Deploy Fusion 5 on Google Kubernetes Engine (GKE)

Table of Contents

Prerequisites
Set up the Google Cloud SDK (one time only)
Create a single-node demo cluster
Create a three-node regional cluster to withstand a zone outage
Verifying the Fusion Installation
More deployment options

Fusion supports deployment on Google Kubernetes Engine (GKE). This topic explains how to deploy a Fusion cluster on GKE using the setup_f5_gke.sh script in the fusion-cloud-native repository.

Prerequisites

This section covers prerequisites and background knowledge needed to help you understand the structure of this document and how the Fusion installation process works with Kubernetes.

Before installing Fusion, you need to choose a Kubernetes namespace to install Fusion into. Think of a K8s namespace as a virtual cluster within a physical cluster. You can install multiple instances of Fusion in the same cluster in separate namespaces. However, please do not install more than one Fusion release in the same namespace.

NOTE: All Fusion services must run in the same namespace, i.e. you should not try to split a Fusion cluster across multiple namespaces.

Use a short name for the namespace, containing only letters, digits, or dashes (no dots or underscores). The setup scripts in this repo use the namespace for the Helm release name by default.

Install Helm

Helm is a package manager for Kubernetes that helps you install and manage applications on your Kubernetes cluster. Regardless of which Kubernetes platform you’re using, you need to install helm as it is required to install Fusion for any K8s platform. On MacOS, you can do:

brew install kubernetes-helm

If you already have helm installed, make sure you’re using the latest version:

brew upgrade kubernetes-helm

For other OS, please refer to the Helm installation docs: https://helm.sh/docs/using_helm/

The Fusion helm chart requires that helm is greater than version 3.0.0; check your Helm version by running helm version --short.

Helm User Permissions

If you require that fusion is installed by a user with minimal permissions, instead of an admin user, then the role and cluster role that will have to be assigned to the user within the namespace that you wish to install fusion in are documented in the install-roles directory.

When working with Kubernetes on the command-line, it’s useful to create a shell alias for kubectl, e.g.:

alias k=kubectl

To use these role in a cluster, as an admin user first create the namespace that you wish to install fusion into:

k create namespace fusion-namespace

Apply the role.yaml and cluster-role.yaml files to that namespace

k apply -f cluster-role.yaml
k config set-context --current --namespace=$NAMESPACE
k apply -f role.yaml

Then bind the rolebinding and clusterolebinding to the install user:

k create --namespace fusion-namespace rolebinding fusion-install-rolebinding --role fusion-installer --user <install_user>
k create clusterrolebinding fusion-install-rolebinding --clusterrole fusion-installer --user <install_user>

You will then be able to run the helm install command as the <install_user>

Clone fusion-cloud-native from Github

You should clone this repo from github as you’ll need to run the scripts on your local workstation:

git clone https://github.com/lucidworks/fusion-cloud-native.git

You should get into the habit of pulling this repo for the latest changes before performing any maintenance operations on your Fusion cluster to ensure you have the latest updates to the scripts.

cd fusion-cloud-native
git pull

Cloning the github repo is preferred so that you can pull in updates to the scripts, but if you are not a git user, then you can download the project: https://github.com/lucidworks/fusion-cloud-native/archive/master.zip. Once downloaded, extract the zip and cd into the fusion-cloud-native-master directory.

The setup_f5_gke.sh script provided in this repo is strictly optional. The script is mainly to help those new to Kubernetes and/or Fusion get started quickly. If you’re already familiar with K8s, Helm, and GKE, then you can skip the script and just use Helm directly to install Fusion into an existing cluster or one you create yourself using the process described here.

If you’re new to Google Cloud Platform (GCP), then you need an account on Google Cloud Platform before you can begin deploying Fusion on GKE.

Set up the Google Cloud SDK (one time only)

If you’ve already installed the gcloud command-line tools, you can skip to Create a Fusion cluster in GKE.

These steps set up your local Google Cloud SDK environment so that you’re ready to use the command-line tools to manage your Fusion deployment.

Usually, you only need to perform these setup steps once. After that, you’re ready to create a cluster.

For a nice getting started tutorial for GKE, see: https://codelabs.developers.google.com/codelabs/cloud-gke-workshop-v2/#1

How to set up the Google Cloud SDK

Enable the Kubernetes Engine API.
Log in to Google Cloud: gcloud auth login
Set up the Google Cloud SDK:
1. gcloud config set compute/zone <zone-name>
  
  If you are working with regional clusters instead of zone clusters, use gcloud config set compute/region <region-name> instead.
2. gcloud config set core/account <email address>
3. New GKE projects only: gcloud projects create <new-project-name>
  
  If you have already created a project, for example in the Google Cloud Platform console, then skip to the next step.
4. gcloud config set project <project-name>

Make sure you install the Kubernetes command-line tool kubectl using:

gcloud components install kubectl
gcloud components update

Create a single-node demo cluster

Run the setup_f5_gke.sh script to install Fusion 5.x in a GKE cluster. To create a new, single-node demo cluster and install Fusion, simply do:

./setup_f5_gke.sh -c <cluster_name> -p <gcp_project_id> --create demo

Use the --help option to see script usage. If you want the script to create a cluster for you, then you need to pass the --create option with either demo or multi_az. If you don’t want the script to create a cluster, then you need to create a cluster before running the script and simply pass the name of the existing cluster using the -c parameter.

If you pass --create demo to the script, then we create a single node GKE cluster (defaults to using n1-standard-8 node type). The minimum node type you’ll need for a 1 node cluster is an n1-standard-8 (on GKE) which has 8 CPU and 30 GB of memory. This is cutting it very close in terms of resources as you also need to host all of the Kubernetes system pods on this same node. Obviously, this works for kicking the tires on Fusion 5.1 but is not sufficient for production workloads.

You can change the instance type using the -i parameter; see: https://cloud.google.com/compute/docs/regions-zones/#available for an list of which machine types are available in your desired region.

Note: If not provided the script generates a custom values file named gke_<cluster>_<namespace>_fusion_values.yaml which you can use to customize the Fusion chart.

WARNING If using Helm V2, the setup_f5_gke.sh script installs Helm’s tiller component into your GKE cluster with the cluster admin role. If you don’t want this, then please upgrade to Helm v3.

If you see an error similar to the following, then wait a few seconds and try running the setup_f5_gke.sh script again with the same arguments as this is usually a transient issue:

Error: could not get apiVersions from Kubernetes: unable to retrieve the complete list of server APIs: metrics.k8s.io/v1beta1: the server is currently unable to handle the request

After running the setup_f5_gke.sh script, proceed to the Verifying the Fusion Installation section below.

When you’re ready to deploy Fusion to a production-like environment, refer to the Planning section of the Survival Guide.

Create a three-node regional cluster to withstand a zone outage

With a three-node regional cluster, nodes are deployed across three separate availability zones.

./setup_f5_gke.sh -c <cluster> -p <project> -n <namespace> --region <region-name> --create multi_az

<cluster> value should be the name of a non-existent cluster; the script will create the new cluster.
<project> must match the name of an existing project in GKE. Run gcloud config get-value project to get this value, or see the GKE setup instructions.
<namespace> Kubernetes namespace to install Fusion into, defaults to default with release f5
<region-name> value should be the name of a GKE region, defaults to us-west1. Run gcloud config get-value compute/zone to get this value, or see the GKE setup instructions to set the value.

In this configuration, Kubernetes deploys a ZooKeeper and Solr pod on each of the three nodes, which allows the cluster to retain ZK quorum and remain operational after losing one node, such as during an outage in one availability zone.

When running in a multi-zone cluster, each Solr node has the solr_zone system property set to the zone it is running in, such as -Dsolr_zone=us-west1-a.

After running the setup_f5_gke.sh script, proceed to the Verifying the Fusion Installation section below.

When you’re ready to deploy Fusion to a production-like environment, refer to the Planning section of the Survival Guide.

GKE Ingress and TLS

The Fusion proxy service provides authentication and serves as an API gateway for accessing all other Fusion services. It’s typical to use an Ingress for TLS termination in front of the proxy service.

The setup_f5_gke.sh supports creating an Ingress with an TLS cert for a domain you own by passing: -t -h <hostname>

After the script runs, you need to create an A record in GCP’s DNS service to map your domain name to the Ingress IP. Once this occurs, our script setup uses Let’s Encrypt to issue a TLS cert for your Ingress.

To see the status of the Let’s Encrypt issued certificate, do:

kubectl get managedcertificates -n <namespace> -o yaml

Please refer to the Kubernetes documentation on configuring an Ingress for GKE: Setting up HTTP Load Balancing with Ingress

The GCP Ingress defaults to a 30 second timeout, which can lead to false negatives for long running requests such as importing apps. To configure the timeout for the backend in kubernetes:

Create a BackendConfig object in your namespace:

---
apiVersion: cloud.google.com/v1beta1
kind: BackendConfig
metadata:
  name: backend_config_name
spec:
  timeoutSec: 120
  connectionDraining:
    drainingTimeoutSec: 60

Then make sure that the following entries are in the right place in your values.yaml file:

api-gateway:
  service:
    annotations:
      beta.cloud.google.com/backend-config: '{"ports": {"6764":"backend_config_name"}}'

and upgrade your release to apply the configuration changes

Ingresses and externalTrafficPolicy

When running a fusion cluster behind an externally controlled LoadBalancer it can be advantageous to configure the externalTrafficPolicy of the proxy service to Local. This preserves the client source IP and avoids a second hop for LoadBalancer and NodePort type services, but risks potentially imbalanced traffic spreading. Although when running in a cluster with a dedicated pool for spark jobs that can scale up and down freely it can prevent unwanted request failures. This behaviour can be altered with the api-gateway.service.externalTrafficPolicy value, which is set to Local if the example values file is used.

You must use externalTrafficPolicy=Local for the Trusted HTTP Realm to work correctly.

If you are already using a custom 'values.yaml' file, create an entry for externalTrafficPolicy under api-gateway service.

api-gateway:
  service:
    externalTrafficPolicy: Local

Considerations when using the nginx ingress controller

If you are using the nginx ingress controller to fulfil your ingress definitions there are a couple of options that are recommended to be set in the configmap:

enable-underscores-in-headers: "true"   # Fusion can return some headers that have underscores, these have to be explicitly enabled in nginx
proxy-body-size: "0"        # By default nginx places a maximum size on request bodies, either increase as needed or disable by setting to 0
proxy-read-timeout: "300"   # Increases the timeout for potential slow queries.

Custom values

There are some example values files that can be used as a starting point for resources, affinity and replica count configuration in the example-values folder. These can be passed to the install script using the --values option, for example:

./setup_f5_gke.sh -c <cluster> -p <project> -r <release> -n <namespace> \
  --values example-values/affinity.yaml --values example-values/resources.yaml --values example-values/replicas.yaml

The --values option can be passed multiple times, if the same configuration property is contained within multiple values files then the values from the latest file passed as a --values option are used.

Connectors custom values

If you are using Fusion 5.9 or later, you can specify resources and replica count per connector. This allows you to set different resource limits for each connector. If you do not set custom values for a connector, that connector uses the default values.

Set each connector’s resource values in the connector-plugin section under pluginValues. The pluginValues section is a list of plugins and its resources. The following sample shows an example.

 pluginValues:
   - id: "plugin-id" (1)
     resources: (2)
       limits:
         cpu: "2"
         memory: "3Gi"
       requests:
         cpu: "250m"
         memory: "2Gi"
     replicaCount: 1 (3)

1	The plugin ID. The plugin ID must match the plugin ID on the plugin ZIP file. without the `lucidworks.` prefix. For example, if the plugin ID on the plugin ZIP file is `lucidworks.sharepoint-optimized`, the plugin ID is `sharepoint-optimized`.
2	The resources settings. You may specify the limits, the requests, and the CPU and memory for each.
3	The number of replicas per connector. This value is 1 by default.

IMPORTANT After editing the connector-plugin section, you must reinstall the affected connector.

Upgrades and Ingress

IMPORTANT If you used the -t -h <hostname> options when installing your cluster, our script created an additional values yaml file named tls-values.yaml.

To make things easier for you when upgrading, you should add the settings from this file into your main custom values yaml file, e.g.:

api-gateway:
  service:
    type: "NodePort"
  ingress:
    enabled: true
    host: "<hostname>"
    tls:
      enabled: true
    annotations:
      "networking.gke.io/managed-certificates": "<RELEASE>-managed-certificate"
      "kubernetes.io/ingress.class": "gce"

This way you don’t have to remember to pass the additional tls-values.yaml file when upgrading.

Verifying the Fusion Installation

In this section, we provide some tips on how to verify the Fusion installation.

Check if the Fusion Admin UI is available at https://<fusion-host>:6764/admin/.

Let’s review some useful kubectl commands.

Enhance the K8s Command-line Experience

Here is a list of tools we found useful for improving your command-line experience with Kubernetes:

krew (kubectl plugin mgr): https://github.com/kubernetes-sigs/krew/
kube-ps1 (show current context on command line prompt): https://github.com/jonmosco/kube-ps1
kubectx / kubens (switch between clusters / namespaces): https://github.com/ahmetb/kubectx

Useful kubectl commands

kubectl reference: https://kubernetes.io/docs/reference/generated/kubectl/kubectl-commands

Set the namespace for kubectl if not using the default:

kubectl config set-context --current --namespace=<NAMESPACE>

This saves you from having to pass -n with every command.

Get a list of running pods: k get pods

Get logs for a pod using a label: k logs –l app.kubernetes.io/component=query-pipeline

Get pod deployment spec and details: k get pods <pod_id> -o yaml

Get details about a pod events: k describe po <pod_id>

Port forward to a specific pod: k port-forward <pod_id> 8983:8983

SSH into a pod: k exec -it <pod_id> — /bin/bash

CPU/Memory usage report for pods: k top pods

Forcefully kill a pod: k delete po <pod_id> --force --grace-period 0

Scale up (or down) a deployment: k scale deployment.v1.apps/<id> --replicas=N

Get a list of pod versions: k get po -o jsonpath='{..image}' | tr -s '' '\n' | sort | uniq

Check Fusion Pods and Services

Once the install script completes, you can check that all pods and services are available using:

kubectl get pods

If all goes well, you should see a list of pods similar to:

NAME                                                        READY   STATUS    RESTARTS   AGE
seldon-controller-manager-6675874894-qxwrv                  1/1     Running   0          8m45s
f5-admin-ui-74d794f4f8-m5jms                                1/1     Running   0          8m45s
f5-ambassador-fd6b9b5dc-7ghf6                               1/1     Running   0          8m43s
f5-api-gateway-6b9998b9c-tmchk                              1/1     Running   0          8m45s
f5-auth-ui-7565564b4c-rdc74                                 1/1     Running   0          8m42s
f5-classic-rest-service-0                                   1/1     Running   3          8m44s
f5-devops-ui-77bb867ffb-fbzxd                               1/1     Running   0          8m42s
f5-fusion-admin-78b8f8fc7f-4d7l8                            1/1     Running   0          8m42s
f5-fusion-indexing-599c8d448-xzsvm                          1/1     Running   0          8m44s
f5-insights-665fd9f6fc-g5psw                                1/1     Running   0          8m43s
f5-job-launcher-84dd4c5c96-p8528                            1/1     Running   0          8m44s
f5-job-rest-server-6d44d964b8-xtnxw                         1/1     Running   0          8m45s
f5-logstash-0                                               1/1     Running   0          8m45s
f5-ml-model-service-6987dc94c9-9ppp8                        2/2     Running   1          8m45s
f5-monitoring-grafana-5d499dbb58-pzw72                      1/1     Running   0          10m
f5-monitoring-prometheus-kube-state-metrics-54d6678dv9h7h   1/1     Running   0          10m
f5-monitoring-prometheus-pushgateway-7d65c65b85-vwrwf       1/1     Running   0          10m
f5-monitoring-prometheus-server-0                           2/2     Running   0          10m
f5-pm-ui-86cbc5bb65-nd2n8                                   1/1     Running   0          8m44s
f5-pulsar-bookkeeper-0                                      1/1     Running   0          8m45s
f5-pulsar-broker-b56cc776f-56msx                            1/1     Running   0          8m45s
f5-query-pipeline-5d75d7d5f4-l2mdf                          1/1     Running   0          8m43s
f5-connectors-7bb6cfc65f-7wfs2                              1/1     Running   0          8m42s
f5-connectors-backend-987fdc648-dldwv                       1/1     Running   0          8m45s
f5-rules-ui-6b9d55b78f-9hzzj                                1/1     Running   0          8m43s
f5-solr-0                                                   1/1     Running   0          8m44s
f5-solr-exporter-c4687c785-jsm7x                            1/1     Running   0          8m45s
f5-ui-6cdbcc68c6-rj9cq                                      1/1     Running   0          8m45s
f5-webapps-6d6bb9bfd-hm4qx                                  1/1     Running   0          8m45s
f5-workflow-controller-7b66679fb7-sjbvp                     1/1     Running   0          8m44s
f5-zookeeper-0                                              1/1     Running   0          8m45s

The number of pods per deployment / statefulset will vary based on your cluster size and replicaCount settings in your custom values YAML file. Also, don’t worry if you see some pods having been restarted as that just means they were too slow to come up and Kubernetes killed and restarted them. You do want to see at least one pod running for every service. If a pod is not running after waiting a sufficient amount of time, use kubectl logs <pod_id> to see the logs for that pod; to see the logs for previous versions of a pod, use: kubectl logs <pod_id> -p. You can also look at the actions Kubernetes performed on the pod using kubectl describe po <pod_id>.

To see a list of Fusion services, do:

kubectl get svc

For an overview of the various Fusion 5 microservices, see: https://doc.lucidworks.com/fusion/5.3/149/fusion-microservices

Once you’re ready to build a Fusion cluster for production, please see the Fusion 5 Survival Guide in this repo.

Upgrading with Zero Downtime

One of the most powerful features provided by Kubernetes and a cloud-native microservices architecture is the ability to do a rolling update on a live cluster. Fusion 5 allows customers to upgrade from Fusion 5.x.y to a later 5.x.z version on a live cluster with zero downtime or disruption of service.

When Kubernetes performs a rolling update to an individual microservice, there will be a mix of old and new services in the cluster concurrently (only briefly in most cases) and requests from other services will be routed to both versions. Consequently, Lucidworks ensures all changes we make to our service do not break the API interface exposed to other services in the same 5.x line of releases. We also ensure stored configuration remains compatible in the same 5.x release line.

Lucidworks releases minor updates to individual services frequently, so our customers can pull in those upgrades using Helm at their discretion.

To upgrade your cluster at any time, use the --upgrade option with our setup scripts in this repo.

The scripts in this repo automatically pull in the latest chart updates from our Helm repository and deploy any updates needed by doing a diff of your current installation and the latest release from Lucidworks. To see what would be upgraded, you can pass the --dry-run option to the script.

Grafana Dashboards

Get the initial Grafana password from a K8s secret by doing:

kubectl get secret --namespace "${NAMESPACE}" ${RELEASE}-monitoring-grafana \
  -o jsonpath="{.data.admin-password}" | base64 --decode ; echo

With Grafana, you can either setup a temporary port-forward to a Grafana pod or expose Grafana on an external IP using a K8s LoadBalancer. To define a LoadBalancer, do (replace ${RELEASE} with your Helm release label):

kubectl expose deployment ${RELEASE}-monitoring-grafana --type=LoadBalancer --name=grafana --port=3000 --target-port=3000

You can use kubectl get services --namespace <namespace> to determine when the load balancer is setup and its IP address. Direct your browser to http://<GrafanaIP>:3000 and enter the username admin@localhost and the password that was returned in the previous step.

This will log you into the application. It is recommended that you create another administrative user with a more desirable password.

The dashboards and datasoure will be setup for you in grafana, simply navigate to Dashboards → Manage to view the vailable dashboards