Platformatic Runtime

Platformatic Runtime provides a unified environment for running multiple Platformatic microservices as a single monolithic deployment unit, for streamlined development.

Features

• Programmatic start: Start Platformatic Runtime programmatically in tests or other applications for enhanced integration.
• Monorepo support: Efficiently manage applications within a monorepo setup.
• Inter-application communication: Enable inter-application communication using private message passing to streamline application interactions.

Standalone usage

If you're only interested in the features available in Platformatic Runtime, you can replace platformatic with @platformatic/runtime in the dependencies of your package.json. This reduces the number of dependencies you need to import for your application.

Example configuration file

The following configuration file can be used to start a new Platformatic Runtime project. For more details on the configuration file, see the configuration documentation.

{
  "$schema": "https://schemas.platformatic.dev/@platformatic/runtime/2.0.0.json",
  "autoload": {
    "path": "./packages",
    "exclude": ["docs"]
  },
  "entrypoint": "entrypointApp"
}

Platformatic Runtime context

Every Platformatic Runtime application can be run as a standalone application or as a Platformatic Runtime application. Runtime application enables certain compile and runtime optimizations, enhancing performance and resource management. You can see the inter-application communication for more features.

Inter-application communication

Platformatic Runtime allows multiple microservice applications to run within a single process. Only the entrypoint binds to an operating system port and can be reached from outside the runtime.

Within the runtime, all inter-application communication happens by injecting HTTP requests into the running servers, without binding them to ports. This injection is handled by fastify-undici-dispatcher and undici-thread-interceptor.

Each microservice is assigned an internal domain name based on its unique ID. For example, a microservice with the ID awesome is given the internal domain of http://awesome.plt.local. The dispatcher packages module map that domain to the Fastify server running the awesome microservice. Any Node.js APIs based on Undici, such as fetch(), will then automatically route requests addressed to awesome.plt.local to the corresponding Fastify server.

Internal Mesh Network Architecture

The communication system is built on a sophisticated mesh network that creates seamless connectivity between all microservices in the runtime:

Domain-Based Routing System

The mesh network uses domain-based routing to intelligently direct requests:

• Domain Configuration: All internal services use the .plt.local domain suffix
• Hostname Matching: The interceptor examines each request's hostname and routes requests ending with .plt.local to the appropriate worker thread
• Case-Insensitive Resolution: Domain matching is case-insensitive for reliability
• Fallback Behavior: Requests not matching the internal domain are passed through to external network handlers

Request Flow and Processing

When a service makes a request to another service (e.g., http://auth.plt.local/verify), the complete flow involves:

1. Request Interception: The undici-thread-interceptor captures the outgoing HTTP request
2. Domain Validation: The system checks if the hostname ends with .plt.local
3. Service Resolution: The hostname auth is mapped to the corresponding worker thread(s)
4. Load Balancing: If multiple workers exist for the service, round-robin selection chooses the target
5. MessageChannel Communication: The request is serialized and sent via Node.js MessageChannel
6. Worker Processing: The target worker thread processes the request through its Fastify server
7. Response Streaming: The response is streamed back through the MessageChannel to the requesting service

Performance Characteristics

The mesh network is optimized for high-performance inter-service communication:

• Efficient Streaming: Large request and response bodies use MessagePort transfers to avoid concatenation overhead
• Concurrent Processing: Multiple worker threads can process requests simultaneously across different CPU cores
• Efficient Load Balancing: Round-robin worker selection operates in O(1) time complexity
• Asynchronous Communication: All inter-thread communication is non-blocking, preserving application responsiveness

Dynamic Mesh Management

The mesh network automatically adapts to runtime changes:

• Automatic Worker Registration: New worker threads are automatically registered with the mesh interceptor
• Dynamic Route Updates: Services can be added or removed during runtime without affecting other services
• Health-Aware Routing: Unhealthy workers are automatically excluded from the routing pool
• Graceful Failover: When workers are restarted or replaced, the mesh maintains service availability

This architecture enables microservices to communicate using standard HTTP APIs while benefiting from the performance and isolation advantages of a multi-threaded execution environment.

For detailed technical documentation about the mesh network implementation, threading architecture, and performance characteristics, see the Multithread Architecture documentation.

Threading and networking model

By default, each application is executed in a separate and dedicated Node.js Worker Thread within the same process. This means that worker.isMainThread will return false and there are some limitations like the inability to use process.chdir.

The application application runtime configuration is accessible via the workerData and globalThis.platformatic objects, which allows to bypass such limitations.

If an application requires to be executed in a separate process, Platformatic Runtime will take care of setting globalThis.platformatic and the interapplication communication automatically.

TrustProxy

For each application in the runtime except the entrypoint, Platformatic will set the Fastify's trustProxy option to true. This will change the ip/hostname in the request object to match the one coming from the entrypoint, rather than the internal xyz.plt.local name.This is useful for applications behind a proxy, ensuring the original client's IP address is preserved. Visit fastify docs for more details.

Issues

If you run into a bug or have a suggestion for improvement, please raise an issue on GitHub or join our Discord feedback channel.