perf(NODE-4726): bench many strings in large batch in parallel

[nbbeeken]

Description

What is changing?

There's still some cruft in here, but its intentional in case someone else want to pull it down and check my work.
The issue reported does reproduce but only when we try to simulate parallelism. I don't think it is the parallelism that effects our BSON library it just offers enough load to be able to measure the difference.

The goal here isn't to fix anything but some hypothesis, some ruled out:

• getValidatedString moved the buffer.toString to a helper, undoing this did not demonstrate a change
• removing all the UTF8 key validation logic may have introduced overhead of allocating a Set, checking has etc.
  • removing it did not appear to have an impact
• The analysis that I would recommend in a follow up is taking a close look at globals / new structures added to deserializeObject that don't exist in v1. Some known ones are:
  • DBRef shape checking
  • DataView for double parsing
  • UTF8 stuff ^

What is the motivation for this change?

This benchmark is meant to demonstrate the perf difference so we can follow up with fixes that can be proven to be effective.

Double check the following

• Ran npm run lint script
• Self-review completed using the steps outlined here
• PR title follows the correct format: <type>(NODE-xxxx)<!>: <description>
• Changes are covered by tests
• New TODOs have a related JIRA ticket

[dariakp]

This PR needs a rebase and a fresh re-install of dependencies.

Let's make sure that we capture the suggested approach in the description in a comment on the follow up ticket (either NODE-4283 or a new one that captures a more specific perf improvement)

[dariakp]

Just one minor thing here

[nbbeeken]

This seems to have added a lot to the lock file I'm happy to make this a "must install before launching benchmaks" style dep, much like the other versions of the BSON lib are installed with --no-save

[dariakp]

no strong preferences here since it's a dev dependency; though if we're intending to run perf tests in the pipeline, may be worth keeping just to avoid surprise failures

[baileympearson]

What is the purpose of the setTimeout here? And why did you choose a setTimeout of 20ms specifically?

[nbbeeken]

It manufactures the delay that the driver does when we promise.all many many finds, 20ms is arbitrary, kept it low so the bench runs in a reasonable amount of time