On-Premise Installation

Installation Flow

The Arize installation is simplified with the use of a Helm Chart and the Arize Operator. The installation flow consists of:

• An Operator Helm Chart which creates the necessary namespaces and roles and then bootstraps the Operator pod. It includes the configuration parameters for the Arize deployment. These parameters are derived from a values.yaml file that the user has crafted and tailored to the environment.

• The Arize Operator which based on the configuration parameters applies the manifests and builds the Arize deployment.

• The arize.sh utility script which assists the user in performing some of the tasks like transferring Arize images to an internal Docker registry (if applicable) and calling Helm to deploy the chart. Note that the user can also call Helm directly if preferred.

Prerequisites

This document assumes that a cluster has been created in GCP, AWS, Azure or another cloud option following specific guidelines or using Terraforms provided by Arize. Consult one of the guides below to get the proper cluster environment in place. These documents are included under ./docs and ./terraform in the distribution TAR file and describe the steps for creating a proper cluster along with supporting resources.

Key requirements:

• Kubernetes cluster

• Two storage buckets

• Storage class for the creation of volumes

Arize recommends a minimum of two node pools to get started, a Base Pool for basic functions and ingestion components and an ArizeDB Pool for the Historicals which serve the data queried by the Arize UI. If the creation of individual node pools is not an option, Arize can be deployed on a shared set of nodes by specifying historicalNodePoolEnabled: false in values.yaml.

<sizing-profile>	High Availability	Node Pool	Min Nodes	vCPU (per node)	RAM (per node)	Example Instance Type	Node Labels	Example Application
nonha	No	Base	1	16	128 GB	n2d-highmem-16(GCP) r5a.4xlarge(AWS) Standard_E16s_v5(AKS)	See diagram	- Integration testing - Staging enviroments
small1b	Yes	Base ArizeDB	3 2	8	64 GB	n2d-highmem-8(GCP) r5a.2xlarge(AWS) Standard_E8s_v5(AKS)	See diagram	- Hundreds of millions of traces or inferences
medium2b	Yes	Base ArizeDB	2 2	16	128 GB	n2d-highmem-16(GCP) r5a.4xlarge(AWS) Standard_E16s_v5(AKS)	See diagram	- Several billions of traces or inferences

Labels should be added to each node pool as shown below. The base pool can be configured for autoscaling.

A Mac or Linux machine with docker.io installed is recommended to run the installation script. Verify that Docker is already running using the command below.

$ docker --version

Verify that Helm is present on the computer.

$ helm version

Verify that curl is present on the computer.

$ curl --version

Verify that openssl is present on the computer.

$ openssl version

Also verify that kubectl is present on the computer.

$ kubectl version

For installation on GCP, verify that the GCP CLI is installed and authenticated.

$ gcloud --version
$ gcloud auth login

For installation on AWS, verify that the AWS CLI is installed and configured.

$ aws --version
$ aws configure

For installation on Azure, verify that the Azure CLI is installed and authenticated.

$ az version
$ az login