Preparation & Prerequisites¶
Checks & steps to ensure a smooth installation.
Obtain a Kinetica License Key
A product license key will be required for install. Please contact Kinetica Support to request a trial key.
Failing to provide a license key at installation time will prevent the DB from starting.
Preparation and prerequisites¶
Free Resources
Your Kubernetes cluster version should be >= 1.22.x and have a minimum of 8 CPU, 8GB Ram and
SSD or SATA 7200RPM hard drive(s) with 4X memory capacity.
GPU Support
For GPU enabled clusters the cards below have been tested in large-scale production environments and provide the best performance for the database.
| GPU | Driver |
|---|---|
| P4/P40/P100 | 525.X (or higher) |
| V100 | 525.X (or higher) |
| T4 | 525.X (or higher) |
| A10/A40/A100 | 525.X (or higher) |
Kubernetes Cluster Connectivity¶
Installation requires Helm3 and access to an on-prem or CSP managed Kubernetes cluster. and the Kubernetes CLI kubectl.
The context for the desired target cluster must be selected from your ~/.kube/config file and set via the KUBECONFIG environment variable or kubectl ctx (if installed). Check to see if you have the correct context with,
and that you can access this cluster correctly with,
Get Nodes
If you do not see a list of nodes for your K8s cluster the helm installation will not work. Please check your Kubernetes installation or access credentials (kubeconfig).
Kinetica Images for an Air-Gapped Environment
If you are installing Kinetica with Helm in an air-gapped environment you will either need a Registry Proxy to pass the requests through or to download the images and push them to your internal Registry.
For information on ways to transfer the files into an air-gapped environment See here.
Required Container Images¶
docker.io (Required Kinetica Images for All Installations)¶
- docker.io/kinetica/kinetica-k8s-operator:{{kinetica_full_version}}
- docker.io/kinetica/kinetica-k8s-cpu:{{kinetica_full_version}} or
- docker.io/kinetica/kinetica-k8s-cpu-avx512:{{kinetica_full_version}} or
- docker.io/kinetica/kinetica-k8s-gpu:{{kinetica_full_version}}
- docker.io/kinetica/workbench-operator:{{kinetica_full_version}}
- docker.io/kinetica/workbench:{{kinetica_full_version}}
- docker.io/kinetica/kinetica-k8s-monitor:{{kinetica_full_version}}
- docker.io/kinetica/busybox:{{kinetica_full_version}}
- docker.io/kinetica/fluent-bit:{{kinetica_full_version}}
- docker.io/kinetica/kagent:7.1.9.15.20230823123615.ga
nvcr.io (Required Kinetica Images for GPU Installations using kinetica-k8s-gpu)¶
- nvcr.io/nvidia/gpu-operator:v23.9.1
registry.k8s.io (Required Kinetica Images for GPU Installations using kinetica-k8s-gpu)¶
- registry.k8s.io/nfd/node-feature-discovery:v0.14.2
docker.io (Required Supporting Images)¶
- docker.io/bitnami/openldap:2.6.7
- docker.io/alpine/openssl:latest (used by bitnami/openldap)
- docker.io/otel/opentelemetry-collector-contrib:0.95.0
quay.io (Required Supporting Images)¶
- quay.io/brancz/kube-rbac-proxy:v0.14.2
Optional Container Images¶
These images are only required if certain features are enabled as part of the Helm installation: -
- CertManager
- ingress-ninx
quay.io (Optional Supporting Images)¶
- quay.io/jetstack/cert-manager-cainjector:v1.13.3 (if optionally installing CertManager via Kinetica Helm Chart)
- quay.io/jetstack/cert-manager-controller:v1.13.3 (if optionally installing CertManager via Kinetica Helm Chart)
- quay.io/jetstack/cert-manager-webhook:v1.13.3 (if optionally installing CertManager via Kinetica Helm Chart)
registry.k8s.io (Optional Supporting Images)¶
- registry.k8s.io/ingress-nginx/controller:v1.9.4 (if optionally installing Ingress nGinx via Kinetica Helm Chart)
- registry.k8s.io/ingress-nginx/controller:v1.9.6@sha256:1405cc613bd95b2c6edd8b2a152510ae91c7e62aea4698500d23b2145960ab9c
Which Kinetica Core Image do I use?¶
| Container Image | Intel (AMD64) | Intel (AMD64 AVX512) | Amd (AMD64) | Graviton (aarch64) | Apple Silicon (aarch64) |
|---|---|---|---|---|---|
| kinetica-k8s-cpu | (1) | ||||
| kinetica-k8s-cpu-avx512 | |||||
| kinetica-k8s-gpu | (2) | (2) | (2) |
- It is preferable on an Intel AVX512 enabled CPU to use the kinetica-k8s-cpu-avx512 container image
- With a supported nVidia GPU.
Label the Kubernetes Nodes¶
Kinetica requires some of the Kubernetes Nodes to be labeled as it splits some of the components into different deployment 'pools'. This enables different physical node types to be present in the Kubernetes Cluster allowing us to target which Kinetica components go where.
e.g. for a GPU installation some nodes in the cluster will have GPUs and others are CPU only. We can put the DB on the GPU nodes and our infrastructure components on CPU only nodes.
The Kubernetes cluster nodes selected to host the Kinetica infrastructure pods i.e. non-DB Pods require the following label app.kinetica.com/pool=infra.
whilst the Kubernetes cluster nodes selected to host the Kinetica DB Pods require the following label app.kinetica.com/pool=compute.
The Kubernetes cluster nodes selected to host the Kinetica infrastructure pods i.e. non-DB Pods require the following label app.kinetica.com/pool=infra.
whilst the Kubernetes cluster nodes selected to host the Kinetica DB Pods require the following label app.kinetica.com/pool=compute-gpu.
On-Prem Kinetica SQLAssistant - Nodes Groups, GPU Counts & VRAM Memory
To run the Kinetica SQLAssistant locally requires additional GPUs to be available in a separate Node Group labeled app.kinetica.com/pool=compute-llm. In order for the On-Prem Kinetica LLM to run it requires 40GB GPU VRAM therefore the number of GPUs automatically allocated to the SQLAssistant pod will ensure that the 40GB VRAM is available e.g. 1x A100 GPU or 2x A10G GPU.
Pods Not Scheduling
If the Kubernetes are not labeled you may have a situation where Kinetica pods not schedule and sit in a 'Pending' state.
Install the kinetica-operators chart¶
This chart will install the Kinetica K8s operators together with a default configured database and workbench UI.
Add the Kinetica chart repository¶
Add the repo locally as kinetica-operators:
Helm Repo Add
Obtain the default Helm values file¶
For the generic Kubernetes install use the following values file without modification. Advanced users with specific requirements may need to adjust parameters in this file.
wget https://raw.githubusercontent.com/kineticadb/charts/72.2.3/kinetica-operators/values.onPrem.k8s.yaml
Determine the following prior to the chart install¶
Default Admin User
the default admin user in the Helm chart is kadmin but this is configurable. Non-ASCII characters and typographical symbols in the password must be escaped with a "\". For example, --set dbAdminUser.password="MyPassword\!"
- Obtain a LICENSE-KEY as described in the introduction above.
- Choose a PASSWORD for the initial administrator user
- As the storage class name varies between K8s flavor and/or there can be multiple, this must be prescribed in the chart installation. Obtain the DEFAULT-STORAGE-CLASS name with the command:
List StorageClass
use the name found after the /, For example, in storageclass.storage.k8s.io/local-path use "local-path" as the parameter.
Amazon EKS
If installing on Amazon EKS See here
Planning access to your Kinetica Cluster¶
Existing Ingress Controller?
If you have an existing Ingress Controller in your Kubernetes cluster and do not want Kinetica to install an ingresss-nginx to expose it's endpoints then you can disable ingresss-nginx installation in the values.yaml by editing the file and setting install: true to install: false: -
```` yaml
nodeSelector: {}
tolerations: []
affinity: {}
ingressNginx:
install: false
````