perf(nodejs): introduce pool into default rng

[puzpuzpuz]

• Significantly improves v4 performance for the default RNG by introducing batch generation

Benchmark master (bc9d4ed):

uuidv1() x 1,868,914 ops/sec ±0.26% (92 runs sampled)
uuidv1() fill existing array x 9,711,758 ops/sec ±0.16% (94 runs sampled)
uuidv4() x 412,330 ops/sec ±0.13% (98 runs sampled)
uuidv4() fill existing array x 513,285 ops/sec ±0.16% (93 runs sampled)
uuidv3() x 321,397 ops/sec ±0.57% (91 runs sampled)
uuidv5() x 301,932 ops/sec ±1.08% (90 runs sampled)
Fastest is uuidv1() fill existing array

Benchmark this branch (pool size 64):

uuidv1() x 1,800,216 ops/sec ±0.80% (96 runs sampled)
uuidv1() fill existing array x 9,679,788 ops/sec ±0.27% (95 runs sampled)
uuidv4() x 819,473 ops/sec ±0.37% (96 runs sampled)
uuidv4() fill existing array x 1,630,070 ops/sec ±0.66% (92 runs sampled)
uuidv3() x 305,889 ops/sec ±0.99% (89 runs sampled)
uuidv5() x 299,425 ops/sec ±0.82% (89 runs sampled)
Fastest is uuidv1() fill existing array

Benchmark this branch (pool size 256) - current version:

uuidv1() x 1,874,252 ops/sec ±0.29% (97 runs sampled)
uuidv1() fill existing array x 9,641,771 ops/sec ±0.48% (94 runs sampled)
uuidv4() x 1,230,123 ops/sec ±0.63% (93 runs sampled)
uuidv4() fill existing array x 4,377,290 ops/sec ±0.71% (98 runs sampled)
uuidv3() x 317,152 ops/sec ±0.76% (93 runs sampled)
uuidv5() x 308,891 ops/sec ±0.85% (93 runs sampled)
Fastest is uuidv1() fill existing array

[broofa]

FWIW, results on my laptop...

Before:

uuidv1() x 1,586,023 ops/sec ±0.73% (91 runs sampled)
uuidv1() fill existing array x 8,903,577 ops/sec ±1.07% (89 runs sampled)
uuidv4() x 387,462 ops/sec ±3.11% (88 runs sampled)
uuidv4() fill existing array x 534,144 ops/sec ±2.71% (94 runs sampled)

After:

uuidv1() x 1,590,931 ops/sec ±0.55% (92 runs sampled)
uuidv1() fill existing array x 8,926,106 ops/sec ±0.56% (91 runs sampled)
uuidv4() x 745,416 ops/sec ±1.83% (82 runs sampled)
uuidv4() fill existing array x 1,685,734 ops/sec ±0.44% (96 runs sampled)

[broofa]

Nice improvement. It looks like we can further significantly improve perf with the attached suggestions. (Mostly a result of increasing pool size from 64 -> 256. Other tweaks to slightly reduce code size / complexity.)

With the attached, I get the following benchmark results:

uuidv1() x 1,604,852 ops/sec ±0.55% (95 runs sampled)
uuidv1() fill existing array x 8,987,992 ops/sec ±0.22% (97 runs sampled)
uuidv4() x 1,126,995 ops/sec ±0.33% (93 runs sampled)
uuidv4() fill existing array x 4,448,781 ops/sec ±1.50% (93 runs sampled)

[puzpuzpuz]

@broofa I've addressed your comments and updated benchmark results in the PR description

[ctavan]

LGTM, thank you @puzpuzpuz 🙏 !

[broofa]

@ctavan 'Not sure why bundlewatch test is hanging here. I was able to get it to pass on my laptop so I'm merging, but I ran into some sort of "call stack exceeded" error initially when doing npm run bundlewatch. Ended up rm -r node_modules'ing and reinstalling to get it to work.

[broofa]

@puzpuzpuz Thanks for the help!

[puzpuzpuz]

@broofa @ctavan thanks for the quick and valuable reviews!

[ctavan]

@broofa bundlewatch needs a token which is not available in CI runs from forks, only from branches within this repo... so PRs from forks will never report bundlewatch results. I haven’t found a workaround so far.

[simlu]

Please consider the regression introduced with this change.

#525

Looking forward to having a good discussion