This document describes the basic architecture of the Pinecone database.
The Pinecone vector database is a cloud-based service deployed partly on Kubernetes. Pinecone serves control plane requests from an API gateway and routes these requests to user indexes; clients make data plane requests directly to pods. See Fig. 1 below.
Figure 1: Pinecone architecture diagram
- Pods. Each index has one or more replicas, each of which is deployed to a pod with assigned SSD and memory capacity. The CPU assigned to the pod performs computation; the SSD stores metadata.
- A stream processor indexes vectors.
- The blob store contains persistent snapshots of replica data.
For more information about security and encryption, see Security.
Updated 3 months ago