MongoDB Search and MongoDB Vector Search on Self-Managed Deployments

关于MongoDB `mongot`

mongot is the MongoDB Search and MongoDB Vector Search process that powers the $search, $searchMeta, and $vectorSearch aggregation stages. Built on Apache Lucene, mongot runs as a separate process from mongod and performs the following functions:

Synchronizes index data from mongod over a permanent connection driven by change streams.
Maintains search indexes on dedicated storage.
Serves search queries proxied through mongod. Clients never connect to mongot directly.

Atlas manages mongot for you. On Community Edition and MongoDB Enterprise, you deploy and operate mongot yourself.

Try MongoDB Search and MongoDB Vector Search Locally

To evaluate MongoDB Search and MongoDB Vector Search, start with the mongodb/mongodb-atlas-local Docker image. This image bundles mongod and mongot in a single container so you can focus on building MongoDB Search and MongoDB Vector Search features.

Use atlas-local when you want to:

Prototype $search and $vectorSearch queries against your own data.
Develop and run integration tests in CI.
Try automated embedding before choosing a production model.

Use atlas-local for evaluation only. The image runs as a single node and lacks high availability, so don't use it as a production deployment target.

Get started with the following commands:

docker pull mongodb/mongodb-atlas-local:preview
docker run -p 27017:27017 mongodb/mongodb-atlas-local

For the full walkthrough, see MongoDB Search and MongoDB Vector Search Local Development Quickstart and Atlas CLI: Deploy a Local Atlas Deployment with Search.

When you're ready to deploy to production, return to this page for guidance on production deployment paths.

Production Deployment Paths

Self-managed MongoDB Search and MongoDB Vector Search deployment options differ between MongoDB editions.

MongoDB Enterprise

The primary deployment pattern uses MongoDB Ops Manager to manage mongod on virtual machines. The MongoDB Controllers for Kubernetes Operator manages mongot on a Kubernetes cluster. This deployment supports replica sets and sharded clusters with multiple mongot instances, and lets you size mongot per workload independently of mongod. To learn more, see MongoDB Controllers for Kubernetes Operator documentation.

You can also run MongoDB Enterprise outside the Kubernetes environment and connect it directly to mongot instances on a Kubernetes cluster. To learn more, see MongoDB Controllers for Kubernetes Operator documentation.

注意

MongoDB Enterprise doesn't support standalone tarball or container deployments of mongot.

Community Edition

Community Edition provides the following installation artifacts:

Linux tarball: Install the mongot binary directly.
Container image: Pull and run the official mongot container.

You can either run these artifacts directly as a standalone or deploy them in Kubernetes using the Kubernetes Operator.

How `mongot` Works with `mongod`

A self-managed deployment runs two mongot processes per replica set:

Architecture diagram showing the relationship between the application, mongod, and mongot processes in a self-managed deployment.

点击放大

The client application connects only to mongod. mongod routes the MongoDB Search and MongoDB Vector Search aggregation stage requests to mongot over gRPC.
mongot connects to mongod to stream change events and maintain the index.
mongot writes index segments to dedicated storage.

后续步骤

目标	Resource
Evaluate or build features for the first time	MongoDB Search and MongoDB Vector Search Local Development Quickstart
Install Community Edition on Linux or Docker	Install or Uninstall `mongot` on Linux or Install or Uninstall MongoDB Search on Community in a Docker Container
Deploy on Kubernetes (Community or Enterprise)	适用于Kubernetes的MongoDB控制器
Confirm an existing installation works	Verify Your mongot Connection
Size before deploying	mongot部署大小调整简介

来年

Compatibility and Requirements