Dataiku DR Site - API Node

Registered Posts: 2

Our Current Dataiku setup in production site consists of these nodes:

  • Design node
  • Deployer node
  • Automation node
  • API node
  • Govern node

and currently we do not have DR site, and we are planning to have one..

Is it possible if in DR site, we only go for the API Node?

So when the Prod site is down, API node will still up in DR site along with any Model that already deployed there and can be accesst from another application via API

This will be crucial for us especially in managing infra and cost…

Operating system used: RHEL

Answers

  • Registered Posts: 15 ✭✭

    Hi!

    I hope that you are doing well.

    Your case would be better discussed in a support ticket, could you please submit one here so we can discuss on the best way to support your needs? ( https://support.dataiku.com/support/tickets/new )

    As general information, DSS API node is natively highly available and scalable. Both high availability and scalability are achieved by deploying multiple instances of the API node (see our documentation for more details : https://doc.dataiku.com/dss/latest/apinode/operations/ha-deployment.html )

    Thank you very much!

    Best,

    Yasmine

  • Dataiku DSS Core Designer, Neuron, Dataiku DSS Adv Designer, Registered, Neuron 2023 Posts: 2,363 Neuron

    Dataiku DR Site - API Node

    It is technically possible but we can't really say for sure without really knowing exactly what's happening in your API services. For instance a lot of people build API services that rely on the Designer node and/or Autonation node. Obviously if your API service relies on these nodes it will work if they are down. Some API services may also rely indirectly on the Designer node and/or Autonation node. This can happen when you use query enrichment and query a dataset in your API node.

    While you may be able to use only API services in the API node in a DR situation this will only cater for realtime scoring models. Batch scoring, data preparation and model retraining is always done using the Designer node and/or Autonation node.

    The final point to make is that the API node allows for more complex deployments using Kubernetes. Kubernetes allows you to have a highly available and scalable deployment which does not need to use a "DR site". So you should look into more modern highly available deployment patterns rather than just having a DR site.

Welcome!

It looks like you're new here. Sign in or register to get started.

Welcome!

It looks like you're new here. Sign in or register to get started.