perf(pm): flush clone unblockers early by elrrrrrrr · Pull Request #2994 · utooland/utoo

elrrrrrrr · 2026-05-19T12:27:36Z

Summary

This is a p3/p4 scheduler experiment on top of #2989 after #2991 integration.

#2991 proved that parent clone completions are on the nested placement critical path. It handles that by prioritizing parent unblockers and putting them in a single-job batch. This PR keeps that early parent signal but avoids dedicating a whole worker batch to it:

parent unblockers are still selected first;
clone workers send CloneProgress immediately after a parent unblocker finishes;
CloneProgress completes that target and wakes nested children, but does not release the worker slot;
remaining leaf clones continue in the same worker and are reported by the final CloneBatch.

Hypothesis

p4 may improve over perf(pm): batch install clone completions #2989/perf(pm): prioritize clone unblockers #2991 if parent children can be queued early while leaf clones remain batched.
ctx may rise slightly because parent progress adds a message, but worker utilization should be better than single-job parent batches.
p3 should be neutral to slightly positive; cold extract/cache work is still the larger p3 cost.

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 26097117399:

phase	utoo wall	utoo vCtx/iCtx
p0_full_cold	7.40s	71.2K / 46.7K
p1_resolve	3.37s	15.8K / 20.2K
p3_cold_install	5.38s	54.4K / 35.0K
p4_warm_link	2.12s	7.4K / 4.4K

Reference #2989 integrated head (fd38cfff):

run	p3	p4
`26094747554`	5.30s, 52.1K / 35.0K	2.07s, 7.5K / 4.5K
`26095851789`	5.66s, 54.7K / 37.6K	2.16s, 7.8K / 4.6K

Conclusion: inconclusive / do not integrate for now. The early-flush implementation is logically reasonable, but it does not beat #2991 clearly enough to justify the extra completion state and worker-slot semantics.

gemini-code-assist

Code Review

This pull request introduces a mechanism to flush clone completions immediately for parent packages that unblock dependencies, rather than waiting for the entire batch to finish. This is implemented via a new CloneBatchJob struct and an OpDone::CloneProgress event. Review feedback suggests simplifying the CloneBatchJob struct by removing the redundant target field, as this information is already accessible through the ready field's specification.

gemini-code-assist · 2026-05-19T12:31:18Z

+struct CloneBatchJob {
+    target: PathBuf,
+    ready: ReadyClone,
+    flush_on_complete: bool,
+}


The target field in CloneBatchJob is redundant because it is already available via ready.spec.target. Removing it simplifies the struct and allows for reducing unnecessary clones in the scheduler loop.

struct CloneBatchJob { ready: ReadyClone, flush_on_complete: bool, }

gemini-code-assist · 2026-05-19T12:31:18Z

+                .filter_map(|batch_job| {
                    let result = clone_package_from_cache_indexed_sync(
-                        &job.spec.package.name,
-                        &job.spec.package.version,
-                        &job.spec.package.tarball_url,
-                        &job.cache.path,
-                        &job.spec.target,
-                        job.cache.index.as_ref(),
+                        &batch_job.ready.spec.package.name,
+                        &batch_job.ready.spec.package.version,
+                        &batch_job.ready.spec.package.tarball_url,
+                        &batch_job.ready.cache.path,
+                        &batch_job.ready.spec.target,
+                        batch_job.ready.cache.index.as_ref(),
                    )
                    .map_err(|e| format!("{e:#}"));
-                    (target, result)
+                    if batch_job.flush_on_complete {
+                        let _ = done_tx.send(OpDone::CloneProgress {
+                            target: batch_job.target,
+                            result,
+                        });
+                        None
+                    } else {
+                        Some((batch_job.target, result))
+                    }
                })


By moving batch_job.ready.spec.target directly into the result messages, we avoid using the redundant target field from the batch job struct.

.filter_map(|batch_job| { let result = clone_package_from_cache_indexed_sync( &batch_job.ready.spec.package.name, &batch_job.ready.spec.package.version, &batch_job.ready.spec.package.tarball_url, &batch_job.ready.cache.path, &batch_job.ready.spec.target, batch_job.ready.cache.index.as_ref(), ) .map_err(|e| format!("{e:#}")); let target = batch_job.ready.spec.target; if batch_job.flush_on_complete { let _ = done_tx.send(OpDone::CloneProgress { target, result, }); None } else { Some((target, result)) } })

gemini-code-assist · 2026-05-19T12:31:19Z

+        assert_eq!(batch[0].target, parent_target);
+        assert!(batch[0].flush_on_complete);
        assert_eq!(
-            state.clone_queue[0].spec.target,
+            batch[1].target,
            PathBuf::from("/tmp/project/node_modules/leaf")
        );
+        assert!(!batch[1].flush_on_complete);
+        assert!(state.clone_active.contains(&parent_target));
+        assert!(state.clone_queue.is_empty());


Updating the test assertions to reflect the removal of the redundant target field in CloneBatchJob.

Suggested change

assert_eq!(batch[0].target, parent_target);

assert!(batch[0].flush_on_complete);

assert_eq!(

state.clone_queue[0].spec.target,

batch[1].target,

PathBuf::from("/tmp/project/node_modules/leaf")

);

assert!(!batch[1].flush_on_complete);

assert!(state.clone_active.contains(&parent_target));

assert!(state.clone_queue.is_empty());

assert_eq!(batch[0].ready.spec.target, parent_target);

assert!(batch[0].flush_on_complete);

assert_eq!(

batch[1].ready.spec.target,

PathBuf::from("/tmp/project/node_modules/leaf")

);

assert!(!batch[1].flush_on_complete);

assert!(state.clone_active.contains(&parent_target));

assert!(state.clone_queue.is_empty());

github-actions · 2026-05-19T13:40:27Z

📊 pm-bench-phases · `e760c43` · 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	9.24s	0.30s	10.49s	10.36s	746M	323.7K
utoo-next	8.74s	1.43s	10.86s	12.88s	1001M	125.7K
utoo-npm	9.43s	1.93s	10.88s	12.58s	997M	122.4K
utoo	7.49s	0.16s	11.38s	11.37s	952M	141.6K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	15.5K	18.4K	1.19G	6M	1.88G	1.76G	1M
utoo-next	133.2K	94.4K	1.16G	5M	1.71G	1.70G	2M
utoo-npm	136.6K	92.1K	1.16G	5M	1.71G	1.70G	2M
utoo	67.6K	47.4K	1.16G	6M	1.71G	1.70G	2M

p1_resolve

PM	wall	±σ	user	sys	RSS	pgMinor
bun	1.97s	0.07s	4.08s	1.04s	532M	160.5K
utoo-next	3.09s	0.03s	5.54s	2.11s	621M	86.1K
utoo-npm	3.07s	0.01s	5.60s	2.11s	603M	83.3K
utoo	2.45s	0.06s	6.14s	1.75s	656M	127.6K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	8.8K	4.4K	202M	3M	107M	-	1M
utoo-next	71.6K	90.2K	200M	2M	7M	3M	2M
utoo-npm	71.9K	93.2K	200M	2M	7M	3M	2M
utoo	14.9K	20.6K	202M	3M	7M	3M	2M

p3_cold_install

PM	wall	±σ	user	sys	RSS	pgMinor
bun	6.75s	0.16s	6.38s	10.16s	608M	210.4K
utoo-next	7.24s	2.23s	5.03s	11.25s	455M	62.0K
utoo-npm	5.94s	0.28s	5.17s	11.14s	508M	59.4K
utoo	5.46s	0.26s	4.97s	9.66s	512M	57.1K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	3.8K	7.1K	1019M	3M	1.76G	1.76G	1M
utoo-next	106.3K	51.2K	989M	3M	1.70G	1.70G	2M
utoo-npm	102.3K	49.2K	989M	3M	1.70G	1.70G	2M
utoo	52.6K	34.0K	989M	2M	1.70G	1.70G	2M

p4_warm_link

PM	wall	±σ	user	sys	RSS	pgMinor
bun	3.40s	0.01s	0.20s	2.42s	134M	31.4K
utoo-next	2.54s	0.49s	0.50s	3.84s	78M	18.1K
utoo-npm	2.36s	0.10s	0.50s	3.83s	79M	18.2K
utoo	2.14s	0.04s	0.34s	3.31s	50M	11.1K

PM	vCtx	iCtx	netRX	netTX	cache	node_mod	lock
bun	286	25	3K	13K	1.87G	1.76G	1M
utoo-next	40.5K	17.3K	7K	10K	1.70G	1.70G	2M
utoo-npm	41.1K	18.4K	6K	7K	1.70G	1.70G	2M
utoo	7.5K	4.3K	6K	6K	1.71G	1.70G	2M

npmmirror.com: no output captured.

perf(pm): flush clone unblockers early

8b4e0eb

elrrrrrrr added A-Pkg Manager Area: Package Manager benchmark Run pm-bench on PR labels May 19, 2026

gemini-code-assist Bot reviewed May 19, 2026

View reviewed changes

elrrrrrrr mentioned this pull request May 19, 2026

perf(pm): batch install clone completions #2989

Draft

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf(pm): flush clone unblockers early#2994

perf(pm): flush clone unblockers early#2994
elrrrrrrr wants to merge 1 commit into
exp/pm-install-clone-batch-completionsfrom
exp/pm-install-flush-clone-unblockers-early

elrrrrrrr commented May 19, 2026 •

edited

Loading

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot May 19, 2026

Uh oh!

gemini-code-assist Bot May 19, 2026

Uh oh!

gemini-code-assist Bot May 19, 2026

Uh oh!

github-actions Bot commented May 19, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

elrrrrrrr commented May 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Hypothesis

Validation

Bench Result

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot May 19, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented May 19, 2026

📊 pm-bench-phases · e760c43 · linux (ubuntu-latest)

npmjs.org

p0_full_cold

p1_resolve

p3_cold_install

p4_warm_link

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

elrrrrrrr commented May 19, 2026 •

edited

Loading

📊 pm-bench-phases · `e760c43` · linux (`ubuntu-latest`)