Overview

dfx is a Linux-based dynamic dataflow executor.

It constructs, runs, and manages directed graphs of computational nodes that exchange messages through typed channels. Each node performs a unit of work, while channels define message flow and execution dependencies between nodes.

The primary responsibilities of dfx are:

• loading and validating dataflow graphs,
• executing nodes asynchronously and safely,
• coordinating message delivery between nodes,
• integrating external event sources (file descriptors, timers, processes),
• managing runtime concurrency and scheduling.

Graph model

A graph is composed of:

• Nodes: stateful computation units with input and output ports,
• Channels: directed connections between node ports,
• Messages: data objects flowing through channels.

Graphs are defined using .dfx configuration files and may be modified at runtime (nodes added/removed, channels rewired) without stopping execution.

Execution model

Execution is fully asynchronous and event-driven.

Key characteristics:

• Nodes are executed on a shared thread pool.
• Only one thread may execute inside a given node at any time.
• Message delivery may be performed inline (fast-path) or deferred, depending on runtime conditions and node policies.
• Deep or cyclic graphs are protected against stack overflow by enforcing bounded inline execution depth.

The runtime uses:

• a dedicated FD listener thread for I/O readiness detection,
• a task-based scheduler to dispatch work to worker threads,
• node-aware serialization to guarantee safe execution.

Runtime control

While running, dfx exposes a Unix domain socket control interface that allows:

• graph inspection,
• dynamic graph modification,
• runtime configuration updates,
• node lifecycle control.

This enables external tools or daemons to control a running instance of dfx without restarting it.

Scalability

dfx is designed to scale across:

• multiple CPU cores,
• varying workloads,
• graphs with heterogeneous node behavior.

Concurrency is controlled centrally by the runtime scheduler, which balances throughput, latency, and safety.

Platform constraints

• Target platform: Linux
• Execution relies on Linux primitives (epoll, eventfd, Unix domain sockets).
• No GUI or display system is required.

What dfx is (and is not)

dfx is:

• a runtime execution engine,
• a message-driven computation framework,
• a foundation for building higher-level tooling and orchestration.

dfx is not:

• a UI application,
• a static batch processor,
• a scripting language runtime.

Higher-level management, orchestration, or visualization is expected to be provided by companion binaries and tools.