rate-limiter-flexible

## Compare fixed window with rolling window **tl;dr** Fixed window algorithm used in [rate-limiter-flexible](https://github.com/animir/node-rate-limiter-flexible) **14x faster** on high traffic than the fastest Rolling window The same benchmarking setting for all tests: `bombardier -c 1000 -l -d 30s -r 2000 -t 1s http://127.0.0.1:3000/endpoint` * 1000 concurrent requests * test duration is 30 seconds * not more than 2000 req/sec 3 libraries from github: 1. this one with fixed window 2. https://github.com/peterkhayes/rolling-rate-limiter 3. https://github.com/tj/node-ratelimiter 4. https://github.com/fastest963/node-redis-rolling-limit There are [4 simple Express 4.x endpoints](https://github.com/animir/simple-express-endpoint-techtask/blob/test-rate-limiters/src/routes/pricing.js) limited by different libraries, which launched in `node:latest` and `redis:alpine` Docker containers by PM2 with 4 workers Docker images are recreated before each test. All limiters created with same rule: maximum 100 requests per 1 second Key for every request is randomly generated number from 0 to 10 ### rate-limiter-flexible ```text Statistics Avg Stdev Max Reqs/sec 1996.58 447.17 8002.31 Latency 10.99ms 18.50ms 406.22ms Latency Distribution 50% 6.39ms 75% 10.07ms 90% 21.44ms 95% 33.62ms 99% 74.38ms HTTP codes: 1xx - 0, 2xx - 30000, 3xx - 0, 4xx - 29934, 5xx - 0 others - 71 Errors: the server closed connection before returning the first response byte. Make sure the server returns 'Connection: close' response header before closing the connection - 71 ``` ### rolling-rate-limiter ```text Statistics Avg Stdev Max Reqs/sec 2002.28 1004.72 25535.21 Latency 281.73ms 2.01s 32.92s Latency Distribution 50% 16.52ms 75% 55.35ms 90% 148.76ms 95% 257.40ms 99% 8.74s HTTP codes: 1xx - 0, 2xx - 55206, 3xx - 0, 4xx - 1197, 5xx - 0 others - 3530 Errors: the server closed connection before returning the first response byte. Make sure the server returns 'Connection: close' response header before closing the connection - 3275 dial tcp 127.0.0.1:3000: connect: operation timed out - 175 dial tcp 127.0.0.1:3000: connect: connection reset by peer - 80 ``` ### ratelimiter ```text Statistics Avg Stdev Max Reqs/sec 2021.58 995.59 22045.06 Latency 219.30ms 1.81s 31.66s Latency Distribution 50% 15.54ms 75% 45.39ms 90% 129.52ms 95% 206.34ms 99% 8.34s HTTP codes: 1xx - 0, 2xx - 54628, 3xx - 0, 4xx - 1038, 5xx - 0 others - 4226 Errors: the server closed connection before returning the first response byte. Make sure the server returns 'Connection: close' response header before closing the connection - 3931 dial tcp 127.0.0.1:3000: connect: connection reset by peer - 150 dial tcp 127.0.0.1:3000: connect: operation timed out - 145 ``` ### redis-rolling-limit ```text Statistics Avg Stdev Max Reqs/sec 1975.33 944.00 25855.43 Latency 148.29ms 1.18s 29.00s Latency Distribution 50% 8.39ms 75% 26.26ms 90% 114.73ms 95% 214.40ms 99% 2.84s HTTP codes: 1xx - 0, 2xx - 56727, 3xx - 0, 4xx - 0, 5xx - 0 others - 2274 Errors: the server closed connection before returning the first response byte. Make sure the server returns 'Connection: close' response header before closing the connection - 2165 dial tcp 127.0.0.1:3000: connect: connection reset by peer - 70 dial tcp 127.0.0.1:3000: connect: operation timed out - 39 ``` ### Conclusion Average latency * `rate-limiter-flexible` 11ms * `rolling-rate-limiter` 282ms * `ratelimiter` 219ms * `redis-rolling-limit` 148ms It is obvious that fixed window is much-much faster