perf(package-manager): parallelize cold-batch slot linking#12249
Merged
Conversation
|
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 (3)
📜 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). (7)
🧰 Additional context used📓 Path-based instructions (1)pacquet/**/*.rs📄 CodeRabbit inference engine (pacquet/AGENTS.md)
Files:
🧠 Learnings (25)📓 Common learnings📚 Learning: 2026-05-23T09:14:43.635ZApplied to files:
📚 Learning: 2026-06-04T14:40:29.444ZApplied to files:
📚 Learning: 2026-05-24T21:11:04.272ZApplied to files:
📚 Learning: 2026-05-23T09:14:43.635ZApplied to files:
📚 Learning: 2026-05-24T21:11:04.272ZApplied to files:
📚 Learning: 2026-05-29T18:03:15.372ZApplied to files:
📚 Learning: 2026-05-24T16:07:54.784ZApplied to files:
📚 Learning: 2026-05-20T23:23:43.636ZApplied to files:
📚 Learning: 2026-06-02T13:18:30.659ZApplied to files:
📚 Learning: 2026-05-19T19:23:00.981ZApplied 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:
📚 Learning: 2026-06-06T18:58:37.156ZApplied to files:
📚 Learning: 2026-05-23T16:55:36.507ZApplied to files:
📚 Learning: 2026-05-20T21:18:56.391ZApplied to files:
📚 Learning: 2026-05-29T18:03:15.372ZApplied to files:
📚 Learning: 2026-05-29T18:03:15.372ZApplied to files:
📚 Learning: 2026-05-29T18:03:15.372ZApplied to files:
📚 Learning: 2026-05-25T14:58:11.105ZApplied to files:
📚 Learning: 2026-06-04T20:24:32.096ZApplied to files:
📚 Learning: 2026-05-20T01:42:44.958ZApplied to files:
📚 Learning: 2026-05-04T17:01:30.322ZApplied to files:
📚 Learning: 2026-06-04T06:04:05.107ZApplied to files:
🔇 Additional comments (7)
📝 WalkthroughWalkthrough
ChangesCold-batch slot linking deferral
Sequence Diagram(s)sequenceDiagram
participant CreateVirtualStore
participant ColdInstall as InstallPackageBySnapshot<br/>(defer_link=true)
participant ColdFetch as Cold fetch/cache
participant ColdLinkPass as Cold link pass<br/>(Rayon parallel)
participant CreateVirtualDir as CreateVirtualDirBySnapshot
CreateVirtualStore->>ColdInstall: install with defer_link=true
ColdInstall->>ColdFetch: fetch & extract, skip slot
ColdFetch-->>CreateVirtualStore: return cas_paths
CreateVirtualStore->>ColdLinkPass: parallel slot linking
ColdLinkPass->>CreateVirtualDir: materialize slots (isolated)
CreateVirtualDir-->>ColdLinkPass: slot populated
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 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 |
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": 10.212459317440002,
"stddev": 0.16210709804143753,
"median": 10.13593329954,
"user": 3.2735745799999996,
"system": 3.2884268399999996,
"min": 10.08858325904,
"max": 10.56932165204,
"times": [
10.37342097604,
10.34087125404,
10.15279356304,
10.11900044504,
10.08858325904,
10.09103396604,
10.17843960504,
10.09205541804,
10.11907303604,
10.56932165204
]
},
{
"command": "pacquet@main",
"mean": 9.90931747134,
"stddev": 0.1425936664770252,
"median": 9.86791610454,
"user": 3.54759868,
"system": 4.300904539999999,
"min": 9.84366085404,
"max": 10.31248548004,
"times": [
9.88692714604,
9.84366085404,
9.84369774904,
9.87077879704,
9.85985611504,
9.86505341204,
9.84910738104,
10.31248548004,
9.89059695904,
9.87101082004
]
},
{
"command": "pnpr@HEAD",
"mean": 5.366586183940001,
"stddev": 0.15476657800863963,
"median": 5.31164598704,
"user": 2.4440086799999996,
"system": 2.96750124,
"min": 5.26085097104,
"max": 5.78930338904,
"times": [
5.78930338904,
5.29576767704,
5.42573014504,
5.29418664704,
5.26085097104,
5.30205793204,
5.31528652304,
5.33497270104,
5.33970040304,
5.30800545104
]
},
{
"command": "pnpr@main",
"mean": 5.126716651440001,
"stddev": 0.14086727968297946,
"median": 5.08508243504,
"user": 2.7235409799999992,
"system": 3.9439124399999996,
"min": 5.02323063504,
"max": 5.50713962404,
"times": [
5.12610444904,
5.04106670804,
5.13302090604,
5.50713962404,
5.02323063504,
5.05289407004,
5.06209018304,
5.15685198504,
5.05669326704,
5.10807468704
]
}
]
}Scenario: Isolated linker: fresh restore, hot cache + hot store
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 0.6492787357600001,
"stddev": 0.01100749143187808,
"median": 0.6471822676600001,
"user": 0.37601122,
"system": 1.2943588799999999,
"min": 0.63189176266,
"max": 0.6665447226600001,
"times": [
0.6665447226600001,
0.6544696776600001,
0.64766798966,
0.66288860266,
0.64507346966,
0.63189176266,
0.6384632786600001,
0.6577256676600001,
0.64669654566,
0.64136564066
]
},
{
"command": "pacquet@main",
"mean": 0.70164067616,
"stddev": 0.0330050551836713,
"median": 0.69087077916,
"user": 0.39105581999999994,
"system": 1.3229494800000001,
"min": 0.67355131566,
"max": 0.78715875566,
"times": [
0.72544138766,
0.67355131566,
0.68766836966,
0.6847879946600001,
0.68989504266,
0.78715875566,
0.69413707566,
0.69184651566,
0.69925473266,
0.6826655716600001
]
},
{
"command": "pnpr@HEAD",
"mean": 0.7979088221599999,
"stddev": 0.06305522350005904,
"median": 0.76793710066,
"user": 0.39382391999999994,
"system": 1.3141035799999998,
"min": 0.7605529896600001,
"max": 0.96459612366,
"times": [
0.80702072666,
0.7653408686600001,
0.96459612366,
0.83267479566,
0.78430876766,
0.7674035276600001,
0.76789384766,
0.76131622066,
0.7605529896600001,
0.7679803536600001
]
},
{
"command": "pnpr@main",
"mean": 0.7816471971600001,
"stddev": 0.016306291826892586,
"median": 0.77500490716,
"user": 0.41015262,
"system": 1.3126817799999997,
"min": 0.76078944666,
"max": 0.81164446966,
"times": [
0.78205740566,
0.79204231666,
0.81164446966,
0.7733678636600001,
0.80514638266,
0.76078944666,
0.7766419506600001,
0.77102867566,
0.7713977416600001,
0.7723557186600001
]
}
]
}Scenario: Isolated linker: fresh install, cold cache + cold store
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 5.35499403886,
"stddev": 0.04771691585975282,
"median": 5.361685317659999,
"user": 3.71947504,
"system": 3.3281736399999993,
"min": 5.28880688966,
"max": 5.43715723466,
"times": [
5.36599188766,
5.33051845766,
5.40323957366,
5.38867771566,
5.357378747659999,
5.36786015566,
5.30072383966,
5.30958588666,
5.43715723466,
5.28880688966
]
},
{
"command": "pacquet@main",
"mean": 5.3684357382600005,
"stddev": 0.049909525486640134,
"median": 5.36137162216,
"user": 3.7881255400000002,
"system": 3.33353514,
"min": 5.32693433366,
"max": 5.50098405266,
"times": [
5.36585538166,
5.35931089666,
5.33290394866,
5.50098405266,
5.37734052566,
5.37593488666,
5.34754359266,
5.32693433366,
5.36343234766,
5.33411741666
]
},
{
"command": "pnpr@HEAD",
"mean": 1.7724795680600003,
"stddev": 0.028164753367486778,
"median": 1.7811691221600001,
"user": 2.4031990399999996,
"system": 2.7506211399999994,
"min": 1.7298924416600001,
"max": 1.8108298856600002,
"times": [
1.7622365476600002,
1.79691764666,
1.7943162006600002,
1.7500980326600002,
1.7772334476600002,
1.7304997316600002,
1.7876669496600002,
1.78510479666,
1.7298924416600001,
1.8108298856600002
]
},
{
"command": "pnpr@main",
"mean": 2.0293010547600003,
"stddev": 0.028902669216662114,
"median": 2.0282530466599997,
"user": 2.49159284,
"system": 3.2863611400000003,
"min": 1.98934768566,
"max": 2.09044191366,
"times": [
2.00674834366,
2.03331789166,
1.9992147856600002,
2.02318820166,
2.0518348996599998,
2.03991600266,
2.0225736096599998,
2.0364272136599997,
2.09044191366,
1.98934768566
]
}
]
}Scenario: Isolated linker: fresh install, hot cache + hot store
BENCHMARK_REPORT.json{
"results": [
{
"command": "pacquet@HEAD",
"mean": 1.41480123434,
"stddev": 0.022782186949906365,
"median": 1.42414865764,
"user": 1.5848426799999997,
"system": 1.7562596600000002,
"min": 1.3614215491400001,
"max": 1.43196247514,
"times": [
1.4299578831400002,
1.42227660014,
1.43196247514,
1.42370714414,
1.40856440414,
1.3614215491400001,
1.38922278914,
1.42459017114,
1.42525408314,
1.4310552441400002
]
},
{
"command": "pacquet@main",
"mean": 1.41777751304,
"stddev": 0.022547579981766656,
"median": 1.42197570014,
"user": 1.5929027799999997,
"system": 1.7660905599999999,
"min": 1.38559587314,
"max": 1.4628841861400002,
"times": [
1.4268732191400002,
1.42202540414,
1.4628841861400002,
1.3973995031400002,
1.4234702621400002,
1.4219259961400001,
1.43618972514,
1.3977309901400001,
1.38559587314,
1.40367997114
]
},
{
"command": "pnpr@HEAD",
"mean": 0.6815159390399999,
"stddev": 0.07013368996278802,
"median": 0.66236200614,
"user": 0.34128188,
"system": 1.2554596600000003,
"min": 0.64459466914,
"max": 0.87976921814,
"times": [
0.66889696314,
0.65113203514,
0.64459466914,
0.87976921814,
0.66099548614,
0.65200971714,
0.65882153714,
0.66372852614,
0.66966949214,
0.66554174614
]
},
{
"command": "pnpr@main",
"mean": 0.69488855614,
"stddev": 0.056728181033914624,
"median": 0.66980846864,
"user": 0.35591158,
"system": 1.27355596,
"min": 0.66132400914,
"max": 0.81021430614,
"times": [
0.66132400914,
0.66970166414,
0.6659152511399999,
0.66182243214,
0.67061199314,
0.81021430614,
0.66824998714,
0.67752592914,
0.66991527314,
0.7936047161399999
]
}
]
}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": 5.00716459532,
"stddev": 0.03560271028702763,
"median": 5.00775816462,
"user": 1.7942784599999997,
"system": 1.8772191799999998,
"min": 4.94588573112,
"max": 5.07472811512,
"times": [
4.98167067012,
4.94588573112,
4.98737537712,
5.01213191212,
5.00974720512,
5.07472811512,
5.03183452312,
4.98506777612,
5.0057691241199995,
5.03743551912
]
},
{
"command": "pacquet@main",
"mean": 5.060775184919999,
"stddev": 0.040828993588567807,
"median": 5.05267591512,
"user": 1.87439896,
"system": 1.90634898,
"min": 5.010615705119999,
"max": 5.117722307119999,
"times": [
5.010615705119999,
5.03132799412,
5.032192215119999,
5.10200987812,
5.08271029012,
5.117722307119999,
5.109882735119999,
5.07315961512,
5.01637072412,
5.03176038512
]
},
{
"command": "pnpr@HEAD",
"mean": 0.66092987922,
"stddev": 0.007385852101453611,
"median": 0.6599400321200001,
"user": 0.3349162600000001,
"system": 1.2490763800000002,
"min": 0.6516556371200001,
"max": 0.6730238931200001,
"times": [
0.6650987621200001,
0.6605767771200001,
0.6709843231200001,
0.6578063591200001,
0.6730238931200001,
0.6640203881200001,
0.6550232021200001,
0.6593032871200001,
0.6518061631200001,
0.6516556371200001
]
},
{
"command": "pnpr@main",
"mean": 0.67127406102,
"stddev": 0.012927603776531502,
"median": 0.6715642311200001,
"user": 0.34681625999999993,
"system": 1.2581707800000002,
"min": 0.6502537261200001,
"max": 0.69708442112,
"times": [
0.66445964812,
0.6502537261200001,
0.65618028112,
0.6722694221200001,
0.6808609881200001,
0.67532445612,
0.6707075651200001,
0.69708442112,
0.6747410621200001,
0.6708590401200001
]
}
]
} |
Contributor
Bencher Report091feeb to
9e80124
Compare
The cold batch in `CreateVirtualStore` ran each per-snapshot `InstallPackageBySnapshot` as a future inside one cooperative `try_join_all` task, and each future called the *blocking* `CreateVirtualDirBySnapshot::run` (a `rayon::join` over CAS import + symlink layout) directly. Because all the futures share a single task, a blocking call in one prevents the others from being polled — so the 1308 slot-link operations on a fresh install serialized one-at-a-time, each paying its own `rayon::join` dispatch. The warm batch already avoids this by linking every slot in a single `rayon` `par_iter`. This was invisible on warm/restore installs (everything goes through the warm batch) but dominated the cold path: a pnpr client install, whose fast offloaded resolution routes every package through the cold batch, spent ~14.5s in `CreateVirtualStore` where an equivalent local fresh install — which pre-fills the store during its slower resolution and so hits the warm batch — spent ~5-7s linking. Defer the slot link out of the per-snapshot download future (new `InstallPackageBySnapshot::defer_link`) and run all the cold links in one parallel `rayon` pass after the downloads, mirroring the warm batch. On a fresh ~1300-package install through a pnpr server this cut `CreateVirtualStore` from ~14.5s to ~9.5s and the end-to-end install from ~18.5s to ~13s — now faster than the equivalent non-pnpr install (~15.5s), which it should be since the pnpr client offloads resolution and downloads less metadata. The progress events are unchanged in kind and order (`fetched`/`found_in_store` still fire during download, `imported` still fires once the slot is linked). --- Written by an agent (Claude Code, claude-opus-4-8).
9e80124 to
7966e73
Compare
Review Summary by QodoParallelize cold-batch slot linking to improve install performance
WalkthroughsDescription• Parallelize cold-batch slot linking in rayon pass after downloads • Defer slot link from per-snapshot futures to avoid serialization • Add defer_link flag to InstallPackageBySnapshot struct • Reduces CreateVirtualStore time by ~34% on fresh installs Diagramflowchart LR
A["Cold batch downloads"] -->|defer_link: true| B["Deferred slot links"]
B -->|rayon par_iter| C["Parallel link pass"]
C -->|isolated only| D["Virtual store slots"]
E["Warm batch"] -->|rayon par_iter| D
style C fill:#90EE90
style B fill:#FFB6C1
File Changes1. pacquet/crates/package-manager/src/create_virtual_store.rs
|
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.
Problem
CreateVirtualStorehas two batches: a warm batch (packages already in the store) and a cold batch (packages it had to download). The warm batch links every slot in a singlerayonpar_iter— fast. The cold batch, however, ran each per-snapshotInstallPackageBySnapshotas a future inside one cooperativetry_join_alltask, and each future called the blockingCreateVirtualDirBySnapshot::run(arayon::joinover CAS import + symlink) directly.Because all those futures share a single task, a blocking call in one prevents the others from being polled — so on a fresh install the 1308 slot-links serialized one-at-a-time, each paying its own
rayon::joindispatch.This is invisible on warm/restore installs (everything is warm), but it dominates the cold path — which is exactly what a pnpr client install hits: its fast offloaded resolution means every package goes through the cold batch. Phase timing on a fresh ~1300-package install:
Same link work, ~2-3x slower, purely from the serialization.
Fix
Defer the slot link out of the per-snapshot download future (new
InstallPackageBySnapshot::defer_link) and run all the cold links in one parallelrayonpass after the downloads — mirroring the warm batch.Result
Fresh ~1300-package install through a pnpr server:
CreateVirtualStoreAnd end-to-end, pnpr now beats the equivalent non-pnpr install:
…which is the expected outcome, since the pnpr client offloads resolution and downloads ~half the metadata. Progress events are unchanged in kind and order (
fetched/found_in_storeduring download,importedonce the slot is linked).Tests
685
pacquet-package-manager+pacquet-clitests pass (incl. thepnpr_installintegration tests, which exercise the cold batch).cargo fmt+clippyclean.Related
Composes with #12248 (dedicated CAS-write pool), which removes a separate extraction↔link rayon-pool contention; both target the same materialize phase.
Written by an agent (Claude Code, claude-opus-4-8).
Summary by CodeRabbit
Refactor
Tests