TechPrufer
Get started
Palantir
Palantir logoPalantir · Feature trail

Ephemeral Kubernetes Compute Instances

1 post · Dec 2024

What this trail covers

The Ephemeral Kubernetes Compute Instances capability has evolved to formalize the management of Kubernetes clusters composed of immutable, short-lived nodes (compute instances) with a maximum lifespan of 48 hours. This approach enhances resilience and simplifies operations by treating nodes like ephemeral pods, enabling automated replacement of problematic or outdated instances, improving security posture by limiting persistent access, and enforcing engineering discipline for high availability. The system includes a termination pipeline with policy-driven node selection and graceful draining, as well as instance group upgrades by replacement to manage OS and configuration changes. Telemetry is used to monitor health during rolling upgrades between instance groups.

All Palantir trails
  1. The Benefits of Running Kubernetes on Ephemeral ComputeThis post details the evolution of Palantir's Rubix platform to manage Kubernetes clusters composed of immutable, short-lived nodes (compute instances) with a maximum lifespan of 48 hours. This approach enhances resilience and simplifies operations by treating nodes like ephemeral pods, enabling automated replacement of problematic or outdated instances, improving security posture by limiting persistent access, and enforcing engineering discipline for high availability. The system includes a termination pipeline with policy-driven node selection and graceful draining, as well as instance group upgrades by replacement to manage OS and configuration changes. Telemetry is used to monitor health during rolling upgrades between instance groups.