Sourcegraph Managed Services Platform (MSP)

The Sourcegraph Managed Services Platform (MSP) is the standardized tooling and infrastructure for deploying and operating managed Sourcegraph services. MSP takes a service specification and generates Terraform manifests and adjacent resources required to operate a service, aiming to provide a simple, Heroku-like experience to spin up infrastructure for standalone managed services.

By adopting MSP for your managed service, it will benefit from an expanding set of features and integrations, alignment with infrastructure and security best practices at Sourcegraph, and support from the Core Services team.

For interacting with existing MSP services, see operating services. Ready to spin up a new service? Check out our Getting started guide!

Use cases

Any “managed service” - internal or customer-facing, for testing or for production - can be operated on Managed Services Platform! Today, MSP operates both internal and external services from many teams across Sourcegraph - see the Managed Services infrastructure page for a generated listing.

For an intro on what “managed services” are and how MSP can help you, check out this Loom introduction: Creating and Operating Managed Services at Sourcegraph (Merge 2024), and refer to features to see what MSP can offer.


MSP supports single-container:

  • stateless, horizontally scaling services
  • scheduled cron jobs

From a simple service configuration YAML (examples) and the sg msp toolchain for managing configuration, we currently support:

  • Generating infrastructure-as-code, deployed via Terraform Cloud
  • Service initialization and runtime boilerplate via sourcegraph/lib/managedservicesplatform, which includes:
    • initialization of OpenTelemetry tracing and metrics, logging, and error reporting (Sentry)
    • integration guidance for provisioned data backends like Redis and PostgreSQL
  • Provisioning of data backends, configured with secure, highly available defaults and regular backups out of the box where applicable:
    • Redis for ephemereal data and synchronization between instances of a service.
    • PostgreSQL for persistent, relational data.
    • BigQuery dataset and tables for high-volume analytics and usage data specific to your feature.
  • Service-specific features
    • Configuring a domain and TLS through Cloudflare and GCP load balancing
    • Scaling capabilities backed by Cloud Run
  • Job-specific features
  • Commands for easy access to infrastructure
    • Shortcuts to relevant UIs in sg msp tfc view, sg msp logs, etc.
    • Securely connect to your PostgreSQL instance using sg msp pg connect
  • Generated infrastructure guidance, rendered in the Managed Services infrastructure pages.
  • Continuous delivery via Cloud Deploy delivery pipelines

See our GitHub roadmap and 2023 Managed Services Platform (MSP) proof-of-concept update for more details on things we will be adding to MSP.

Operating services

All infrastructure manifests are managed in sourcegraph/managed-services, and the tooling is being developed in sourcegraph/sourcegraph/dev/sg/msp.

Getting started

To get started, you will need to write some code and build the service for distribution in MSP. Then, you can refer to creating and configuring infrastructure to get your service up and running!

Service code

The Core Services team recommends building your service in Go to leverage the service initialization and runtime boilerplate provided by the standalone module.

The runtime.Start function outlines the expected “contract” the MSP runtime expects services to fulfill, and ensures your service is compatible with MSP infrastructure:

import (

  // Your implementation!

func main() {

In your implementation of runtime.Service, the primary entrypoint Initialize provides a runtime.Contract that is pre-configured with MSP defaults and offers helpers to integrating with MSP-provisioned resources. For example:

  • to serve your service, you must use (runtime.Contract).Port, listening on all network interfaces, i.e.$PORT, or :$PORT.
    • do not use localhost:$PORT or$PORT.
  • to get a securely authenticated PostgreSQL connection, you should use (runtime.Contract).PostgreSQL.OpenDatabase(...)
  • Sentry error reporting integration for error-level logs is automatically configured when using the provided logger instance

A full example service is available in cmd/msp-example that makes use of most MSP functionality.

Service images

Every MSP service requires a runnable server in a Docker image whose platform is linux/amd64.

When publishing images for MSP to consume, you can use the public Docker registry, or an Artifact Registry repository within the Sourcegraph GCP organization. Image repositories published by the sourcegraph/sourcegraph monorepo work as well.

When using a private image registry within GCP, MSP will automatically provision the prerequisite permissions for MSP to access your images.

Creating and configuring infrastructure

Refer to the sourcegraph/managed-services README for all documentation related to creating configuring MSP deployments and getting started with sg msp.