perf(pm): fan out registry http clients

[elrrrrrrr]

Summary

Review-sized split from #3028 / #2948, stacked after #3035.

Fans registry manifest requests across a small native reqwest client pool:

• keeps wasm on a single browser-backed client;
• uses four native clients with independent keep-alive pools;
• round-robins get_client() with a relaxed atomic counter;
• preserves proxy/DNS/client-builder error handling.

This isolates the GHA/npmjs Cloudflare single-pool bottleneck experiment into one small PR.

Size

• Diff against perf(pm): add mock manifest provider #3035: 1 file changed, 38 insertions(+), 4 deletions(-)

Validation

• cargo fmt
• cargo check -p utoo-ruborist
• cargo clippy --all-targets -- -D warnings --no-deps

pack-napi still warns locally because next.js is a symlink in this worktree; clippy exits successfully.

[gemini-code-assist]

Code Review

This pull request introduces a pool of HTTP clients for non-WASM targets to improve request distribution across multiple IP addresses, utilizing a round-robin selection mechanism via an atomic counter. For WASM targets, the single-client implementation is maintained. Feedback was provided to include a compile-time assertion ensuring the client pool size is greater than zero, which prevents a potential division-by-zero panic during client retrieval.

[gemini-code-assist]

While CLIENT_POOL_SIZE is currently a constant set to 4, it would be safer to add a compile-time assertion to ensure it is never accidentally set to 0. If it were 0, the HTTP_CLIENTS initialization would succeed with an empty Vec, but get_client() would subsequently panic due to a division by zero in the modulo operation on line 118. This approach ensures the bug is caught at compile-time, adhering to the principle that panics should be treated as bugs to be fixed.

Suggested change

	const CLIENT_POOL_SIZE: usize = 4;
	const CLIENT_POOL_SIZE: usize = 4;
	#[cfg(not(target_arch = "wasm32"))]
	const _: () = assert!(CLIENT_POOL_SIZE > 0, "CLIENT_POOL_SIZE must be at least 1");

References

1. Panics should be treated as unrecoverable bugs that need to be fixed, not as transient, recoverable errors. A compile-time assertion ensures the configuration is valid and prevents a runtime panic.

[github-actions]

📊 pm-bench-phases · 599034d · linux (ubuntu-latest)

Workflow run — ant-design

PMs: utoo (this branch) · utoo-npm (latest published) · bun (latest)

npmjs.org

p0_full_cold

PM	wall	±σ	user	sys	RSS	pgMinor
bun	12.24s	2.44s	10.28s	10.20s	578M	301.4K
utoo-next	20.59s	17.40s	10.99s	13.38s	951M	119.2K
utoo-npm	19.25s	12.51s	11.09s	13.67s	963M	121.0K
utoo	10.45s	2.26s	10.84s	13.23s	1.02G	117.7K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	24.0K	14.7K	1.19G	7M	1.86G	1.75G	1M
utoo-next	172.0K	103.6K	1.16G	6M	1.71G	1.70G	2M
utoo-npm	186.6K	123.3K	1.16G	6M	1.71G	1.70G	2M
utoo	153.7K	98.4K	1.16G	6M	1.71G	1.70G	2M

p1_resolve

PM	wall	±σ	user	sys	RSS	pgMinor
bun	4.38s	2.18s	3.86s	1.09s	472M	179.6K
utoo-next	4.96s	2.59s	5.52s	1.88s	611M	89.0K
utoo-npm	3.60s	0.15s	5.68s	2.21s	600M	80.0K
utoo	6.19s	3.01s	5.51s	1.92s	610M	80.2K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	15.0K	2.9K	202M	3M	107M	-	1M
utoo-next	57.5K	71.0K	200M	3M	7M	3M	2M
utoo-npm	86.4K	85.2K	200M	3M	7M	3M	2M
utoo	58.9K	72.1K	200M	3M	7M	3M	2M

p3_cold_install

PM	wall	±σ	user	sys	RSS	pgMinor
bun	6.67s	0.07s	6.37s	10.25s	537M	196.4K
utoo-next	8.45s	0.18s	5.09s	11.33s	469M	58.4K
utoo-npm	9.06s	1.25s	5.21s	11.46s	433M	57.4K
utoo	8.24s	0.32s	5.09s	11.23s	460M	59.2K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	7.5K	8.4K	1019M	4M	1.76G	1.76G	1M
utoo-next	120.1K	58.7K	990M	3M	1.70G	1.70G	2M
utoo-npm	130.5K	57.8K	990M	3M	1.70G	1.70G	2M
utoo	120.3K	62.8K	990M	3M	1.70G	1.70G	2M

p4_warm_link

PM	wall	±σ	user	sys	RSS	pgMinor
bun	3.28s	0.15s	0.19s	2.34s	135M	32.7K
utoo-next	2.20s	0.10s	0.49s	3.74s	78M	18.3K
utoo-npm	2.17s	0.04s	0.49s	3.68s	79M	18.1K
utoo	2.17s	0.07s	0.50s	3.74s	79M	18.9K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	403	23	5M	56K	1.91G	1.75G	1M
utoo-next	40.5K	18.7K	9K	11K	1.70G	1.70G	2M
utoo-npm	41.7K	19.2K	8K	5K	1.70G	1.70G	2M
utoo	43.1K	19.7K	8K	24K	1.71G	1.70G	2M

npmmirror.com: no output captured.

[elrrrrrrr]

Closing this temporary fine-grained draft split. We are replacing it with a smaller 9-PR review stack based on the source PRs.