Compact child observation

[SandroMaglione]

Summary

• replace the replaying child-registry PubSub and complete registry snapshots with compact, scoped observer handoffs
• preserve initial replay, independent subscribers, registry-wide emission ticks, ordered backlog delivery, and lifecycle race safety
• keep the unobserved representation allocation-free and release observer state when a stream is finalized
• add coverage for multiple subscribers, compatibility ticks, and a stalled observer backlog

Why

Observing one child previously cloned and retained the complete child registry behind a replaying PubSub. The public stream only needs the selected child value, so retaining full snapshots made observation substantially more expensive than the underlying registry.

Changeset

• Added a patch changeset for the runtime memory improvement.

Validation

• pnpm check
• pnpm perf:types
• Three independent local runtime benchmark processes for both main and this branch
• Relevant example checks are covered by pnpm check
• Reviewed the independent CI runtime performance report

The local median reduces observed child-registry retained heap from 28.0 KiB to 21.2 KiB (-24.3%). Unobserved machine and child-family memory are unchanged within measurement precision; control throughput benchmarks show no material regression.

[github-actions]

Type performance

Measured with TypeScript 6.0.3 and skipLibCheck=true.

Scenario	Base	PR	Difference
Effect only	55	55	0 (0.0%)
Import effect-machine	55	55	0 (0.0%)
Machine.defineStates (3 states)	2,965	2,965	0 (0.0%)
Machine.make (3 states, 2 events)	8,715	8,715	0 (0.0%)
machine.handle (3 states, 2 transitions)	23,492	23,492	0 (0.0%)
machine.handle (depth 8)	117,075	117,075	0 (0.0%)
machine.handle (depth 12)	129,713	129,713	0 (0.0%)
machine.handle (depth 16)	145,055	145,055	0 (0.0%)
machine.handle (depth 24)	183,851	183,851	0 (0.0%)
machine.handle (wide depth 16)	214,189	214,189	0 (0.0%)
machine.handle (parallel/history/choice)	132,667	132,667	0 (0.0%)
machine.handle (4 successive calls)	128,674	128,674	0 (0.0%)
machine exact input/output/error/services	110,584	110,584	0 (0.0%)
execution adapter readiness	127,854	127,854	0 (0.0%)

Marginal instantiations are measured against the matching setup without that API call:

Scenario	Base	PR	Difference
Import effect-machine	0	0	0
Machine.defineStates (3 states)	2,910	2,910	0 (0.0%)
Machine.make (3 states, 2 events)	5,742	5,742	0 (0.0%)
machine.handle (3 states, 2 transitions)	14,777	14,777	0 (0.0%)
machine.handle (depth 8)	109,601	109,601	0 (0.0%)
machine.handle (depth 12)	121,039	121,039	0 (0.0%)
machine.handle (depth 16)	135,181	135,181	0 (0.0%)
machine.handle (depth 24)	171,577	171,577	0 (0.0%)
machine.handle (wide depth 16)	202,224	202,224	0 (0.0%)
machine.handle (parallel/history/choice)	114,331	114,331	0 (0.0%)
machine.handle (4 successive calls)	113,703	113,703	0 (0.0%)
machine exact input/output/error/services	100,670	100,670	0 (0.0%)
execution adapter readiness	102,092	102,092	0 (0.0%)

Check times (informational)

Scenario	Base	PR
Effect only	0.03s	0.03s
Import effect-machine	0.04s	0.04s
Machine.defineStates (3 states)	0.11s	0.11s
Machine.make (3 states, 2 events)	0.17s	0.16s
machine.handle (3 states, 2 transitions)	0.27s	0.25s
machine.handle (depth 8)	0.56s	0.52s
machine.handle (depth 12)	0.59s	0.59s
machine.handle (depth 16)	0.64s	0.61s
machine.handle (depth 24)	0.73s	0.70s
machine.handle (wide depth 16)	0.81s	0.79s
machine.handle (parallel/history/choice)	0.66s	0.64s
machine.handle (4 successive calls)	0.63s	0.59s
machine exact input/output/error/services	0.53s	0.54s
execution adapter readiness	0.60s	0.60s

Type instantiations are the comparison metric. Check time varies with runner load and is informational only.

[github-actions]

Runtime performance

Median of 5 independent benchmark processes on AMD EPYC 7763 64-Core Processor with Node v24.18.0.

Pull request baseline

Scenario	Effect Machine	XState 5	XState 6 alpha
Plan counter transitions	107,347 transitions/s	140,612 transitions/s	125,849 transitions/s
Drain burst with terminal fence	89,863 increments/s	413,435 increments/s	114,661 increments/s
Drain burst with a change observer	66,607 increments/s	410,990 increments/s	114,359 increments/s
Lookup and send to one child	84,908 increments/s	409,371 increments/s	109,054 increments/s
Start and stop a machine	18,776 machines/s	191,975 machines/s	73,883 machines/s
Start and stop a parent with one child	6,569 families/s	70,641 families/s	27,235 families/s

Effect runtime reference points

Scenario	Effect Machine	Effect runtime primitives
Start and stop a raw compiled process	30,863 processes/s	—
Start and interrupt a suspended fiber	—	168,606 fibers/s
Start and stop a queue worker	—	117,980 workers/s
Start and stop an actor shell	—	108,968 actors/s
Start and stop two actor shells	—	68,087 families/s
Update an owner-only mutable snapshot	—	316,856,781 updates/s
Update a synchronized snapshot	—	1,547,755 updates/s
Create, resolve, and await a terminal latch	—	1,438,518 latches/s

Memory profile	Effect Machine	XState 5	XState 6 alpha	Effect runtime primitives
Idle machine	7.5 KiB	3.8 KiB	6.1 KiB	—
Raw generic managed process	12.5 KiB	—	—	—
Raw compiled process	7.8 KiB	—	—	—
Two independent idle machines	14.5 KiB	7.3 KiB	11.8 KiB	—
Idle parent with one child	17.0 KiB	5.5 KiB	8.8 KiB	—
Parent with observed child registry	21.2 KiB	—	—	—
Parent with observed invoked child snapshots	17.2 KiB	—	—	—
Suspended Effect fiber	—	—	—	0.7 KiB
Effect queue with waiting fiber	—	—	—	2.8 KiB
Effect mailbox actor shell	—	—	—	3.2 KiB
Two Effect actor shells	—	—	—	6.9 KiB

Effect Machine change from base

Metric	Base	Base variability	PR	PR variability	Difference
Plan counter transitions	110,215 transitions/s	0.7% MAD	107,347 transitions/s	4.1% MAD	-2.6%
Drain burst with terminal fence	91,695 increments/s	1.5% MAD	89,863 increments/s	2.1% MAD	-2.0%
Drain burst with a change observer	66,678 increments/s	2.2% MAD	66,607 increments/s	0.3% MAD	-0.1%
Lookup and send to one child	85,613 increments/s	0.7% MAD	84,908 increments/s	0.9% MAD	-0.8%
Start and stop a machine	18,368 machines/s	0.9% MAD	18,776 machines/s	1.5% MAD	+2.2%
Start and stop a parent with one child	6,519 families/s	0.4% MAD	6,569 families/s	3.1% MAD	+0.8%
Idle machine heap per unit	7.5 KiB	0.0% MAD	7.5 KiB	0.0% MAD	0.0%
Raw generic managed process heap per unit	12.5 KiB	0.0% MAD	12.5 KiB	0.0% MAD	-0.0%
Raw compiled process heap per unit	7.8 KiB	0.3% MAD	7.8 KiB	0.2% MAD	-0.0%
Two independent idle machines heap per unit	14.5 KiB	0.0% MAD	14.5 KiB	0.0% MAD	0.0%
Idle parent with one child heap per unit	17.0 KiB	0.0% MAD	17.0 KiB	0.0% MAD	-0.0%
Parent with observed child registry heap per unit	28.0 KiB	0.0% MAD	21.2 KiB	0.0% MAD	-24.3%
Parent with observed invoked child snapshots heap per unit	17.2 KiB	0.0% MAD	17.2 KiB	0.0% MAD	-0.0%

Effect runtime reference change from base

Metric	Base	Base variability	PR	PR variability	Difference
Start and stop a raw compiled process	31,378 processes/s	0.2% MAD	30,863 processes/s	0.6% MAD	-1.6%
Start and interrupt a suspended fiber	169,463 fibers/s	0.5% MAD	168,606 fibers/s	0.3% MAD	-0.5%
Start and stop a queue worker	118,120 workers/s	1.3% MAD	117,980 workers/s	0.6% MAD	-0.1%
Start and stop an actor shell	109,087 actors/s	2.1% MAD	108,968 actors/s	0.3% MAD	-0.1%
Start and stop two actor shells	66,765 families/s	4.4% MAD	68,087 families/s	1.3% MAD	+2.0%
Update an owner-only mutable snapshot	316,856,781 updates/s	0.0% MAD	316,856,781 updates/s	0.0% MAD	0.0%
Update a synchronized snapshot	1,541,680 updates/s	1.2% MAD	1,547,755 updates/s	1.0% MAD	+0.4%
Create, resolve, and await a terminal latch	1,428,514 latches/s	2.3% MAD	1,438,518 latches/s	0.8% MAD	+0.7%

Versions and interpretation

• Effect Machine: 0.3.0
• XState 5: 5.32.5
• XState 6 alpha: 6.0.0-alpha.27
• Effect runtime primitives: 4.0.0-beta.105

Higher throughput is better; lower heap is better. Variability is the median absolute deviation across independent processes, relative to their median. Runtime measurements on shared GitHub-hosted hardware remain informational, so small differences should be confirmed across multiple workflow runs.