perf(tarball): run parallel CAS writes on a dedicated rayon pool#12248
Merged
Conversation
The per-file CAS-write parallelism added in #12247 ran on rayon's global pool. But the install pipeline overlaps tarball extraction with linking each resolved package into `node_modules`, and the linker drives its per-package work through `rayon::join` / `par_iter` on that same global pool. When a batch of downloads finished at once (hundreds of tarballs entering extraction together), the extraction work queued ahead of the linker's jobs and stalled linking for seconds. Aligning the download/extract trace with the `imported` progress events on a ~1300-package fresh install showed the linker dropping to zero completions for ~1s right as an extraction surge landed, then grinding the rest out afterward — extraction had gotten faster, but it stuttered the concurrent linker, so the net win on the pipeline was lost. Route the parallel CAS writes through a dedicated rayon pool (sized to the core count; the work is CPU-bound SHA-512 + CAFS write) so an extraction burst can't monopolize the global pool the linker uses. The two phases now run concurrently without one starving the other: on the same fixture the linker no longer stalls (continuous completions through the extraction window) and the big-package extraction tail stays parallelized. Falls back to the global pool if the dedicated pool can't be built. --- Written by an agent (Claude Code, claude-opus-4-8).
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (1)
📜 Recent review details⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (2)
🧰 Additional context used📓 Path-based instructions (1)pacquet/**/*.rs📄 CodeRabbit inference engine (pacquet/AGENTS.md)
Files:
🧠 Learnings (5)📓 Common learnings📚 Learning: 2026-05-25T14:58:11.105ZApplied to files:
📚 Learning: 2026-05-20T19:40:55.051ZApplied to files:
📚 Learning: 2026-05-22T00:08:44.646ZApplied to files:
📚 Learning: 2026-05-20T23:07:58.444ZApplied to files:
🔇 Additional comments (2)
📝 WalkthroughWalkthroughThis PR adds a dedicated Rayon thread pool for CAS-write operations during tarball extraction. When the pending file count exceeds a threshold, writes execute on the dedicated pool; otherwise, small batches run serially. Pool creation failures gracefully fall back to existing behavior. ChangesDedicated CAS-write thread pool for tarball extraction
Sequence Diagram(No diagram needed for this change—the logic flow is simple and self-contained within a single function.) Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
Micro-Benchmark ResultsLinux |
Review Summary by QodoIsolate tarball CAS writes to dedicated rayon pool
WalkthroughsDescription• Isolate CAS-write parallelism to dedicated rayon pool • Prevents extraction bursts from starving linker on global pool • Improves concurrent pipeline performance during install • Falls back gracefully if dedicated pool creation fails Diagramflowchart LR
A["Tarball Extraction"] -->|CAS writes| B["Dedicated Pool"]
B -->|CPU-bound work| C["SHA-512 + CAFS write"]
D["Package Linker"] -->|per-package work| E["Global Pool"]
C -.->|no contention| E
File Changes1. pacquet/crates/tarball/src/lib.rs
|
Contributor
Integrated-Benchmark Report (Linux)Each scenario has pacquet rows (direct install) and pnpr rows (the same client through the pnpr install accelerator), so pnpr@HEAD vs pacquet@HEAD is the pnpr-vs-direct ratio. Cold-store scenarios wipe the client store between runs (warm server); hot-store scenarios keep it warm. The pacquet@HEAD rows feed the pacquet Bencher testbed; the pnpr@HEAD rows feed the pnpr testbed. Scenario: Isolated linker: fresh restore, cold cache + cold store
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 10.010998551799998,
"stddev": 0.1481741587551117,
"median": 9.943700258499998,
"user": 3.6415678,
"system": 4.2730365799999985,
"min": 9.9156281455,
"max": 10.3964415155,
"times": [
10.1200656095,
9.941981845499999,
10.0172028455,
9.945418671499999,
9.940264384499999,
9.9334557685,
9.9351929365,
9.9643337955,
10.3964415155,
9.9156281455
]
},
{
"command": "pacquet@main",
"mean": 10.116023478499999,
"stddev": 0.16565828733731636,
"median": 10.0548988205,
"user": 3.6641863999999997,
"system": 4.31429418,
"min": 9.9790666715,
"max": 10.524537707499999,
"times": [
9.9790666715,
10.524537707499999,
10.035888182499999,
10.0326563465,
10.2610263905,
10.0018176475,
10.0739094585,
10.153635615499999,
10.0197867245,
10.077910040499999
]
},
{
"command": "pnpr@HEAD",
"mean": 5.079941864100001,
"stddev": 0.045950783414495176,
"median": 5.070211239500001,
"user": 2.8467627,
"system": 3.8858840800000003,
"min": 5.0307381035,
"max": 5.1800158675,
"times": [
5.0382895925,
5.0558838255000005,
5.1375636345,
5.1800158675,
5.072305210500001,
5.0681172685,
5.0307381035,
5.0824884815,
5.0529671855,
5.0810494715
]
},
{
"command": "pnpr@main",
"mean": 5.174951543400001,
"stddev": 0.08494575066894672,
"median": 5.146631060500001,
"user": 2.8768315,
"system": 3.93716988,
"min": 5.1066102685,
"max": 5.3719003455,
"times": [
5.1538591105,
5.1439517835,
5.2765226345,
5.184206731500001,
5.1345223155,
5.1066102685,
5.1181304115,
5.1493103375,
5.3719003455,
5.1105014955
]
}
]
}Scenario: Isolated linker: fresh restore, hot cache + hot store
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 0.6751936207800001,
"stddev": 0.030065317416254456,
"median": 0.6674702190800001,
"user": 0.36508996,
"system": 1.32183664,
"min": 0.6577981955800001,
"max": 0.7591978075800001,
"times": [
0.7591978075800001,
0.67317357058,
0.6689526105800001,
0.6657430695800001,
0.67228710558,
0.6577981955800001,
0.6588049625800001,
0.6589178185800001,
0.6659878275800001,
0.6710732395800001
]
},
{
"command": "pacquet@main",
"mean": 0.69696154008,
"stddev": 0.015316989658648009,
"median": 0.6962639395800001,
"user": 0.37917646,
"system": 1.3360063400000002,
"min": 0.67128961758,
"max": 0.7243295005800001,
"times": [
0.7143976535800001,
0.6782702435800001,
0.69220863458,
0.7243295005800001,
0.70155570158,
0.6961345935800001,
0.69639328558,
0.67128961758,
0.69597967758,
0.69905649258
]
},
{
"command": "pnpr@HEAD",
"mean": 0.81981227978,
"stddev": 0.10392973796117223,
"median": 0.78248091408,
"user": 0.39651006000000005,
"system": 1.33988594,
"min": 0.76692927958,
"max": 1.10929334158,
"times": [
0.8400093655800001,
0.7835602325800001,
0.77727588158,
0.76692927958,
1.10929334158,
0.8075006425800001,
0.78202173358,
0.78294009458,
0.77539973758,
0.7731924885800001
]
},
{
"command": "pnpr@main",
"mean": 0.81790815378,
"stddev": 0.08559372757680923,
"median": 0.78442232858,
"user": 0.39047136,
"system": 1.3384131400000001,
"min": 0.76408355258,
"max": 1.0307366245799998,
"times": [
0.90537900658,
1.0307366245799998,
0.7856570755800001,
0.7840149715800001,
0.76408355258,
0.8122181475800001,
0.7848296855800001,
0.7784758795800001,
0.76868660058,
0.76499999358
]
}
]
}Scenario: Isolated linker: fresh install, cold cache + cold store
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 5.37445151214,
"stddev": 0.049126776742596294,
"median": 5.39200366784,
"user": 3.8321015000000003,
"system": 3.28458396,
"min": 5.28219729634,
"max": 5.43675273634,
"times": [
5.28219729634,
5.41231844934,
5.40118171934,
5.43675273634,
5.3236248353399995,
5.32531298134,
5.36643891734,
5.41268085034,
5.39650274334,
5.38750459234
]
},
{
"command": "pacquet@main",
"mean": 5.4182296896399995,
"stddev": 0.05360788291160449,
"median": 5.41600819934,
"user": 3.8778683000000003,
"system": 3.34451916,
"min": 5.33400619134,
"max": 5.50172752834,
"times": [
5.33400619134,
5.50070352634,
5.4044781533399995,
5.38365102034,
5.37874210834,
5.43236534534,
5.50172752834,
5.4379621263399995,
5.42753824534,
5.38112265134
]
},
{
"command": "pnpr@HEAD",
"mean": 1.99874442574,
"stddev": 0.02968313808334842,
"median": 2.00023253984,
"user": 2.6341473,
"system": 3.1626518599999995,
"min": 1.96274379534,
"max": 2.0569664303399997,
"times": [
2.01862548234,
2.0101737473399997,
2.01670157134,
1.97441909234,
1.96485635534,
1.9794157733400002,
2.01325067734,
2.0569664303399997,
1.9902913323400002,
1.96274379534
]
},
{
"command": "pnpr@main",
"mean": 2.02294484104,
"stddev": 0.020230621638806925,
"median": 2.02198711334,
"user": 2.6870373,
"system": 3.2713739599999996,
"min": 1.98885453634,
"max": 2.0520475223399997,
"times": [
2.0252330433399997,
2.0196573563399998,
1.98885453634,
2.01234211834,
2.05191929234,
2.0520475223399997,
2.02431687034,
2.0136968503399997,
2.03813797234,
2.0032428483399998
]
}
]
}Scenario: Isolated linker: fresh install, hot cache + hot store
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 1.4439011771,
"stddev": 0.019959774908014545,
"median": 1.4440132891,
"user": 1.57225428,
"system": 1.7813421400000002,
"min": 1.4149171051,
"max": 1.4884554731000001,
"times": [
1.4268267931,
1.4507420321000002,
1.4511244151,
1.4479200051,
1.4357730391,
1.4301187171,
1.4149171051,
1.4884554731000001,
1.4530276181000001,
1.4401065731
]
},
{
"command": "pacquet@main",
"mean": 1.4725261376000003,
"stddev": 0.021160226831649045,
"median": 1.4678062896,
"user": 1.5804691800000001,
"system": 1.8215798399999996,
"min": 1.4546816851,
"max": 1.5239479721,
"times": [
1.4871367361,
1.4552896921,
1.5239479721,
1.4553874691000002,
1.4818059191000001,
1.4695403191,
1.4616385591,
1.4546816851,
1.4697607641000001,
1.4660722601
]
},
{
"command": "pnpr@HEAD",
"mean": 0.673489305,
"stddev": 0.03178121765781224,
"median": 0.6625335376,
"user": 0.3232257799999999,
"system": 1.26366944,
"min": 0.6547989090999999,
"max": 0.7625212731,
"times": [
0.6693393151,
0.6581831681,
0.6547989090999999,
0.7625212731,
0.6630288351,
0.6617710411,
0.6668957761,
0.6746662661,
0.6616502261,
0.6620382401
]
},
{
"command": "pnpr@main",
"mean": 0.6563374208,
"stddev": 0.007448383359137579,
"median": 0.6571438231,
"user": 0.31936138000000003,
"system": 1.25111124,
"min": 0.6450081851,
"max": 0.6695899541,
"times": [
0.6450081851,
0.6484303901,
0.6576026771,
0.6485384031,
0.6625730081,
0.6695899541,
0.6603231141,
0.6547050231,
0.6566849691,
0.6599184841
]
}
]
}Scenario: Isolated linker: fresh install, cold cache + hot storeResolution-only: cold packument cache (full re-resolve over the registry link) with a hot store (no tarball download), so this isolates pnpr offloading the client resolution to its warm server.
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 5.05599541454,
"stddev": 0.04340728444597576,
"median": 5.04051559344,
"user": 1.7090591399999997,
"system": 1.95851306,
"min": 5.02475625144,
"max": 5.17455418644,
"times": [
5.062395529440001,
5.02475625144,
5.03651120244,
5.06039393044,
5.17455418644,
5.04451998444,
5.03254392844,
5.03646766444,
5.05198066344,
5.035830804440001
]
},
{
"command": "pacquet@main",
"mean": 5.01019596164,
"stddev": 0.02089344345133252,
"median": 5.01239145744,
"user": 1.6955970399999998,
"system": 1.9470569599999998,
"min": 4.97931634744,
"max": 5.04773231144,
"times": [
5.04773231144,
5.01281057944,
4.99166816444,
5.01655851044,
5.03446882244,
5.01880644344,
5.01197233544,
4.97931634744,
4.99278060444,
4.99584549744
]
},
{
"command": "pnpr@HEAD",
"mean": 0.6478643085400001,
"stddev": 0.007928669086402353,
"median": 0.64523806044,
"user": 0.32489914000000003,
"system": 1.2348576599999999,
"min": 0.64076980944,
"max": 0.66348540744,
"times": [
0.66348540744,
0.64550863844,
0.64591157444,
0.64496748244,
0.64344670244,
0.64433552544,
0.64245501844,
0.64076980944,
0.64613455244,
0.66162837444
]
},
{
"command": "pnpr@main",
"mean": 0.65244192644,
"stddev": 0.011565644880580555,
"median": 0.65045053444,
"user": 0.31418273999999996,
"system": 1.2571509599999997,
"min": 0.6351119494399999,
"max": 0.67584896844,
"times": [
0.6444408914399999,
0.64635296044,
0.6351119494399999,
0.64554171244,
0.65177428644,
0.65045496744,
0.67584896844,
0.66412249644,
0.65044610144,
0.66032493044
]
}
]
} |
Contributor
Bencher Report
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Follow-up to #12247.
Problem
The per-file CAS-write parallelism added in #12247 ran on rayon's global pool. But the install pipeline overlaps tarball extraction with linking each resolved package into
node_modules, and the linker drives its per-package work throughrayon::join/par_iteron that same global pool. When a batch of downloads finishes at once (hundreds of tarballs entering extraction together), the extraction work queues ahead of the linker's jobs and stalls linking.Aligning the download/extract trace with the
importedprogress events on a ~1300-package fresh install, the linker dropped to zero completions for ~1s right as an extraction surge landed, then ground out the rest afterward:Extraction had gotten faster, but it stuttered the concurrent linker — so the pipeline win was lost.
Fix
Route the parallel CAS writes through a dedicated rayon pool (sized to the core count; the work is CPU-bound SHA-512 + CAFS write) so an extraction burst can't monopolize the global pool the linker uses. The two phases now run concurrently without starving each other — on the same fixture the linker no longer stalls (continuous completions through the extraction window, cumulative links by sec 6: 592 vs 537), the big-package extraction tail stays parallelized, and total nudged 17s→16s. Falls back to the global pool if the dedicated pool can't be built.
(Note: absolute time on macOS is dominated by an APFS metadata-op floor in the link phase that's identical with and without this change; the benefit of removing the stall shows most where linking parallelizes, i.e. Linux.)
Tests
All 54
pacquet-tarballtests pass; clippy clean. Output is unchanged — this only moves which thread pool the writes run on.Written by an agent (Claude Code, claude-opus-4-8).
Summary by CodeRabbit