like the k8s namespace are widely used across all commands. To keep things DRY and flexible, we centralized them using a few variables. Modify and set them according to your setup:

# Temp docker registry with IBMs images pushed to
DOCKERREGISTRY=registry.example.com
REGISTRYUSER=admin
REGISTRYPASSWORD=xxx
# Path to extracted IC-ComponentPack-6.0.0.7 folder
DOWNLOADPATH=/install/IC-ComponentPack-6.0.0.7
# IBM Connections
CNXDOMAIN=cnx.local
ICADMIN=icadmin
# Important to escape dollar sign since otherwise bash would recognize it as variable - Or use single ticks 'value'
ICADMINPW=yyy
ICCNXHOST=cnx.local
# Host or IP of your kubernetes master node
K8SMASTERIP=k8s.local
CLUSTERNAME=cluster01.internal
STORAGECLASS=nfs-client
NAMESPACE=connections-test
# 'General' password used for redis, solr and in some places of elasticsearch like ca password
COMPPASSWORD=zzz

Now you can deploy the packed helm charts using install.sh. Since we ran into different issues, I'd recomment not to execute the hole script. Instead, copy/execute them command by command.

You may ask why I generate long, complex --set calls on CLI instead of a clean yaml file passed to helm. Well, I tried this but gave it up since the IBM charts sometimes have default values set in their charts (or parent chart). Currently it doesn't seem possible to override them, expect using --set. A fix is already merged, so I'd expect the release soon.

Fix for custom-named clusters

Dont know the name of your cluster? Well, it's not possible to simply ask the cluster. But when using rke, you can see it in the cluster configuration file in services.kubelet.cluster_domain. If you haven't change it, skip this paragraph.

Thanks to stoeps13, IBM got informed that not all people let the default name. So this component pack update let users choose their cluster name. This is important when your k8s cluster is named something other than cluster.local. Sadly this wasn't done consistently. The Mongodb Statefulset still contains hardcoded cluster.local code, which leads to crashing containers.

To fix this, the following hacky changes on the yml files are required: Go to microservices_connections/hybridcloud/helmbuilds and unpack the infrastructure tgz file

tar xfvz infrastructure-0.1.0-20190205-020035.tgz

Now apply two changes on infrastructure/charts/mongodb/templates/statefulset.yaml

Set regular cluster domain

mongo-k8s-sidecar provides a KUBERNETES_CLUSTER_DOMAIN environment variable, which IBM doesn't forward to the pod. Go to env section of mongo-sidecar pod and add it:

- name: mongo-sidecar
  # ...
  env:
    # ...
    - name: KUBERNETES_CLUSTER_DOMAIN
      value: "customized-cluster.internal"

Replace in mongo.js file

/opt/cvallance/mongo-k8s-sidecar/src/lib/mongo.js must be changed. I don't see any other way than modifying the entrypoint. This way replacing is guaranteered to happen before mongo starts. Search for name: mongo-sidecar and modify command to ["/bin/bash"]. Add the following args:

- name: mongo-sidecar
  command: ["/bin/bash"]
  args:
    [
      "-c",
      "sed -i -e s/cluster.local/customized-cluster.local/g /opt/cvallance/mongo-k8s-sidecar/src/lib/mongo.js; /opt/cvallance/mongo-k8s-sidecar/entrypoint.sh",
    ]

Re-Pack

Now re-pack the modified infrastructure folder:

tar cvzf infrastructure-0.1.0-20190205-020035.tgz infrastructure

Disclaimer

I'm not associated to IBM, HZL or any other company behind Connections. This scripts and documentations are provided in the hope that they will be useful to others and spare headaches. I don't provide any warranty. Brands are copyrighted and only used to describe how to get the product up and running. A valid licence is required to run proprietary software legally.