feat: expose pool aggregate statistics

[mnutt]

This relates to...

In achieving parity with the node.js agent, I'd like to be able to get aggregate statistics about requests and connections in undici. This is possible today, but requires using private symbols.

Addresses #1254, relates to #693 -- I just now noticed that one, so will review it to see if there's any feedback that can be applied here.

Rationale

I currently use the node.js agent and send its stats to prometheus; I'd like to send the same ones for undici to be able to instrument the switchover and ensure everything is working properly.

Changes

• Expose a stats getter on pools.

Features

• Access aggregate statistics about undici Pools via a stats object.

Bug Fixes

N/A

Breaking Changes and Deprecations

N/A

Status

KEY: S = Skipped, x = complete

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• [S] Benchmarked (optional)
  • I didn't benchmark it because the new methods aren't in the normal request path and are only intended to be called periodically; eyeballing it I believe it should scale roughly O(n) with number of active requests.
• Documented
• [] Review ready
• In review
• Merge ready

[codecov-commenter]

Codecov Report

Merging #1255 (48d99d8) into main (662fd82) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main    #1255      +/-   ##
==========================================
+ Coverage   94.08%   94.10%   +0.02%     
==========================================
  Files          43       44       +1     
  Lines        4071     4086      +15     
==========================================
+ Hits         3830     3845      +15     
  Misses        241      241              

Impacted Files	Coverage Δ
lib/core/symbols.js	100.00% <ø> (ø)
lib/pool-base.js	97.74% <100.00%> (+0.08%)	⬆️
lib/pool-stats.js	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 662fd82...48d99d8. Read the comment docs.

[ronag]

Could we provide more info through diagnostic channels so that these stats can be built in user space?

[mnutt]

That's an interesting idea, I guess in metrics terms these are more like gauges, while diagnostics channels are counters. I think it's possible that one could use diagnostic channels counters to emulate some of these by doing some bookkeeping, though I think it might require quite a few more diagnostic channels since some of these are overlapping?

[ronag]

few more diagnostic channels since some of these are overlapping?

I'm happy to add more channels!

[mcollina]

lgtm

[ronag]

I still think for best performance and reactive API adding more diagnostic channels and implementing this on top of that is a better idea.

[ronag]

@mnutt Do you think you could add something similar to Client and maybe Agent?