Quickstart with GKE

This is a quick start guide for deploying TigerGraph on Kubernetes with Google Kubernetes Engine (GKE).

Prerequisites

Single-server deployment

This section describes the steps to deploy, verify, and remove a single-server deployment of TigerGraph on GKE.

Deploy single server

Step 1: Generate deployment manifest. Clone the TigerGraph ecosystems repository and change into the k8s directory. You can edit the kustimization.yaml file in the gke folder to change the namespace and image name for your deployment. The default namespace is default. No need to edit the files if no changes are needed.

Next, run the ./tg script in the k8s directory to generate the deployment manifest for a single-server deployment. You can use the --prefix option to specify a prefix for your pods. The default prefix is tigergraph. A deploy directory will be created automatically, and you should find the manifest named tigergraph-gke-default.yaml in the directory.

$ ./tg gke kustomize -s 1

Step 2: Deploy manifest. Run kubectl apply to create the deployment using the manifest you generated in Step1.

$ kubectl apply -f deploy/tigergraph-gke-default.yaml

Verify single server

Run kubectl get pods to confirm that the pods were created successfully:

$ kubectl get pods
NAME              READY   STATUS    RESTARTS   AGE
installer-zsnb4   1/1     Running   0          4m11s
tigergraph-0      1/1     Running   0          4m10s

Run kubectl get services to find the IP addresses of the RESTPP service as well as the GUI service. You can then make curl calls to the IP address of tg-rest-service at port 9000 to make sure that RESTPP is running:

$ curl <restpp_ip>:9000/echo | jq .
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100    39  100    39    0     0    120      0 --:--:-- --:--:-- --:--:--   120
{
  "error": false,
  "message": "Hello GSQL"
}

You can also copy the IP address of the GUI service into your browser and visit port 14240 to make sure that GraphStudio is working.

Connect to single server

You can use kubectl to get a shell to the container or log in via ssh

# Via kubectl
kubectl exec -it tigergraph-0 -- /bin/bash

# Via ssh
ip_m1=$(kubectl get pod -o wide |grep tigergraph-0| awk '{print $6}')
ssh tigergraph@ip_m1

Remove single server resources

Use the tg script in the k8s directory of the repo to delete all cluster resources. Replace <namespace_name> with the name of the namespace within which you want to delete the resources. If you don’t specify a namespace, the command will delete the resources in the namespace default:

$ ./tg gke delete -n <namespace_name>

Cluster deployment

Once your GKE cluster is ready, you can start following the below steps to deploy a TigerGraph cluster on Kubernetes.

Deploy TigerGraph cluster

1. Generate Kubernetes manifest

Clone the TigerGraph ecosystem repository and change into the k8s directory:

$ git clone https://github.com/tigergraph/ecosys.git
$ cd ecosys/k8s

You can customize your deployment by editing the kustomize.yaml file in the gke directory. The tg script in the k8s folder offers a convenient way to make common customizations such as namespace, TigerGraph version, as well as cluster size. Run ./tg -h to view the help text on how to use the script.

Use the tg script in the k8s directory of the repo to create a Kubernetes manifest. Use -s or --size to indicate the number of nodes in the cluster. Use the --ha option to indicate the replication factor of the cluster, and the partitioning factor will be the number of nodes divided by the replication factor.

For example, the following command will create a manifest that will deploy a 3*2 cluster with a replication factor of 2 and a partitioning factor of 3.

$ ./tg gke kustomize -s 6 --ha 2

The command will create a directory named deployment with the manifest inside.

2. Deploy the cluster

Run kubectl apply to create the deployment

$ kubectl apply -f ./deploy/tigergraph-gke-default.yaml

Verify cluster

Run kubectl get pods to verify the pods were created successfully:

$ kubectl get pods
NAME              READY   STATUS    RESTARTS   AGE
installer-zsnb4   1/1     Running   0          4m11s
tigergraph-0      1/1     Running   0          4m10s
tigergraph-1      1/1     Running   0          75s

Run kubectl get services to find the IP addresses of the RESTPP service as well as the GUI service. You can then make curl calls to the IP address of tg-rest-service at port 9000 to make sure that RESTPP is running:

$ curl <restpp_ip>:9000/echo | jq .
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100    39  100    39    0     0    120      0 --:--:-- --:--:-- --:--:--   120
{
  "error": false,
  "message": "Hello GSQL"
}

You can also copy the IP address of the GUI service into your browser and visit port 14240 to make sure that GraphStudio is working.

Connect to instances

You can use kubectl to get a shell to the container or log in via ssh

# Via kubectl
kubectl exec -it tigergraph-0 -- /bin/bash

# Via ssh
ip_m1=$(kubectl get pod -o wide |grep tigergraph-0| awk '{print $6}')
ssh tigergraph@ip_m1

Delete cluster resources

Use the tg script in the k8s directory of the repo to delete all cluster resources. Replace <namespace_name> with the name of the namespace within which you want to delete the resources. If you don’t specify a namespace, the command will delete the resources in the namespace default:

$ ./tg gke delete -n <namespace_name>