fs: improve promise based readFile performance for big files

[BridgeAR]

This significantly reduces the peak memory for the promise
based readFile operation by reusing a single memory chunk after
each read and strinigifying that chunk immediately.

Refs: https://github.com/nodejs/node/discussions/44239#discussioncomment-3428693

Signed-off-by: Ruben Bridgewater ruben@bridgewater.de

Benchmark with a few runs:

                                                              confidence improvement accuracy (*)    (**)   (***)
fs/readfile-promises.js concurrent=1 len=1024 duration=5                     -1.18 %       ±7.07%  ±9.67% ±13.12%
fs/readfile-promises.js concurrent=1 len=16777216 duration=5                 -2.11 %       ±3.34%  ±4.56%  ±6.20%
fs/readfile-promises.js concurrent=1 len=33554432 duration=5         ***     59.48 %       ±5.41%  ±7.45% ±10.25%
fs/readfile-promises.js concurrent=1 len=4194304 duration=5          ***      9.95 %       ±5.10%  ±6.93%  ±9.32%
fs/readfile-promises.js concurrent=1 len=524288 duration=5                    1.88 %       ±6.77%  ±9.22% ±12.45%
fs/readfile-promises.js concurrent=1 len=8388608 duration=5           **      6.18 %       ±4.19%  ±5.70%  ±7.69%
fs/readfile-promises.js concurrent=10 len=1024 duration=5                    -1.21 %       ±8.11% ±11.12% ±15.15%
fs/readfile-promises.js concurrent=10 len=16777216 duration=5        ***      9.72 %       ±5.07%  ±6.92%  ±9.36%
fs/readfile-promises.js concurrent=10 len=33554432 duration=5        ***     26.58 %       ±6.29%  ±8.56% ±11.56%
fs/readfile-promises.js concurrent=10 len=4194304 duration=5          **      9.66 %       ±5.88%  ±8.00% ±10.77%
fs/readfile-promises.js concurrent=10 len=524288 duration=5                   1.38 %       ±5.49%  ±7.47% ±10.07%
fs/readfile-promises.js concurrent=10 len=8388608 duration=5         ***     16.84 %       ±4.35%  ±5.93%  ±8.02%

Be aware that when doing many comparisons the risk of a false-positive
result increases. In this case, there are 12 comparisons, you can thus
expect the following amount of false-positive results:
  0.60 false positives, when considering a   5% risk acceptance (*, **, ***),
  0.12 false positives, when considering a   1% risk acceptance (**, ***),
  0.01 false positives, when considering a 0.1% risk acceptance (***)

[mcollina]

lgtm

[BridgeAR]

We might consider a similar fix for #41435. If I checked correct, it outperforms the regular readFile variant for big files.
The lower peak memory and chunked stringify operations is better than less C++/JS boundary crossings.

[mcollina]

Does this also impact "old" callback- based readFile?

[BridgeAR]

Does this also impact "old" callback- based readFile?

No. This is solely about the promise based readFile with encodings. The non encoding part is also not impacted.
The implementation of the callback-based readFile differs from the promise based one.

23:39:15                                                                               confidence improvement accuracy (*)    (**)   (***)
23:39:15 fs/readfile-promises.js concurrent=10 encoding='utf8' len=1024 duration=5              *     -9.10 %       ±8.50% ±11.31% ±14.73%
23:39:15 fs/readfile-promises.js concurrent=10 encoding='utf8' len=16777216 duration=5         **      7.96 %       ±5.27%  ±7.02%  ±9.16%
23:39:15 fs/readfile-promises.js concurrent=10 encoding='utf8' len=33554432 duration=5        ***     63.48 %       ±6.59%  ±8.81% ±11.56%
23:39:15 fs/readfile-promises.js concurrent=10 encoding='utf8' len=4194304 duration=5                  2.20 %       ±3.73%  ±4.97%  ±6.47%
23:39:15 fs/readfile-promises.js concurrent=10 encoding='utf8' len=524288 duration=5                  -0.89 %       ±2.44%  ±3.25%  ±4.23%
23:39:15 fs/readfile-promises.js concurrent=10 encoding='utf8' len=8388608 duration=5                  4.81 %       ±5.08%  ±6.77%  ±8.84%
23:39:15 fs/readfile-promises.js concurrent=1 encoding='utf8' len=1024 duration=5                     -0.63 %       ±2.87%  ±3.82%  ±4.98%
23:39:15 fs/readfile-promises.js concurrent=1 encoding='utf8' len=16777216 duration=5         ***     -9.93 %       ±4.71%  ±6.28%  ±8.21%
23:39:15 fs/readfile-promises.js concurrent=1 encoding='utf8' len=33554432 duration=5         ***     75.47 %      ±10.06% ±13.56% ±17.98%
23:39:15 fs/readfile-promises.js concurrent=1 encoding='utf8' len=4194304 duration=5          ***    -11.12 %       ±5.45%  ±7.30%  ±9.61%
23:39:15 fs/readfile-promises.js concurrent=1 encoding='utf8' len=524288 duration=5                   -2.80 %       ±5.00%  ±6.65%  ±8.67%
23:39:15 fs/readfile-promises.js concurrent=1 encoding='utf8' len=8388608 duration=5          ***      9.33 %       ±3.50%  ±4.66%  ±6.07%

I ran the benchmark on our machines and there seems to be a weird effect for some file sizes where there's a performance drop. I guess it has to do with how V8 handles some things internally.
Reads up to 512kb are not impacted (the one star is misleading) and big reads (all tests above 32mb) are always significantly faster. In-between there are some file sizes that profit and some where there's a ~10% performance hit.

[BridgeAR]

Benchmark: https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/1176/

[mcollina]

lgtm

[mcollina]

@BridgeAR are you planning to do the same for the sync and callback versions too?

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/47083/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/47093/

[BridgeAR]

are you planning to do the same for the sync and callback versions too?

@mcollina I suggest to merge this and I am looking into the other part afterwards.

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/47101/

[mcollina]

Yes, let's do it.

[nodejs-github-bot]

Commit Queue failed

- Loading data for nodejs/node/pull/44295
✔  Done loading data for nodejs/node/pull/44295
----------------------------------- PR info ------------------------------------
Title      fs: improve promise based readFile performance for big files (#44295)
Author     Ruben Bridgewater  (@BridgeAR)
Branch     BridgeAR:improve-fs-promises-readfile -> nodejs:main
Labels     fs
Commits    1
 - fs: improve promise based readFile performance for big files
Committers 1
 - Ruben Bridgewater 
PR-URL: https://github.com/nodejs/node/pull/44295
Reviewed-By: Matteo Collina 
Reviewed-By: James M Snell 
------------------------------ Generated metadata ------------------------------
PR-URL: https://github.com/nodejs/node/pull/44295
Reviewed-By: Matteo Collina 
Reviewed-By: James M Snell 
--------------------------------------------------------------------------------
   ⚠  Commits were pushed since the last review:
   ⚠  - fs: improve promise based readFile performance for big files
   ℹ  This PR was created on Fri, 19 Aug 2022 15:12:52 GMT
   ✔  Approvals: 2
   ✔  - Matteo Collina (@mcollina) (TSC): https://github.com/nodejs/node/pull/44295#pullrequestreview-1080779016
   ✔  - James M Snell (@jasnell) (TSC): https://github.com/nodejs/node/pull/44295#pullrequestreview-1087869731
   ✔  Last GitHub CI successful
   ℹ  Last Benchmark CI on 2022-08-20T10:00:08Z: https://ci.nodejs.org/view/Node.js%20benchmark/job/benchmark-node-micro-benchmarks/1176/
   ℹ  Last Full PR CI on 2022-10-06T13:18:58Z: https://ci.nodejs.org/job/node-test-pull-request/47101/
- Querying data for job/node-test-pull-request/47101/
   ✔  Last Jenkins CI successful
--------------------------------------------------------------------------------
   ✔  Aborted `git node land` session in /home/runner/work/node/node/.ncu

https://github.com/nodejs/node/actions/runs/3198926242

[mcollina]

lgtm

[nodejs-github-bot]

Landed in d3dd49f