perf(NODE-5955): use pooled memory when possible

[nbbeeken]

Description

What is changing?

Use allocUnsafe which uses Node.js' pool of pre-allocated memory and does not santize it before returning it to the caller. We can use it in cases where we overwrite the entire returned Buffer to improve the performance of creating small memory allocations for ObjectIds and Decimal128 instances.

Is there new documentation needed for these changes?

No.

What is the motivation for this change?

Using the memory pool is faster than creating a new allocation.

Release Highlight

Use allocUnsafe for ObjectIds and Decimal128

For small allocations Node.js performance can be improved by using pre-allocated pooled memory. ObjectIds and Decimal128 instance will now use allocUnsafe on Node.js.

Double check the following

• Ran npm run check:lint script
• Self-review completed using the steps outlined here
• PR title follows the correct format: type(NODE-xxxx)[!]: description
  • Example: feat(NODE-1234)!: rewriting everything in coffeescript
• Changes are covered by tests
• New TODOs have a related JIRA ticket

[nbbeeken]

Node.js v20.11.0
OS: linux
CPUs: 8
Arch: x64
RAM: 33.105743872 GB

test single ObjectId
bson_deserialize_current_main x 2,735,078 ops/sec ±0.11% (395 runs sampled)
bson_deserialize_pr x 3,568,042 ops/sec ±0.46% (394 runs sampled)
bson_serialize_current_main x 2,260,330 ops/sec ±0.16% (393 runs sampled)
bson_serialize_pr x 2,227,234 ops/sec ±0.14% (394 runs sampled)
Fastest to slowest is:
- serialize on main is faster by 1.47%
- deserialize on pr is faster by 26.43%

test 1000 ObjectId
bson_deserialize_current_main x 3,987 ops/sec ±0.09% (395 runs sampled)
bson_deserialize_pr x 5,369 ops/sec ±0.17% (393 runs sampled)
bson_serialize_current_main x 8,484 ops/sec ±0.29% (391 runs sampled)
bson_serialize_pr x 8,655 ops/sec ±0.28% (395 runs sampled)
Fastest to slowest is:
- serialize on pr is faster by 1.99%
- deserialize on pr is faster by 29.54%

test single Decimal128
bson_deserialize_current_main x 1,834,433 ops/sec ±0.11% (393 runs sampled)
bson_deserialize_pr x 2,303,206 ops/sec ±0.09% (394 runs sampled)
bson_serialize_current_main x 1,995,501 ops/sec ±0.13% (394 runs sampled)
bson_serialize_pr x 1,987,482 ops/sec ±0.43% (394 runs sampled)
Fastest to slowest is:
- serialize on main is faster by 0.40%
- deserialize on pr is faster by 22.66%

test 1000 Decimal128
bson_deserialize_current_main x 2,396 ops/sec ±0.08% (394 runs sampled)
bson_deserialize_pr x 3,190 ops/sec ±0.09% (395 runs sampled)
bson_serialize_current_main x 6,776 ops/sec ±0.40% (393 runs sampled)
bson_serialize_pr x 7,085 ops/sec ±0.31% (393 runs sampled)
Fastest to slowest is:
- serialize on pr is faster by 4.45%
- deserialize on pr is faster by 28.41%

Granular benchmarks

[billouboq]

Nice improvement, didn't think it would be that much 👍

[nathan-knight]

Hey @nbbeeken, I just wanted to say that this appears to have been a massive boost in performance for my use case so thank you for this!