perf(internal): add small buffer tiers and route put by capacity

[darakanoit]

Reason for This PR

Follow-up to roadrunner-server/roadrunner#2395 and #220. Problem statement: roadrunner-server/roadrunner#2401. Two problems in the tiered buffer pool in internal/bpool.go, which since #220 serves both the send and the receive path.

The smallest tier is 1 MB. Every payload up to 1 MB, and every options read of 4 to 40 bytes, takes a 1 MB buffer while in flight. Resident memory grows with the number of workers, not with the size of the data. Measured with N goroutines each holding a buffer for a 200-byte payload, which is what N workers do while their responses are in flight:

Workers in receive	master	this PR
8	8.4 MB	0.03 MB
64	67.2 MB	0.3 MB
256	268.7 MB	1.3 MB

The memory survives one GC in the victim cache, so under steady load it is permanent. A pool miss also zeroes the whole tier: 37 µs for 1 MB on an M3.

put routes by the requested size, not by the buffer's capacity. A buffer above 10 MB is allocated with make in get, but put stores it in the 10 MB tier. From then on it is handed out for every 5 to 10 MB request and never released while the pool is warm. The default max_request_size in the http plugin is 1 GB, so a single large upload or response reaches this path. Reproduction on master: get(50 << 20), put(50 << 20, buf), then get(6 << 20) returns a buffer with cap == 50 MB.

Description of Changes

• Add 4 KB, 16 KB, 64 KB and 256 KB tiers below the existing 1 MB, 5 MB and 10 MB. A request is served from the smallest tier that fits it.
• Hold the tiers in an array indexed in parallel with a sorted tierSizes array instead of a sync.Map keyed by size. get is a scan over seven uint32 comparisons.
• put(data) routes by cap(*data) and returns a buffer only to the tier whose size equals its capacity. Anything else, including an oversized make result, is dropped. This is the same rule as putDataBufferChunk in x/net/http2 and BinaryTieredBufferPool.Put in grpc-go, which both key on the buffer rather than on the request.
• Tests moved from receive_test.go into bpool_test.go: every tier boundary, an oversized buffer never entering the 10 MB tier, a foreign slice never entering a tier, a 4 KB buffer returning to the 4 KB tier. The last three pin GOMAXPROCS(1) so the sync.Pool private slot makes the outcome deterministic. Ported onto master they fail on every run without the race detector.
• Wire format, public API and the Preallocate contract are unchanged. Only internal is touched.

Benchmarks, BenchmarkReceivePath and BenchmarkSendPath, master and this branch interleaved over 5 rounds, benchstat over 10 samples each, M3:

Path	master	this PR	Δ
ReceivePath/1KB	95.3 ns	71.5 ns	−25% (p=0.000)
ReceivePath/64KB	2.02 µs	1.73 µs	−14% (p=0.023)
ReceivePath/1MB	37.8 µs	37.5 µs	~
SendPath/1KB	115 ns	91.6 ns	−20% (p=0.000)
SendPath/64KB	4.32 µs	4.29 µs	~
SendPath/1MB	56.3 µs	55.3 µs	~

Allocations per op are identical in every case. The gain on small frames is the sync.Map lookup and the two type assertions that are gone from get and put.

closes: roadrunner-server/roadrunner#2401

License Acceptance

By submitting this pull request, I confirm that my contribution is made under
the terms of the MIT license.

PR Checklist

• All commits in this PR are signed (git commit -s) or (git commit -S).
• The reason for this PR is clearly provided (issue no. or explanation).
• The description of changes is clear and encompassing.
• Any required documentation changes (code and docs) are included in this PR.
• Any user-facing changes are mentioned in CHANGELOG.md.
• All added/changed functionality is tested.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.37%. Comparing base (d1d60f5) to head (18b18f1).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #221      +/-   ##
==========================================
- Coverage   78.56%   78.37%   -0.19%     
==========================================
  Files           9        9              
  Lines         695      689       -6     
==========================================
- Hits          546      540       -6     
  Misses        149      149              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[rustatian]

Hey @darakanoit 👋🏻
I'd suggest you change the base branch and merge this PR into your PR #222, since they both update the same files.

[darakanoit]

Makes sense, thanks. Closing this one; #222 already carries this change and its description now covers both.