Use a single require-in-the-middle hook

[pgayvallet]

Summary

Doing investigations on Kibana startup time, we discovered that the usage of require-in-the-middle has some significant performance drawback for initial start / module load time.

After investigating, we discovered that the perf hit of the library is directly related to the number of hooks that are being defined (as the patching logic is executed once PER hook, surprisingly).

I will be opening an issue upstream to discuss about the problem, but for now, we'll be working around it from there.

This PR brings some perf improvements by changing the way we're using the lib, and using a single hook now delegating to the various patching methods instead of creating one hook per module patch.

Perf gain

Tests were only performed locally (we will confirm in other environments once the PR is merged), but the results are quite encouraging.

Without the PR

• Process startup: ~2000ms
• Preboot: ~5000ms
• Setup: ~6000ms
• Start: ~1000ms

With the PR

• Process startup: ~1650ms
• Preboot: ~4100ms
• Setup: ~5150ms
• Start: ~1000ms (no changes, expected given we don't load any module there)

So TLDR: ~18% in term of initial code loading time.

Thanks to @dgieselaar and @rudolf for the group effort on the investigation

[pgayvallet]

/ci

[pgayvallet]

Tests are failing because of elastic/require-in-the-middle#79, I guess we will have to fix that upstream before going forward

[pgayvallet]

/ci

[elasticmachine]

Pinging @elastic/kibana-core (Team:Core)

[kibana-ci]

💚 Build Succeeded

• Buildkite Build
• Commit: 7bf22f5

Metrics [docs]

✅ unchanged

History

• 💛 Build #197374 was flaky 280c501
• 💔 Build #196596 failed 50dd5b9

To update your PR or re-run it, just comment with:
@elasticmachine merge upstream