feat(runtime): report replay cost per step batch, flagged by retained-VM mode by shalabhc · Pull Request #3490 · vercel/workflow

shalabhc · 2026-08-12T16:38:07Z

This changes metrics and spans only:

More spans get replayCost
Added retained as an optimization tag so it is reported by the backend spans.

Details

## Summary

Two changes, both in @workflow/core:

Replay cost is now reported for every batch that qualifies for TTFS or STSO, not just the run's first step. replayMs was gated on RSFS eligibility, so every ordinary step-to-step transition — the STSO population, which is the bulk of step scheduling — had no replay-cost telemetry at all. The value was already measured at every suspension (replayDurationMs in runtime.ts) and then discarded.
Whether a retained VM served the batch is reported as a retained entry in the existing optimizations list, alongside turbo/lazyStepStart/optimisticStart. finalSchedulingReplay measures a resume on the retained path and a full replay otherwise, so without this dimension the metric is knowingly bimodal — the existing doc comment says so.

Why `optimizations` instead of a dedicated field

An earlier version of this PR added a replayMode: 'replay' | 'retained' wire field, which cost two extra packages here (world-vercel, world) plus a parse/sanitize path and a metric tag in workflow-server.

Riding the existing optimizations array instead means the SDK change is one line and the server needs no new parsing at all: reportClientStepLatency already turns each known entry into a bounded boolean tag applied to every latency metric. So the companion backend change — vercel/workflow-server#767 — collapses to adding one entry to OPTIMIZATION_TAGS, and retained dimensions ttfs_ms, stso_ms, rsfs_ms and final_scheduling_replay_ms rather than only the last of those. Unknown entries are already ignored server-side, so old and new SDKs interoperate; client_version is already a tag on these metrics, so "no flag because full replay" stays distinguishable from "no flag because old SDK".

Why the retained flag is computed here rather than read off a span

traceExecution tags the workflow.run span retained when the span opens, before session.resume() can come back with { type: 'replay' } on an internal cache miss and fall through to a real replay. The flag in this PR is computed after the resume result is known, so it does not inherit that mislabeling.

Motivation

Production analysis of the v0chat STSO tail: retained batches were 88.5% of sampled steps with a 2.0 s STSO p75, full replays 11.4% with a 58.7 s p75, and full replay accounted for 66% of everything above p90. Mode is the strongest tail discriminator found, and it was only reachable via a cross-span temporal join over ~7% sampled spans.

Test plan

pnpm exec vitest run packages/core/src/runtime/step-latency.test.ts — 38 passed, including the STSO broadening and both retained states
pnpm exec vitest run packages/core/src/runtime.test.ts — 44 passed; the turbo latency test now pins that the first step's pass (a full replay that built the VM) has no retained flag while the second step's pass, resumed from the session the same invocation retained, does
pnpm exec vitest run packages/core/src/workflow.test.ts — 89 passed, 1 expected fail
pnpm --filter @workflow/core typecheck
Confirm in staging that step_completed/step_failed events for ordinary (non-first-step) suspensions carry finalSchedulingReplay, and that retained appears in optimizations

changeset-bot · 2026-08-12T16:38:12Z

🦋 Changeset detected

Latest commit: d098a38

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 16 packages

Name	Type
@workflow/core	Patch
@workflow/builders	Patch
@workflow/cli	Patch
@workflow/next	Patch
@workflow/nitro	Patch
@workflow/vitest	Patch
@workflow/web-shared	Patch
@workflow/web	Patch
workflow	Patch
@workflow/world-testing	Patch
@workflow/astro	Patch
@workflow/nest	Patch
@workflow/rollup	Patch
@workflow/sveltekit	Patch
@workflow/vite	Patch
@workflow/nuxt	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

vercel · 2026-08-12T16:38:13Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
example-nextjs-workflow-turbopack	Ready	Preview, v0	Aug 25, 2026 2:25am
example-nextjs-workflow-webpack	Ready	Preview, v0	Aug 25, 2026 2:25am
example-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-astro-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-express-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-fastify-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-hono-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-nestjs-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-nitro-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-nuxt-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-python-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-sveltekit-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-tanstack-start-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workbench-vite-workflow	Ready	Preview, v0	Aug 25, 2026 2:25am
workflow-docs	Ready	Preview, v0	Aug 25, 2026 2:25am
workflow-swc-playground	Ready	Preview, v0	Aug 25, 2026 2:25am
workflow-tarballs	Ready	Preview, v0	Aug 25, 2026 2:25am
workflow-web	Ready	Preview, v0	Aug 25, 2026 2:25am

github-actions · 2026-08-12T16:38:33Z

🧪 E2E Test Results

❌ Some tests failed

❌ Failed E2E Tests

▲ Vercel Production (8 failed)

python-node (8 failed):

promiseAllWorkflow | wrun_41M0VBPBP70GT1Q21CC3JJNVGC | 🔍 observability
sleepingWorkflow | wrun_41M0VBQ0VS0GM7A3J76GT6HYFK | 🔍 observability
parallelSleepWorkflow | wrun_41M0VBQ1870GPTSJDC6GN0JG40 | 🔍 observability
nullByteWorkflow | wrun_41M0VBQAAT0GS7PKQA5BBV5NRP | 🔍 observability
cancelRun - cancelling a running workflow | wrun_41M0VBW2J70GR0B55AQ4CG1P5V | 🔍 observability
cancelRun via CLI - cancelling a running workflow | wrun_41M0VBW3BH0GZQFXK405MCYXBJ | 🔍 observability
sleepInLoopWorkflow - sleep inside loop with steps actually delays each iteration | wrun_41M0VBWB5P0GST5NWVFS8ZCEBF | 🔍 observability
resilient start: addTenWorkflow completes when run_created returns 500 | wrun_41M0VBWTZ10GJZ0P8E92XS7BE5 | 🔍 observability

🌐 Cross-language Conformance (9 failed)

python (9 failed):

deploymentId: 'latest' is a no-op in non-Vercel worlds | wrun_01M0VBWZ511X73CJFBDR7971ZN
promiseAllWorkflow | wrun_41M0VBPBP70GT1Q21CC3JJNVGC
sleepingWorkflow | wrun_41M0VBQ0VS0GM7A3J76GT6HYFK
parallelSleepWorkflow | wrun_41M0VBQ1870GPTSJDC6GN0JG40
nullByteWorkflow | wrun_41M0VBQAAT0GS7PKQA5BBV5NRP
cancelRun - cancelling a running workflow | wrun_41M0VBW2J70GR0B55AQ4CG1P5V
cancelRun via CLI - cancelling a running workflow | wrun_41M0VBW3BH0GZQFXK405MCYXBJ
sleepInLoopWorkflow - sleep inside loop with steps actually delays each iteration | wrun_41M0VBWB5P0GST5NWVFS8ZCEBF
resilient start: addTenWorkflow completes when run_created returns 500 | wrun_41M0VBWTZ10GJZ0P8E92XS7BE5

⚠️ Flaky E2E Tests (passed on retry)

These tests failed at least once and passed on a retry. A recurring entry here is a real race worth investigating.

addTenWorkflow (nitro)
cancelRun via CLI - cancelling a running workflow (nitro)
getTailIndex returns correct index after stream completes (nextjs-webpack)
no startIndex (reads all chunks) (hono)
webhookWorkflow (sveltekit)

🛠 Infra Events (absorbed by the harness)

Platform anomalies the e2e harness detected and worked around (e.g. a run the queue never picked up, replaced by a fresh run). Clustered timestamps indicate a backend blip; a steady drip indicates a platform issue worth escalating.

32 infra events

cold-start-warmup · suite warmup (tanstack-start) · at 02:25:08Z · abandoned wrun_01M0VBQVVSYQXV7BFB3P3W6DF9
cold-start-warmup · suite warmup (python) · at 02:25:32Z · abandoned wrun_41M0VBNRQD0GVXC4WWMM466F68 · (+7 more)
run-pickup-stall · sleepingWorkflow (python) · at 02:25:48Z · abandoned wrun_41M0VBSDMM0GS8EF550NJ0R1QB
run-pickup-stall · parallelSleepWorkflow (python) · at 02:25:48Z · abandoned wrun_41M0VBSDMM0GS8EF550NJ0R1QC
run-pickup-stall · nullByteWorkflow (python) · at 02:25:48Z · abandoned wrun_41M0VBSDMR0GSGQ0KXBHEY20Q0
run-pickup-stall · promiseAllWorkflow (python) · at 02:25:48Z · abandoned wrun_41M0VBSDME0GG72CQQBCXKZH8E
run-pickup-stall · cancelRun - cancelling a running workflow (python) · at 02:25:48Z · abandoned wrun_41M0VBSE0Q0GTH9YZH084Z8QSP
cold-start-warmup · suite warmup (python) · at 02:26:13Z · abandoned wrun_01M0VBQ0Q6PE7TGWHPP35NHP2S · (+7 more)
run-pickup-stall · cancelRun - cancelling a running workflow (python) · at 02:26:21Z · abandoned wrun_41M0VBTDHK0GJP6GT5R76EJREE
run-pickup-stall · sleepingWorkflow (python) · at 02:26:28Z · abandoned wrun_01M0VBTNW1BR42K2J6SJQD7TQC
run-pickup-stall · deploymentId: 'latest' is a no-op in non-Vercel worlds (python) · at 02:26:28Z · abandoned wrun_01M0VBTNVWY313Z0S53STSTYZG
run-pickup-stall · parallelSleepWorkflow (python) · at 02:26:28Z · abandoned wrun_01M0VBTNW3DBVA7VYGRSZ5E78D
run-pickup-stall · promiseAllWorkflow (python) · at 02:26:28Z · abandoned wrun_01M0VBTNVWY313Z0S53STSTYZH
run-pickup-stall · nullByteWorkflow (python) · at 02:26:28Z · abandoned wrun_01M0VBTNW5DA4RMM1VC1XYF2H4
run-pickup-stall · sleepingWorkflow (python) · at 02:26:49Z · abandoned wrun_41M0VBV94V0GJKG8XF36PWQ4HG
run-pickup-stall · promiseAllWorkflow (python) · at 02:26:49Z · abandoned wrun_41M0VBV95R0GXZ8TY09FAD32VT
run-pickup-stall · nullByteWorkflow (python) · at 02:26:49Z · abandoned wrun_41M0VBV9AY0GYSZRKAA12NPQ5J
run-pickup-stall · parallelSleepWorkflow (python) · at 02:26:52Z · abandoned wrun_41M0VBVCJV0GM8Y9T83QGHJVQF
run-pickup-stall · deploymentId: 'latest' is a no-op in non-Vercel worlds (python) · at 02:27:28Z · abandoned wrun_01M0VBWGFHT5HC3QJ1JX74BHTK
run-pickup-stall · sleepingWorkflow (python) · at 02:27:28Z · abandoned wrun_01M0VBWGFNEVZJQ6K1M7CKHCNC
run-pickup-stall · promiseAllWorkflow (python) · at 02:27:28Z · abandoned wrun_01M0VBWGFK2VTQB12AESNPXXSS
run-pickup-stall · parallelSleepWorkflow (python) · at 02:27:28Z · abandoned wrun_01M0VBWGFW6996ANDE95S5RFC3
run-pickup-stall · nullByteWorkflow (python) · at 02:27:28Z · abandoned wrun_01M0VBWGG45XM3MXBVSW70YKR9
run-pickup-stall · cancelRun via CLI - cancelling a running workflow (python) · at 02:27:40Z · abandoned wrun_41M0VBVD830GVRJJ74GPKHS14T
run-pickup-stall · cancelRun via CLI - cancelling a running workflow (python) · at 02:28:28Z · abandoned wrun_01M0VBYB3BQTEMRZQRJK01VV90
run-pickup-stall · sleepInLoopWorkflow - sleep inside loop with steps actually delays each iteration (python) · at 02:28:28Z · abandoned wrun_01M0VBYB3HWH91ZW8VQN6KZ6KC
run-pickup-stall · cancelRun - cancelling a running workflow (python) · at 02:28:28Z · abandoned wrun_01M0VBYB37V7AS3JK8E93GSBG8
run-pickup-stall · cancelRun via CLI - cancelling a running workflow (python) · at 02:28:58Z · abandoned wrun_01M0VBZ8FA9C02E12W8GS9715N
run-pickup-stall · cancelRun - cancelling a running workflow (python) · at 02:28:58Z · abandoned wrun_01M0VBZ8F84ATAMPKEH3BJD1EA
run-pickup-stall · cancelRun via CLI - cancelling a running workflow (python) · at 02:29:25Z · abandoned wrun_41M0VBXES80GSN3YBQ5D9EVKVZ
run-pickup-stall · sleepInLoopWorkflow - sleep inside loop with steps actually delays each iteration (python) · at 02:29:26Z · abandoned wrun_41M0VBXE3K0GY4M29BXF5885R7
run-pickup-stall · sleepInLoopWorkflow - sleep inside loop with steps actually delays each iteration (python) · at 02:29:28Z · abandoned wrun_01M0VC05PNQPYQ3R7TK7P7FWGF

E2E Test Summary

Summary

	Passed	Failed	Skipped	Total
❌ ▲ Vercel Production	3570	8	742	4320
✅ 💻 Local Development	3922	0	558	4480
✅ 📦 Local Production	3922	0	558	4480
✅ 🐘 Local Postgres	3922	0	558	4480
✅ 🪟 Windows	320	0	0	320
❌ 🌐 Cross-language Conformance	0	9	132	141
✅ vercel-http-transport	817	0	143	960
✅ vercel-multi-region	27	0	0	27
✅ vercel-ws-transport	553	0	87	640
Total	17053	17	2778	19848

Details by Category

❌ ▲ Vercel Production

App	Passed	Failed	Skipped
✅ astro-node	132	0	28
✅ astro-quickjs	132	0	28
✅ example-node	132	0	28
✅ example-quickjs	132	0	28
✅ express-node	132	0	28
✅ express-quickjs	132	0	28
✅ fastify-node	132	0	28
✅ fastify-quickjs	132	0	28
✅ hono-node	132	0	28
✅ hono-quickjs	132	0	28
✅ nest-node	132	0	28
✅ nest-quickjs	132	0	28
✅ nextjs-turbopack-node	157	0	3
✅ nextjs-turbopack-quickjs	157	0	3
✅ nextjs-webpack-node	157	0	3
✅ nextjs-webpack-quickjs	157	0	3
✅ nitro-node	132	0	28
✅ nitro-quickjs	132	0	28
✅ nuxt-node	132	0	28
✅ nuxt-quickjs	132	0	28
❌ python-node	0	8	152
✅ sveltekit-node	151	0	9
✅ sveltekit-quickjs	151	0	9
✅ tanstack-start-node	132	0	28
✅ tanstack-start-quickjs	132	0	28
✅ vite-node	132	0	28
✅ vite-quickjs	132	0	28

✅ 💻 Local Development

App	Passed	Skipped
✅ astro-stable-node	134	26
✅ astro-stable-quickjs	134	26
✅ express-stable-node	134	26
✅ express-stable-quickjs	134	26
✅ fastify-stable-node	134	26
✅ fastify-stable-quickjs	134	26
✅ hono-stable-node	134	26
✅ hono-stable-quickjs	134	26
✅ nest-stable-node	134	26
✅ nest-stable-quickjs	134	26
✅ nextjs-turbopack-canary-node	141	19
✅ nextjs-turbopack-canary-quickjs	141	19
✅ nextjs-turbopack-stable-node	160	0
✅ nextjs-turbopack-stable-quickjs	160	0
✅ nextjs-webpack-canary-node	141	19
✅ nextjs-webpack-canary-quickjs	141	19
✅ nextjs-webpack-stable-node	160	0
✅ nextjs-webpack-stable-quickjs	160	0
✅ nitro-stable-node	134	26
✅ nitro-stable-quickjs	134	26
✅ nuxt-stable-node	134	26
✅ nuxt-stable-quickjs	134	26
✅ sveltekit-stable-node	153	7
✅ sveltekit-stable-quickjs	153	7
✅ tanstack-start-node	134	26
✅ tanstack-start-quickjs	134	26
✅ vite-stable-node	134	26
✅ vite-stable-quickjs	134	26

✅ 📦 Local Production

App	Passed	Skipped
✅ astro-stable-node	134	26
✅ astro-stable-quickjs	134	26
✅ express-stable-node	134	26
✅ express-stable-quickjs	134	26
✅ fastify-stable-node	134	26
✅ fastify-stable-quickjs	134	26
✅ hono-stable-node	134	26
✅ hono-stable-quickjs	134	26
✅ nest-stable-node	134	26
✅ nest-stable-quickjs	134	26
✅ nextjs-turbopack-canary-node	141	19
✅ nextjs-turbopack-canary-quickjs	141	19
✅ nextjs-turbopack-stable-node	160	0
✅ nextjs-turbopack-stable-quickjs	160	0
✅ nextjs-webpack-canary-node	141	19
✅ nextjs-webpack-canary-quickjs	141	19
✅ nextjs-webpack-stable-node	160	0
✅ nextjs-webpack-stable-quickjs	160	0
✅ nitro-stable-node	134	26
✅ nitro-stable-quickjs	134	26
✅ nuxt-stable-node	134	26
✅ nuxt-stable-quickjs	134	26
✅ sveltekit-stable-node	153	7
✅ sveltekit-stable-quickjs	153	7
✅ tanstack-start-node	134	26
✅ tanstack-start-quickjs	134	26
✅ vite-stable-node	134	26
✅ vite-stable-quickjs	134	26

✅ 🐘 Local Postgres

App	Passed	Skipped
✅ astro-stable-node	134	26
✅ astro-stable-quickjs	134	26
✅ express-stable-node	134	26
✅ express-stable-quickjs	134	26
✅ fastify-stable-node	134	26
✅ fastify-stable-quickjs	134	26
✅ hono-stable-node	134	26
✅ hono-stable-quickjs	134	26
✅ nest-stable-node	134	26
✅ nest-stable-quickjs	134	26
✅ nextjs-turbopack-canary-node	141	19
✅ nextjs-turbopack-canary-quickjs	141	19
✅ nextjs-turbopack-stable-node	160	0
✅ nextjs-turbopack-stable-quickjs	160	0
✅ nextjs-webpack-canary-node	141	19
✅ nextjs-webpack-canary-quickjs	141	19
✅ nextjs-webpack-stable-node	160	0
✅ nextjs-webpack-stable-quickjs	160	0
✅ nitro-stable-node	134	26
✅ nitro-stable-quickjs	134	26
✅ nuxt-stable-node	134	26
✅ nuxt-stable-quickjs	134	26
✅ sveltekit-stable-node	153	7
✅ sveltekit-stable-quickjs	153	7
✅ tanstack-start-node	134	26
✅ tanstack-start-quickjs	134	26
✅ vite-stable-node	134	26
✅ vite-stable-quickjs	134	26

✅ 🪟 Windows

App	Passed	Failed	Skipped
✅ nextjs-turbopack-node	160	0	0
✅ nextjs-turbopack-quickjs	160	0	0

❌ 🌐 Cross-language Conformance

App	Passed	Failed	Skipped
❌ python	0	9	132

✅ vercel-http-transport

App	Passed	Skipped
✅ example	132	28
✅ express	132	28
✅ hono	132	28
✅ nextjs-turbopack	157	3
✅ nitro	132	28
✅ vite	132	28

✅ vercel-multi-region

App	Passed	Failed	Skipped
✅ nextjs-turbopack	27	0	0

✅ vercel-ws-transport

App	Passed	Skipped
✅ example	132	28
✅ express	132	28
✅ nextjs-turbopack	157	3
✅ vite	132	28

📋 View full workflow run

github-actions · 2026-08-12T16:38:38Z

📊 Workflow Benchmarks

commit d098a38 · Tue, 25 Aug 2026 02:47:33 GMT · run logs

Backend: vercel · app: nextjs-turbopack

Metric	Scenario	Best (ms)	P75 (ms)	P90 (ms)	P99 (ms)	Samples
TTFS	step	274 (-74%) 💚	1367 🔴 (+20%) 🔻	1477 🔴 (+22%) 🔻	1502 🔴 (+16%) 🔻	30
TTFS	stream	269 (-65%) 💚	1446 🔴 (+31%) 🔻	1486 🔴 (+30%) 🔻	1517 🔴 (+26%) 🔻	30
TTFS	hook + stream	1608 (+243%) 🔻	1796 🔴 (+24%) 🔻	1922 🔴 (+22%) 🔻	5904 🔴 (+26%) 🔻	30
Fan-out TTFS	Promise.all(100 steps)	666 (+23%) 🔻	2220 (+129%) 🔻	2302 (+46%) 🔻	2716 (+47%) 🔻	10
Fan-out TTLS	Promise.all(100 steps)	2044 (+3.3%)	3782 (+29%) 🔻	3789 (+9.0%)	8368 (+11%)	10
STSO	1020 steps (inline)	115 (+19%) 🔻	140 (-7.9%)	156 (-12%)	247 (+3.8%)	1019
WO	1020 steps	140422 (-3.8%)	140422 (-3.8%)	140422 (-3.8%)	140422 (-3.8%)	1
CRTT	first chunk (pooled)	102 (-9.7%)	198 (+11%)	235 (+13%)	246 (-39%) 💚	28

Streams

Scenario	CRTT 1st	p75	p90	p99	CDV max	iters
paced control (100/s, 60B)	168 (+19%)	149 (-29%)	194 (-53%)	334 (-43%)	130 (-35%)	10
size sweep (100/s, 160B-12KB)	128 (-17%)	142 (-44%)	200 (-51%)	302 (-65%)	132 (-40%)	10
replay gateway-gpt-5.4-nano-2000t (1x)	165 (+33%)	119 (-44%)	150 (-53%)	350 (-50%)	184 (-60%)	3
replay eve-gpt-5.6-sol-2000t (1x)	180 (+6%)	132 (-77%)	191 (-92%)	635 (-85%)	497 (-77%)	2
replay eve-gpt-5.6-sol-2000t (2x)	198 (+8%)	170 (-44%)	230 (-47%)	346 (-46%)	229 (-43%)	3

📈 STSO distribution vs main (inline / queue-hop histograms)

1020 steps (inline)

Cumulative STSO time: main 145414ms → this run 140245ms (Δ -5169ms, -4%)

 50-100 ms  ┃                         main   1  this   0    -1
100-150 ms  ████████████████████░░░┃  main 731  this 877  +146
150-200 ms  ██┃████                   main 247  this 116  -131
200-250 ms  ┃                         main  33  this  17   -16
250-300 ms  ┃                         main   3  this   8    +5
300-350 ms  ┃                         main   2  this   0    -2
350-400 ms  ┃                         main   2  this   1    -1

📈 CRTT drill-down vs main (RTT distributions & profiles)

variant  RTT 1ms→5s+             avg         p50         p90         p99     n
control  ······▄█▁····  122.8 (-28%)  115 (-21%)  194 (-53%)  334 (-43%)  3000
sweep    ······▅█▁····  121.2 (-39%)  111 (-31%)  200 (-51%)  302 (-65%)  3000
gw 1x    ·····▁██▁····  107.8 (-38%)  100 (-30%)  150 (-53%)  350 (-50%)  5295
eve 1x   ·····▁██▂▁···  122.3 (-73%)  101 (-34%)  191 (-92%)  635 (-85%)  5186
eve 2x   ·····▁▄█▂····  139.8 (-35%)  124 (-33%)  230 (-47%)  346 (-46%)  7779

RTT over stream progress (avg per tenth of stream, bars scaled min→max):

control  █▃▃▃▅▄▃▁▃▃  104–153ms
sweep    █▄▄▄▁▂▃▁▄▄  102–158ms
gw 1x    ██▆▆▁▁▃▂▇▂  98–118ms
eve 1x   █▂▃▅▅▂▃▄▂▁  100–161ms
eve 2x   ▃▂▅▁▁▅▇█▅▂  122–165ms

RTT by chunk size (avg per log size bin, ~160B → ~12KB serialized, bars scaled min→max):

sweep  ▂██▇▄▅▁  119–123ms

Delivery jitter over stream progress (avg positive CDV per tenth of stream, bars scaled min→max):

control  ▁▆█▇▆▇▄▄▄▅  28–45ms
sweep    ▇▃█▃▁▃▆▃▄▂  41–61ms
gw 1x    █▄▄▅▂▁█▃▇▁  29–35ms
eve 1x   █▄▅▆▆▆▆▅▃▁  20–29ms
eve 2x   █▅▆▄▅▆▁▃▇▆  20–26ms

ℹ️ Metric definitions & methodology

_{Streams: first-chunk RTT (the stream-open path, before any buffering/backpressure), CRTT percentiles, and worst delivery stall (CDV max). Cells are medians across iterations; per-run values in the artifacts. No 🔴/🟢 marks until targets attach.}

_{The collapsed STSO distribution section above buckets every step gap, split inline (same warm process — pure framework overhead) vs queue-hop (fresh process — dispatch, reinit, replay). █ = main, ┃ = this run, ░ = fill.}

_{The collapsed CRTT drill-down: per-variant RTT histograms (fixed log bins, · = empty) and mean RTT/positive-CDV profile lines over stream progress and chunk size. Histograms, avgs, and profiles merge exactly across runs; p50–p99 are percentile-of-percentiles. Per-index rows live in the artifacts.}

_{Best/P75/P90/P99 deltas compare against the most recent benchmark run on main at the time of this run. 🔻 flags a delta worse than +15%, 💚 one better than −15%.}

_{Metrics — TTFS: time to first step body (in-deployment start() → first step body) · Fan-out TTFS: fan-out time to first step (in-deployment start() → first of the parallel step bodies to complete) · Fan-out TTLS: fan-out time to last step (in-deployment start() → last of the parallel step bodies to complete, i.e. when the Promise.all resolves) · STSO: step-to-step overhead (gap between consecutive step bodies) · WO: workflow overhead (whole-run time outside step bodies, in-deployment anchored) · CRTT: chunk round-trip time (per-chunk write → read latency, one clock domain: deployment → stream backend → same deployment) · CDV: chunk delay variation / delivery jitter (inter-arrival gap minus inter-write gap per seq-adjacent pair; skew-free; the row is each run's MAX positive value, so one stall moves it)}

_{Scenarios — step: one trivial no-op step, no stream; no hooks, so the run stays in turbo mode (in-process fast path) · stream: one streaming step; no hooks, so the run stays in turbo mode (in-process fast path) · hook + stream: registers a hook before one step, which exits turbo mode (dispatch path) · 1020 steps: 1020 trivial sequential steps; STSO is measured between consecutive steps in the given step ranges, and WO is the whole-run overhead outside step bodies · Promise.all(100 steps): 100 trivial no-op steps started together in a single Promise.all; Fan-out TTFS is the first of them to complete and Fan-out TTLS the last, both from the in-deployment clientStart, so their gap is the spread the runtime adds across the fan-out · paced control (100/s, 60B): the control: 300 tiny (~60B) deltas metronome-paced at 100/s — zero workload structure, so it reads the transport floor and flush cadence, and disambiguates transport-wide vs workload-specific when a replay row moves · size sweep (100/s, 160B-12KB): same pacing as the control with deltas padded in rotation across seven log-spaced sizes (~160B–12KB) — rotation decouples size from stream position, so it isolates whether chunk size causes latency · replay gateway-gpt-5.4-nano-2000t (1x): raw provider SSE cadence captured at the AI gateway boundary (gpt-5.4-nano, the most popular gateway model; per-token deltas p50 208B = the modal production chunk size), replayed exactly as measured — the typical customer's workload; its CDV is the typical customer's real delivery jitter · replay eve-gpt-5.6-sol-2000t (1x): a captured eve turn (gpt-5.6-sol, the most-used demanding eve model; ~2000 output tokens = production p50 turn length) replayed exactly as measured — eve's envelope protocol re-ships the cumulative message so sizes ramp 142B→13KB; the demanding outlier tenant's reality · replay eve-gpt-5.6-sol-2000t (2x): the same eve capture at 2x — the headroom/stress row; real fast-tier models emit the same chunk sizes at proportionally higher rate, so time compression is a faithful speed model · first chunk (pooled): every run's seq-0 RTT pooled across all stream scenarios — the first chunk precedes any workload differentiation, so pooling samples one shared stream-open path with exact percentiles}

_{Replay cadences (semantic sha256) — eve-gpt-5.6-sol-2000t eaf22f5946e7c61f3c65c7006d550df180cfabd4e706254a09f22aec0cfb420d · gateway-gpt-5.4-nano-2000t 6f24ac518b6b83ff1d0e85a5fe78230db192716d66a7fc6b2fe022752001d041}

_{🔴 marks a percentile over its target (within target is left unmarked). Targets (p75/p90/p99, ms) — TTFS 200/300/600}

_{All timestamps are deployment-side; runs are triggered in-deployment, so the CI runner and api.vercel.com sit outside every measured window. TTFS = start() → first step body (includes dispatch + any cold start); Fan-out TTFS/TTLS = first/last step completion of one Promise.all from the same anchor (the gap is the runtime’s fan-out spread); STSO/WO between step bodies; CRTT inside the workflow (excludes the api.vercel.com read path).}

_{Cold starts stay in the numbers (real bursty-workload latency, inflates P75+); Best is the warm floor.}

github-actions · 2026-08-12T16:41:37Z

Sim World

Simulated world deterministic testing for races. Traces

🟠 world-sim scenario book — 1 fail of 41 total

fence=per-spec

scenario	outcome	events	virt	replay
✅ `smoke-no-steps`	completed	3	0ms	ok
✅ `smoke-one-step`	completed	6	0ms	ok
✅ `hook-at-step-started`	completed	12	0ms	ok
✅ `hook-at-step-completed`	completed	12	0ms	ok
✅ `hook-at-hook-created`	completed	12	0ms	ok
✅ `deadline-hook-wins`	completed	7	1.0h	ok
✅ `deadline-expires`	completed	7	1.0h	ok
✅ `long-sleep`	completed	11	30.0d	ok
✅ `hook-never-arrives`	stalled	3	0ms	skipped
✅ `step-retries-twice`	completed	10	2.0s	ok
✅ `parallel-steps`	completed	9	0ms	ok
✅ `hook-on-execution-state`	completed	12	0ms	ok
✅ `peek-hook-before-branch`	completed	12	0ms	ok
✅ `peek-hook-after-branch`	completed	12	0ms	ok
✅ `peek-hook-at-registration`	completed	12	0ms	ok
✅ `race-hook-before-probe`	completed	12	0ms	ok
✅ `race-hook-after-probe`	completed	12	0ms	ok
✅ `race-duplicate-delivery`	completed	13	0ms	ok
✅ `attr-hook-before-step`	completed	11	0ms	ok
✅ `attr-hook-after-step`	completed	11	0ms	ok
✅ `attr-from-step-body`	completed	13	0ms	ok
✅ `fork-hook-after-timeout`	completed	14	1.0m	ok
✅ `fork-hook-before-timeout`	completed	14	1.0m	ok
✅ `count-hook-after-timeout`	completed	17	1.0m	ok
✅ `count-hook-before-timeout`	completed	20	1.0m	ok
✅ `stale-read-step-count-fork`	completed	20	1.0m	ok
✅ `stale-read-equal-step-counts`	completed	14	1.0m	ok
✅ `step-vs-step-fork`	completed	12	0ms	ok
✅ `step-vs-step-fork-fenced`	completed	12	0ms	ok
✅ `fence-catches-benign-direction`	completed	12	5ms	ok
✅ `in-flight-before-decision`	completed	17	1.0m	ok
❌ `in-flight-before-decision-counted`	completed	17	1.0m	ok
✅ `in-flight-after-decision`	completed	19	2.0m	ok
✅ `stale-read-step-count-fork-fenced`	completed	20	1.0m	ok
✅ `fork-hook-wins`	completed	13	1.0m	ok
✅ `fork-timeout-wins`	completed	13	1.0m	ok
✅ `unclaimed-payload-under-fork`	completed	17	1.0m	ok
✅ `claimed-payload-under-fork`	completed	17	1.0m	ok
✅ `writers-independent-step-bodies`	completed	12	0ms	ok
✅ `writers-scripted-tempo`	completed	12	0ms	ok
✅ `cancel-mid-step`	cancelled	7	0ms	skipped

Full trace: world-sim.txt

…tained-VM mode Co-Authored-By: Shalabh Chaturvedi <7066873+shalabhc@users.noreply.github.com> Co-Authored-By: shalabhc <shalabh.chaturvedi@vercel.com>

github-actions · 2026-08-25T02:50:39Z

No backport to stable for f7fb012 (AI decision).

This commit adds new telemetry surface — it broadens replayMs/finalSchedulingReplay reporting to all STSO batches and introduces a new retained entry in the optimizations wire array — which is a feature enhancement (observability capability) rather than a fix to broken behavior on stable. It also depends on a companion backend change (vercel/workflow-server#767) to consume the new dimension. No correctness, crash, or data-loss defect is being repaired, so it belongs on main only.

To override, re-run the Backport to stable workflow manually via workflow_dispatch and paste this commit SHA into the ref input:

f7fb0126528fac2168492c3143fe0379b6f5ef1a

vercel Bot had a problem deploying to Preview – workbench-python-workflow August 12, 2026 16:38 Failure

vercel Bot deployed to Preview – workflow-web August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-fastify-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-nitro-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-vite-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-nestjs-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-hono-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workflow-tarballs August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-sveltekit-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-nuxt-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-astro-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – workbench-express-workflow August 12, 2026 16:39 View deployment

vercel Bot deployed to Preview – example-workflow August 12, 2026 16:40 View deployment

vercel Bot deployed to Preview – workbench-tanstack-start-workflow August 12, 2026 16:40 View deployment

vercel Bot deployed to Preview – workflow-docs August 12, 2026 16:40 View deployment

vercel Bot deployed to Preview – example-nextjs-workflow-turbopack August 12, 2026 16:40 View deployment

vercel Bot deployed to Preview – example-nextjs-workflow-webpack August 12, 2026 16:40 View deployment

vercel Bot deployed to Preview – workflow-swc-playground August 12, 2026 16:42 View deployment

vercel Bot had a problem deploying to Preview – workbench-python-workflow August 12, 2026 20:19 Failure

vercel Bot deployed to Preview – workflow-web August 12, 2026 20:20 View deployment

vercel Bot deployed to Preview – workflow-tarballs August 12, 2026 20:20 View deployment

vercel Bot deployed to Preview – workbench-astro-workflow August 12, 2026 20:20 View deployment

vercel Bot deployed to Preview – workbench-nestjs-workflow August 12, 2026 20:20 View deployment

vercel Bot deployed to Preview – workbench-tanstack-start-workflow August 12, 2026 20:20 View deployment

vercel Bot deployed to Preview – workbench-nuxt-workflow August 12, 2026 20:20 View deployment

shalabhc marked this pull request as ready for review August 12, 2026 20:48

shalabhc requested a review from a team as a code owner August 12, 2026 20:48

shalabhc marked this pull request as draft August 12, 2026 20:52

shalabhc force-pushed the shalabhc/replay-cost-metrics branch from 64da8d0 to efc2b5f Compare August 17, 2026 20:24

shalabhc changed the title ~~feat(runtime): report replay-vs-retained mode alongside finalSchedulingReplay~~ feat(runtime): report replay cost per step batch, flagged by retained-VM mode Aug 17, 2026

vercel Bot deployed to Preview – workbench-python-workflow August 17, 2026 20:25 View deployment

vercel Bot deployed to Preview – workflow-web August 17, 2026 20:25 View deployment

vercel Bot deployed to Preview – workflow-tarballs August 17, 2026 20:25 View deployment

vercel Bot deployed to Preview – workbench-express-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-tanstack-start-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-nestjs-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-fastify-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-hono-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-nitro-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-astro-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-vite-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – example-nextjs-workflow-turbopack August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-sveltekit-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – workbench-nuxt-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – example-workflow August 17, 2026 20:26 View deployment

vercel Bot deployed to Preview – example-nextjs-workflow-webpack August 17, 2026 20:27 View deployment

vercel Bot deployed to Preview – workflow-docs August 17, 2026 20:27 View deployment

vercel Bot deployed to Preview – workflow-swc-playground August 17, 2026 20:29 View deployment

shalabhc marked this pull request as ready for review August 17, 2026 21:49

VaguelySerious approved these changes Aug 21, 2026

View reviewed changes

feat(runtime): report replay cost for every step batch, flagged by re…

d098a38

…tained-VM mode Co-Authored-By: Shalabh Chaturvedi <7066873+shalabhc@users.noreply.github.com> Co-Authored-By: shalabhc <shalabh.chaturvedi@vercel.com>

This was referenced Aug 25, 2026

Version Packages (beta) #3775

Merged

Version Packages (beta) ben-vargas/workflow#104

Open

Version Packages (beta) Jokacar10/jkworkflow#519

Open

Navigation Menu

Uh oh!

feat(runtime): report replay cost per step batch, flagged by retained-VM mode - #3490

feat(runtime): report replay cost per step batch, flagged by retained-VM mode#3490
shalabhc merged 1 commit into
mainfrom
shalabhc/replay-cost-metrics

shalabhc commented Aug 12, 2026 •

edited

Loading

Uh oh!

changeset-bot Bot commented Aug 12, 2026 •

edited

Loading

Uh oh!

vercel Bot commented Aug 12, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 12, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 12, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 12, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Aug 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

shalabhc commented Aug 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why optimizations instead of a dedicated field

Why the retained flag is computed here rather than read off a span

Motivation

Test plan

Uh oh!

changeset-bot Bot commented Aug 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

vercel Bot commented Aug 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Aug 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🧪 E2E Test Results

❌ Failed E2E Tests

▲ Vercel Production (8 failed)

🌐 Cross-language Conformance (9 failed)

⚠️ Flaky E2E Tests (passed on retry)

🛠 Infra Events (absorbed by the harness)

E2E Test Summary

Uh oh!

github-actions Bot commented Aug 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

📊 Workflow Benchmarks

Uh oh!

github-actions Bot commented Aug 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Sim World

Uh oh!

github-actions Bot commented Aug 25, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

shalabhc commented Aug 12, 2026 •

edited

Loading

Why `optimizations` instead of a dedicated field

changeset-bot Bot commented Aug 12, 2026 •

edited

Loading

vercel Bot commented Aug 12, 2026 •

edited

Loading

github-actions Bot commented Aug 12, 2026 •

edited

Loading

github-actions Bot commented Aug 12, 2026 •

edited

Loading

github-actions Bot commented Aug 12, 2026 •

edited

Loading