Skip to content

Latest commit

 

History

History
103 lines (79 loc) · 5.02 KB

design.md

File metadata and controls

103 lines (79 loc) · 5.02 KB
Raw

Overview and Motivation

node-webworkers aims to implement as much of the HTML5 Web Workers API as is practical and useful in the context of NodeJS. Extensions to the HTML5 API are provided where it makes sense (e.g. to allow file descriptor passing).

The motivation for providing this API to NodeJS is as follows

  • A set of standard (well, emerging standard anyway) platform-independency concurrency APIs is a useful abstraction. Particularly as HTML5 gains wider adoption and JavaScript developers are likely to familiar with Web Workers from doing browser development. The set of Node.JS primitives for managing processes, child_process provides a lot of utility, but is easily misunderstood by developers who have not developed for a UNIX platofrm before (e.g. "why does kill() not kill my process?").

    In addition, the error reporting APIs in the Web Workers spec are more full-featured and JavaScript-specific than that provided natively by child_process (e.g. one can get a stack trace, etc).

  • Existing communicaiton mechanisms with child processes involve communicating over stdin/stdout. These are opaque byte streams and require the application to implement their own framing logic to discern message boundaries. Further, use of these built-in streams prevents sys.puts() and friends from working.

  • Shared Workers provide a useful naming service for communicating with other workers by name. Without this, the application must maintain its own routing structure.

    Note that shared workers are not implemented yet.

Design

The design that follows for Web Workers is motivated by a handful of underlying assumptions / philosophies:

  • Worker instances should be relatively long-lived. That is, it is not considered an important workload to be able to create and destroy thousands of workers as quickly as possible. Passing messages to existing workers to dispatch work items is favored over creating a new worker for each work item.

  • In the future, it will be desirable to run workers off-box, and to implement workers in other application frameworks / languages. This is particularly relevant in the choice of communication medium.

In addition, there is a general preference to embracing standards and existing building blocks, particularly those that are geared towards JavaScript and/or HTTP.

Worker Processes

Each worker executes in its own self-contained node process rather than as a separate thread and V8 context within the master process.

Benefits of this approach include fault isolation (any worker running out of memory or triggering some buggy C++ code will not take down other workers); avoiding the complexity of managing multiple event loops in a single process; and typical OSes are more likely to schedule different processes on different CPUs (this may not always happen for multiple threads within the same process), allowing the application to utilize multiple CPUs.

Drawbkacks include the cost of context switching between workers being more expensive when using a process-per-worker model than it would be in a thread-per-worker model; passing messages between processes typically requires a data copy and always requires serializing data; and overhead of spawning a new process.

Worker Context

Each worker is launched by webworker-child.js.

This script is passed to node as the entry point for the process and is responsible for constructing a V8 script context populated with bits relevant to the Web Worker API (e.g. the postMessage(), close() and location primitives, etc).

Inter-worker Communication

The Web Workers spec provides a simple message passing API.

Under the covers, this is implemented by connecting each dedicated worker to its parent process with a UNIX domain socket. This is lower overhead than TCP, and allows for UNIX goodies like file descriptor passing.

Message passing is done over this UNIX socket by negotiating an HTML5 Web Socket connection over this transport. This is done to provide a reasonably-performant standards-based message framing implementation and to lay the groundwork or communicating with off-box workers via HTTP over TCP, which may be implemented in another application stack entirely (e.g. Java, etc). The overhead of negotiating and maintaining the Web Socket connection is 1 round trip for handshaking and the overhead of maintaining HTTP state objects (http_parser and such). The handshaking overhead is not considered an undue burden given that workers are expected to be relatively long-lived and the HTTP state overhead considered small.

The format of the messages themselves is JSON, serialized using JSON.stringify() and de-serialized using JSON.parse(). Significantly, the use of a framing protocol allows the Web Workers implementation to wait for an entire, complete JSON blob to arrive before invoking JSON.parse().