perf(pm): prioritize install clone pumping

[elrrrrrrr]

Summary

This is an independent p3 scheduler experiment on top of #2989.

The install scheduler currently pumps ready work in this order:

1. downloads
2. extracts
3. clones/materialize

When p3 has both extracted packages and ready clone placements, new extract workers can enter the global rayon queue before materialize workers. This PR keeps network prefetch first, but gives ready clone/materialize work a chance to enter rayon before starting more cache writes.

Hypothesis

• p3 may improve if clone/materialize critical-path work is less often queued behind extract/cache-write work.
• p4 should be neutral because warm link has no extract queue.
• If p3 regresses, it means keeping extract/cache writes fully ahead of materialize is better for this workload.

Validation

• cargo fmt
• cargo test -p utoo-pm install_scheduler
• cargo clippy -p utoo-pm --all-targets -- -D warnings --no-deps

Bench Result

GHA Linux npmjs run 26093648834:

phase	utoo wall	vCtx	iCtx
p3 cold install	6.22s	56.8K	36.5K
p4 warm link	1.93s	7.4K	4.4K

p4 is a no-op control for this PR because there is no extract queue in warm link, so its movement is GHA noise. p3 regressed versus #2989 latest npmjs reference (5.29s, 53.0K/32.9K).

Conclusion: reject for now. Pure pump-order priority does not look like the right p3 fix.

[gemini-code-assist]

Code Review

This pull request reorders the execution of pump_clones and pump_extracts within the SchedulerState::run loop in crates/pm/src/service/install_scheduler.rs. This change prioritizes network prefetch and materialization work before starting cache writes. There are no review comments to address, and I have no further feedback to provide.

[github-actions]

📊 pm-bench-phases · d374339 · 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	9.53s	0.18s	10.31s	10.53s	590M	291.1K
utoo-next	8.94s	1.14s	10.49s	12.81s	988M	119.8K
utoo-npm	9.40s	1.87s	10.83s	12.98s	1002M	124.3K
utoo	7.73s	0.08s	11.22s	11.59s	934M	147.9K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	19.6K	17.9K	1.19G	7M	1.84G	1.73G	1M
utoo-next	137.5K	98.5K	1.16G	5M	1.71G	1.70G	2M
utoo-npm	147.6K	106.4K	1.16G	6M	1.71G	1.70G	2M
utoo	77.0K	50.9K	1.16G	6M	1.71G	1.70G	2M

p1_resolve

PM	wall	±σ	user	sys	RSS	pgMinor
bun	2.58s	0.01s	3.89s	1.17s	498M	188.8K
utoo-next	3.30s	0.08s	5.48s	2.27s	613M	82.0K
utoo-npm	3.37s	0.03s	5.63s	2.27s	624M	80.4K
utoo	2.49s	0.04s	6.03s	1.79s	648M	123.7K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	13.9K	3.3K	202M	3M	107M	-	1M
utoo-next	82.0K	88.7K	200M	3M	7M	3M	2M
utoo-npm	83.0K	90.7K	200M	3M	7M	3M	2M
utoo	17.5K	20.3K	202M	3M	7M	3M	2M

p3_cold_install

PM	wall	±σ	user	sys	RSS	pgMinor
bun	6.80s	0.21s	6.33s	10.49s	570M	199.3K
utoo-next	8.22s	1.11s	5.08s	11.35s	442M	59.9K
utoo-npm	7.36s	1.71s	5.23s	11.25s	489M	62.1K
utoo	5.57s	0.12s	4.84s	9.64s	485M	62.0K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	7.2K	8.7K	1019M	5M	1.76G	1.76G	1M
utoo-next	123.0K	52.9K	990M	3M	1.70G	1.70G	3M
utoo-npm	115.2K	55.0K	990M	3M	1.70G	1.70G	3M
utoo	52.1K	34.7K	989M	2M	1.70G	1.70G	3M

p4_warm_link

PM	wall	±σ	user	sys	RSS	pgMinor
bun	3.23s	0.14s	0.20s	2.32s	134M	32.5K
utoo-next	2.48s	0.45s	0.49s	3.85s	79M	18.4K
utoo-npm	2.38s	0.12s	0.50s	3.78s	79M	18.7K
utoo	2.10s	0.13s	0.34s	3.35s	50M	11.3K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	390	22	5M	57K	1.91G	1.75G	1M
utoo-next	40.9K	17.6K	1K	4K	1.70G	1.70G	2M
utoo-npm	42.1K	19.2K	2K	5K	1.70G	1.70G	2M
utoo	7.6K	4.4K	3K	9K	1.71G	1.70G	2M

npmmirror.com: no output captured.