Skip to content

[core] Prepare replay payloads as event frames arrive - #3548

Open
NathanColosimo wants to merge 12 commits into
codex/replay-sync-decodefrom
codex/replay-stream-preparation
Open

[core] Prepare replay payloads as event frames arrive#3548
NathanColosimo wants to merge 12 commits into
codex/replay-sync-decodefrom
codex/replay-stream-preparation

Conversation

@NathanColosimo

@NathanColosimo NathanColosimo commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

Summary

  • resolve the run key as soon as a deployment-bearing replay event arrives
  • buffer decoded event references until that one key resolution completes, then decrypt and decompress them in one pass without per-payload key promises
  • prepare subsequent payloads synchronously from the validated-frame callback, with natural response backpressure
  • keep one invocation-scoped cache of prepared payloads across fresh VMs and use a full event-array scan as a simple correctness safety net
  • reuse immutable primitive values for successful results, errors, hooks, and aborts while rehydrating object graphs per VM
  • let ordered consumers await only the genuinely asynchronous payload formats they touch

Performance signal

The latest benchmark and previous behavior run were recorded on the original cumulative branch before this review split. Across those runs, inline STSO improved 4–20% at p75, 10–24% at p90, and 39–51% at p99; total 1020-step workflow overhead improved 7–23%.

Cold TTFS regressed in both runs, including the turbo control that does not preload or prepare replay events, so that signal is dominated by deployment cold-start variability. The benchmarked branch also contained the small tracing layer in #3523.

Stack

  1. [world-vercel] Resume partial replay streams after validated events #3546 — replay stream recovery
  2. [core] Decode replay payloads synchronously #3547 — synchronous replay decode
  3. [core] Prepare replay payloads as event frames arrive #3548 — streamed replay preparation
  4. [core] Trace fresh workflow replay phases #3523 — replay phase tracing
  5. [core] Prune schema modules from workflow bundles #3550 — prune sandbox dependency graph
  6. [workflow] Load per-source replay bundles #3551 — per-source lazy VM bundles

Validation

  • core typecheck passed
  • full core suite: 2,112 tests passed, with 3 expected failures
  • focused replay cache, key helper, delivery-ordering, hook, and abort suites: 198 tests passed

@changeset-bot

changeset-bot Bot commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown

🦋 Changeset detected

Latest commit: f0cf664

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

This PR includes changesets to release 20 packages
Name Type
@workflow/core Patch
@workflow/world Patch
@workflow/world-vercel 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/world-local Patch
@workflow/world-postgres 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

vercel Bot commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

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

Project Deployment Actions Updated (UTC)
example-nextjs-workflow-turbopack Ready Ready Preview Aug 21, 2026 10:54am
example-nextjs-workflow-webpack Ready Ready Preview Aug 21, 2026 10:54am
example-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-astro-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-express-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-fastify-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-hono-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-nestjs-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-nitro-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-nuxt-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-python-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-sveltekit-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-tanstack-start-workflow Ready Ready Preview Aug 21, 2026 10:54am
workbench-vite-workflow Ready Ready Preview Aug 21, 2026 10:54am
workflow-docs Ready Ready Preview, v0 Aug 21, 2026 10:54am
workflow-swc-playground Ready Ready Preview Aug 21, 2026 10:54am
workflow-tarballs Ready Ready Preview Aug 21, 2026 10:54am
workflow-web Ready Ready Preview Aug 21, 2026 10:54am

@github-actions

github-actions Bot commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

🧪 E2E Test Results

❌ Some tests failed

❌ Failed E2E Tests

▲ Vercel Production (11 failed)

astro-quickjs (1 failed):

  • concurrent hook token conflict - two workflows cannot use the same hook token simultaneously | wrun_41M0HZF92J0GK2H5Z624W84T79 | 🔍 observability

example-node (1 failed):

example-quickjs (1 failed):

  • error handling serialization failures step-argument serialization failure is catchable in workflow code

express-node (1 failed):

  • AbortController abortHookOrderingWorkflow [listener-first-hook-first]: addEventListener → hook.then → resumeHook → abort()

express-quickjs (1 failed):

  • error handling serialization failures uncaught step-return-value serialization failure fails the run as USER_ERROR

hono-node (2 failed):

  • error handling retry behavior FatalError fails immediately without retries
  • hookGetConflictWorkflow - hook.getConflict() resolves with the conflicting run when token is already registered | wrun_41M0HZEGQZ0GQXP3N8GZDYQKGQ | 🔍 observability

nitro-node (1 failed):

  • AbortController abortDeterministicBranchFromStepWorkflow: branches stay consistent when abort comes from a step

nuxt-node (1 failed):

  • hookClaimOnlyMutexWorkflow - hook works as a pure run mutex without payload data | wrun_41M0HZF6KS0GMCQ57REM2Z4V18 | 🔍 observability

vite-node (2 failed):

  • AbortController abortSurvivesReplayWorkflow: controller state consistent across replay
  • AbortController abortHookOrderingWorkflow [hook-first-abort-first]: hook.then → addEventListener → abort() → resumeHook

⚠️ 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.

77 flaky tests
  • addTenWorkflow (fastify) — flaked in 2 jobs
  • 'hookGetConflictWithParallelStepWorkfl…' - hook.getConflict() does not block step execution (fastify)
  • 'hookGetConflictWithPriorStepWorkflow' - hook.getConflict() does not block step execution (nextjs-webpack)
  • abortExternalSignalInFlightWorkflow: external abort fires mid-flight, propagates to nested steps (nest)
  • abortExternalSignalInFlightWorkflow: external abort fires mid-flight, propagates to nested steps (nextjs-webpack)
  • abortExternalSignalWorkflow: signal passed as workflow input (nextjs-webpack)
  • abortListenerWorkflow: signal.addEventListener fires on the deserialized step signal (express)
  • abortParallelWorkflow: abort cancels all parallel steps (astro)
  • abortReasonTypesWorkflow: various abort reason types propagate correctly (express)
  • abortReasonTypesWorkflow: various abort reason types propagate correctly (sveltekit)
  • abortSurvivesReplayWorkflow: controller state consistent across replay (nuxt)
  • abortThrowIfAbortedWorkflow: throwIfAborted causes FatalError, no retries (fastify)
  • abortViaHookWorkflow: external hook triggers abort on in-flight step (fastify)
  • abortVoidSleepTimeoutWorkflow: documented void sleep().then(abort) pattern works (nuxt)
  • abortVoidSleepTimeoutWorkflow: documented void sleep().then(abort) pattern works (sveltekit)
  • addTenWorkflow (sveltekit)
  • AllInOneService.processNumber - static workflow method using sibling static step methods (vite)
  • basic step error preserves message and stack trace (hono)
  • Calculator.calculate - static workflow method using static step methods from another class (sveltekit)
  • cancelRun via CLI - cancelling a running workflow (sveltekit)
  • ChainableService.processWithThis - static step methods using this to reference the class (example)
  • ChainableService.processWithThis - static step methods using this to reference the class (vite)
  • completes but callbacks are not called (GAP) (sveltekit)
  • concurrent hook token conflict - two workflows cannot use the same hook token simultaneously (hono)
  • concurrent hook token conflict - two workflows cannot use the same hook token simultaneously (tanstack-start)
  • concurrent hook token conflict - two workflows cannot use the same hook token simultaneously (vite)
  • cross-file imports preserve message and stack trace (tanstack-start)
  • customSerializationWorkflow - custom class serialization with WORKFLOW_SERIALIZE/WORKFLOW_DESERIALIZE (nextjs-webpack)
  • distributedAbortController - TTL expiration triggers signal (fastify)
  • FatalError can be caught and detected with FatalError.is() (hono)
  • FatalError can be caught and detected with FatalError.is() (vite)
  • fire-and-forget: void setAttributes lands without awaiting (fastify)
  • getChunks returns same content as reading the stream (vite)
  • getterStepWorkflow - getter functions with "use step" directive (nitro)
  • hookAdoptOwnerResultWorkflow - duplicate adopts the owner result via conflict.returnValue (express)
  • hookClaimOnlyMutexWorkflow - hook works as a pure run mutex without payload data (nuxt)
  • hookCleanupTestWorkflow - hook token reuse after workflow completion (astro)
  • hookCleanupTestWorkflow - hook token reuse after workflow completion (nitro)
  • hookCleanupTestWorkflow - hook token reuse after workflow completion (tanstack-start)
  • hookDisposeTestWorkflow - hook token reuse after explicit disposal while workflow still running (hono)
  • hookGetConflictWorkflow - awaiting hook.getConflict() registers hook without payload (example)
  • hookSignalOwnerWorkflow - duplicate forwards its payload to the owner via resumeHook (hono)
  • hookSupersedeOwnerWorkflow - duplicate cancels the owner and claims the released token (astro)
  • hookSupersedeOwnerWorkflow - duplicate cancels the owner and claims the released token (hono)
  • hookSupersedeOwnerWorkflow - duplicate cancels the owner and claims the released token (nextjs-webpack)
  • hookWorkflow (astro)
  • hookWorkflow (fastify)
  • hookWorkflow (hono)
  • hookWorkflow (nest)
  • hookWorkflow (nitro)
  • maxRetries=0 disables retries (hono)
  • no startIndex (reads all chunks) (fastify)
  • nullByteWorkflow (sveltekit)
  • parallelStepsThenWebhookWorkflow - no hook_conflict from same-tick replay race (hono)
  • pathsAliasWorkflow - TypeScript path aliases resolve correctly (nextjs-turbopack)
  • pathsAliasWorkflow - TypeScript path aliases resolve correctly (sveltekit)
  • positive startIndex (skips first chunk) (nuxt)
  • promiseRaceStressTestWorkflow (vite)
  • resume-or-start route pattern - resumeHook retried after start() reaches the new run (fastify)
  • resume-or-start route pattern - resumeHook retried after start() reaches the new run (hono)
  • resume-or-start route pattern - resumeHook retried after start() reaches the new run (tanstack-start)
  • setAttributesInsideStepWorkflow: step-body calls append attributed native events (vite)
  • sleepWinsRaceWorkflow (astro)
  • sleepWinsRaceWorkflow (hono)
  • sleepWinsRaceWorkflow (vite)
  • step throw of a non-Error value preserves it as cause on the wrapping FatalError (nitro)
  • step-argument serialization failure is catchable in workflow code (express)
  • step-argument serialization failure is catchable in workflow code (fastify)
  • step-return-value serialization failure is catchable in workflow code (astro)
  • stepFunctionPassingWorkflow - step function references can be passed as arguments (without closure vars) (astro)
  • stepFunctionPassingWorkflow - step function references can be passed as arguments (without closure vars) (fastify)
  • stepFunctionWithClosureWorkflow - step function with closure variables passed as argument (nest)
  • stepWinsRaceWorkflow (nest)
  • thisSerializationWorkflow - step function invoked with .call() and .apply() (astro)
  • uncaught step-return-value serialization failure fails the run as USER_ERROR (vite)
  • workflow throw of a non-Error value round-trips verbatim as cause (nuxt)
  • WorkflowNotRegisteredError fails the run when workflow does not exist (nextjs-webpack)

🛠 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.

  • cold-start-warmup · suite warmup (tanstack-start) · at 10:56:36Z · abandoned wrun_01M0HZDG2KZR4J2Q290ECWS0S4
  • run-pickup-stall · resume-or-start route pattern - resumeHook retried after start() reaches the new run (hono) · at 10:59:48Z · abandoned wrun_41M0HZGS400GG8NEM51PEHHXZ2

E2E Test Summary

Summary
Passed Failed Skipped Total
❌ ▲ Vercel Production 3567 11 742 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 9 0 132 141
✅ vercel-multi-region 27 0 0 27
Total 15689 11 2548 18248
Details by Category

❌ ▲ Vercel Production

App Passed Failed Skipped
✅ astro-node 132 0 28
❌ astro-quickjs 131 1 28
❌ example-node 131 1 28
❌ example-quickjs 131 1 28
❌ express-node 131 1 28
❌ express-quickjs 131 1 28
✅ fastify-node 132 0 28
✅ fastify-quickjs 132 0 28
❌ hono-node 130 2 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 131 1 28
✅ nitro-quickjs 132 0 28
❌ nuxt-node 131 1 28
✅ nuxt-quickjs 132 0 28
✅ python-node 8 0 152
✅ sveltekit-node 151 0 9
✅ sveltekit-quickjs 151 0 9
✅ tanstack-start-node 132 0 28
✅ tanstack-start-quickjs 132 0 28
❌ vite-node 130 2 28
✅ vite-quickjs 132 0 28

✅ 💻 Local Development

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

✅ 📦 Local Production

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

✅ 🐘 Local Postgres

App Passed Failed Skipped
✅ astro-stable-node 134 0 26
✅ astro-stable-quickjs 134 0 26
✅ express-stable-node 134 0 26
✅ express-stable-quickjs 134 0 26
✅ fastify-stable-node 134 0 26
✅ fastify-stable-quickjs 134 0 26
✅ hono-stable-node 134 0 26
✅ hono-stable-quickjs 134 0 26
✅ nest-stable-node 134 0 26
✅ nest-stable-quickjs 134 0 26
✅ nextjs-turbopack-canary-node 141 0 19
✅ nextjs-turbopack-canary-quickjs 141 0 19
✅ nextjs-turbopack-stable-node 160 0 0
✅ nextjs-turbopack-stable-quickjs 160 0 0
✅ nextjs-webpack-canary-node 141 0 19
✅ nextjs-webpack-canary-quickjs 141 0 19
✅ nextjs-webpack-stable-node 160 0 0
✅ nextjs-webpack-stable-quickjs 160 0 0
✅ nitro-stable-node 134 0 26
✅ nitro-stable-quickjs 134 0 26
✅ nuxt-stable-node 134 0 26
✅ nuxt-stable-quickjs 134 0 26
✅ sveltekit-stable-node 153 0 7
✅ sveltekit-stable-quickjs 153 0 7
✅ tanstack-start-node 134 0 26
✅ tanstack-start-quickjs 134 0 26
✅ vite-stable-node 134 0 26
✅ vite-stable-quickjs 134 0 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 9 0 132

✅ vercel-multi-region

App Passed Failed Skipped
✅ nextjs-turbopack 27 0 0

📋 View full workflow run

@NathanColosimo NathanColosimo changed the title Prepare replay payloads during event streaming [core] Prepare replay payloads as event frames arrive Aug 14, 2026
This was referenced Aug 14, 2026
Open
Open
Open
@NathanColosimo
NathanColosimo force-pushed the codex/replay-stream-preparation branch from 388b6ea to 294d4b7 Compare August 14, 2026 03:56
@github-actions

github-actions Bot commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

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

Full trace: world-sim.txt

@github-actions

github-actions Bot commented Aug 14, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

📊 Workflow Benchmarks

commit f0cf664 · Fri, 21 Aug 2026 11:12:49 GMT · run logs

Backend: vercel · app: nextjs-turbopack

Metric Scenario Best (ms) P75 (ms) P90 (ms) P99 (ms) Samples
TTFS step 374 (-45%) 💚 1378 🔴 (+34%) 🔻 1410 🔴 (+27%) 🔻 1431 🔴 (-4.9%) 30
TTFS stream 949 (+245%) 🔻 1404 🔴 (+29%) 🔻 1419 🔴 (+29%) 🔻 1428 🔴 (+26%) 🔻 30
TTFS hook + stream 546 (-57%) 💚 1663 🔴 (+16%) 🔻 1745 🔴 (+4.9%) 1968 🔴 (-58%) 💚 30
Fan-out TTFS Promise.all(100 steps) 478 (+9.9%) 715 (+15%) 🔻 2093 (+211%) 🔻 2242 (+15%) 10
Fan-out TTLS Promise.all(100 steps) 4533 (+3.4%) 5379 (-28%) 💚 6801 (-11%) 12761 (+45%) 🔻 10
STSO 1020 steps (inline) 119 (-6.3%) 219 (-10%) 250 (-12%) 356 (-22%) 💚 1019
WO 1020 steps 213147 (-8.0%) 213147 (-8.0%) 213147 (-8.0%) 213147 (-8.0%) 1
CRTT first chunk (pooled) 92 (+4.5%) 138 (+7.0%) 209 (+39%) 🔻 294 (-27%) 💚 28

Streams

Scenario CRTT 1st p75 p90 p99 CDV max iters
paced control (100/s, 60B) 110 (+3%) 114 (-7%) 132 (-38%) 356 (-30%) 76.5 (-39%) 10
size sweep (100/s, 160B-12KB) 116 (+12%) 117 (-11%) 221 (±0%) 356 (-34%) 106 (-36%) 10
replay gateway-gpt-5.4-nano-2000t (1x) 155 (+14%) 108 (-21%) 132 (-35%) 207 (-63%) 135 (-6%) 3
replay eve-gpt-5.6-sol-2000t (1x) 127 (-50%) 155 (-37%) 265 (-81%) 430 (-90%) 373 (-84%) 2
replay eve-gpt-5.6-sol-2000t (2x) 137 (-9%) 187 (+5%) 377 (+21%) 715 (-27%) 452 (-22%) 3
📈 STSO distribution vs main (inline / queue-hop histograms)

1020 steps (inline)

Cumulative STSO time: main 231540ms → this run 212860ms (Δ -18680ms, -8%)

  100-150 ms  ┃                         main   6  this  26   +20
  150-200 ms  █████████████████░░░░░░┃  main 344  this 495  +151
  200-250 ms  ██████████████████┃███    main 451  this 394   -57
  250-300 ms  ███┃███                   main 144  this  73   -71
  300-350 ms  ┃█                        main  45  this  20   -25
  350-400 ms  ┃                         main   7  this   3    -4
  400-450 ms  ┃                         main  10  this   4    -6
  450-500 ms  ┃                         main   5  this   1    -4
  500-550 ms  ┃                         main   1  this   0    -1
  550-600 ms  ┃                         main   2  this   1    -1
  600-650 ms  ┃                         main   1  this   0    -1
  650-700 ms  ┃                         main   0  this   1    +1
  700-750 ms  ┃                         main   2  this   0    -2
  800-850 ms  ┃                         main   1  this   0    -1
1250-1300 ms  ┃                         main   0  this   1    +1
📈 CRTT drill-down vs main (RTT distributions & profiles) 
variant  RTT 1ms→5s+             avg         p50         p90         p99     n
control  ······█▅▁····   98.7 (-10%)    94 (-4%)  132 (-38%)  356 (-30%)  3000
sweep    ·····▁█▅▁····  102.2 (-14%)   93 (-11%)   221 (±0%)  356 (-34%)  3000
gw 1x    ·····▁█▃▁····   90.3 (-20%)   86 (-13%)  132 (-35%)  207 (-63%)  5295
eve 1x   ·····▁█▆▂▁···  118.2 (-62%)   92 (-12%)  265 (-81%)  430 (-90%)  5186
eve 2x   ·····▁██▃▁···   145.8 (-5%)  107 (-16%)  377 (+21%)  715 (-27%)  7779

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

control  █▂▁▁▂▇▇▄▃▂  90–111ms
sweep    █▂▁██▅▂▄▁▂  87–121ms
gw 1x    █▄▆▇▅▇▄▁▄▄  85–94ms
eve 1x   █▂▁▄▅▆▅▂▁▂  92–163ms
eve 2x   ▅▂▁▁▁█▅▄▃▂  99–240ms

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

sweep  ▇█▆▄▁▂▁  99–107ms

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

control  ▁▂▄▄▂█▂▄▆▇  25–33ms
sweep    ▃▄▂█▃▄▂▄▁▂  34–50ms
gw 1x    ▂▅▄▃█▅█▂▁▅  23–28ms
eve 1x   █▂▁▃██▆▄▃▃  21–36ms
eve 2x   ▄▃▁▂▂█▆▂▄▄  18–42ms
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.

@NathanColosimo
NathanColosimo force-pushed the codex/replay-stream-preparation branch from 7471e77 to 184d506 Compare August 14, 2026 20:01
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@NathanColosimo