Using Worker Threads

Since Node 10.5.0, we have a new mechanism for running CPU-intensive algorithms outside of the main event loop called worker threads. Worker threads can be seen as a lightweight alternative to child_process.fork() with some extra advantages. Compared to processes, worker threads have a smaller memory footprint and a faster startup time since they run within the main process but inside different threads.

Even though worker threads are based on real threads, they don’t allow the deep synchronization and sharing capabilities supported by other languages such as Java or Python. This is because JavaScript is a single-threaded language and it doesn’t have any built-in mechanism to synchronize access to variables from multiple threads. JavaScript with threads simply wouldn’t be JavaScript. The solution to bring all the advantages of threads within Node.js without actually changing the language is worker threads.

Worker threads are essentially threads that, by default, don’t share anything with the main application thread; they run within their own V8 instance, with an independent Node.js runtime and event loop. Communication with the main thread is possible thanks to message-based communication channels, the transfer of ArrayBuffer objects, and the use of SharedArrayBuffer objects whose synchronization is managed by the user (usually with the help of Atomics).

This extensive level of isolation of worker threads from the main thread preserves the integrity of the language. At the same time, the basic communication facilities and data-sharing capabilities are more than enough for 99% of use cases.

Next, let’s use worker threads in our SubsetSum example.

Running the subset sum task in a worker thread

The worker threads API has a lot in common with the ChildProcess API, so the changes to our code will be minimal.

First, we need to create a new ThreadPool named class, which is our ProcessPool adapted to operate with worker threads instead of processes. The following code shows the differences between the new ThreadPool class and the ProcessPool class. There are only a few differences in the acquire() method, which are highlighted; the rest of the code is identical.