Performance and untrusted input

Every function is synchronous: it blocks the calling thread (the Node.js event loop, or the browser’s main thread) until it returns. Most calls on short strings take microseconds. Several functions, however, take time proportional to the product of their input lengths, so long input can block for seconds, for example text pasted into a search box or sent to your server.

Measured with the native addon on the benchmark machine:

Call	Input	Time
partialRatio, weightedRatio	1,000 vs 10,000 characters	0.4 s
partialRatio, weightedRatio	2,000 vs 20,000 characters	2.9 s
partialRatio	5,000 vs 50,000 characters	43 s
damerauLevenshtein	two 10,000-character strings	0.4 s
damerauLevenshtein	two 30,000-character strings	3.8 s
levenshtein	two 100,000-character strings	0.5 s
tokenSetRatio	two 100,000-character strings	0.45 s
search	1,000 items of 10,000 characters	0.5 s
jaroWinkler, sorensenDice	two 100,000-character strings	< 0.05 s

The WebAssembly build is slower than the native addon.

Keeping an application responsive

• Cap input lengths before calling rapid-fuzzy. Queries rarely need more than 100–200 characters, and partialRatio, weightedRatio and damerauLevenshtein should only see strings of bounded length.
• Use an index (FuzzyIndex or FuzzyObjectIndex) for repeated searches over the same items, and pass maxResults.
• Prefer *Many and typed-array variants for one string against many candidates.
• Move large jobs off the main thread with worker_threads in Node.js or a Web Worker in the browser. FuzzyIndex.fromAsync() builds an index on the libuv thread pool in Node.js.

Threads

With the native addon, a search that is estimated to take more than about a quarter of a millisecond is split across a thread pool, with one thread per CPU. Set RAYON_NUM_THREADS to limit the pool. Smaller searches run on the calling thread. The WebAssembly build is single-threaded.

Limits

• The WebAssembly build can use at most 4 GB of memory (65,536 pages of 64 KB).
• Only index construction has an asynchronous variant, FuzzyIndex.fromAsync().