fix(yieldable-parser): parseAsync sometimes blocks for a *long* time #36

[lukas8219]

Fixes #36

Diagnosis:

1. Run node --prof parse-text.js
2. With the generated file, run node --prof-process <file.name> >> benchmark.txt
3. Check the code path that is being executed the most.

I've encapsulated the possible aggressors into processOne, processTwo functions so the profiler could gather more precise information... and re-run the same diagnosis steps - which lead me to this benchmark results bench-indexOf.txt

It's pretty clear that the main aggressor was while loop that had synchronous indexOf operations, causing performance hits in long strings.

Solution:

Split the indexOf method into it's own generator-style - yielding at each N sized chunk.

Results:

Using the same script shared by the issue author, event-loop was unblocked. But there was some performance hit, expected due to generators yielding.

Using master code: 1.92s user 3.03s system 109% cpu 4.534 total

Using new code: 5.17s user 3.95s system 134% cpu 6.785 total

[lukas8219]

Updates:

Finally got some time to run it on a non-virtualized environment. Just ran the same test on GCP using Linux ebpf-playground 5.15.0-1060-gcp #68~20.04.1-Ubuntu SMP Wed May 1 14:35:27 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux

And as expected, performance wasn't a huge deal.

Noticed a small increase of 2 seconds between approaches.

@gireeshpunathil Would you consider this a NO-GO?

[lukas8219]

This one may require attention into the JSON info. I copied and pasted from the shared file in the issue