coordinator auto-poll streaming conversion:dispatch 後自動 ingest viewer_url - #98
Merged
monkey1sai merged 1 commit intoMay 22, 2026
Conversation
…_url(coordinator-auto-poll-streaming-conversion)
完整 OpenSpec change(scaffold + apply 同 commit)。
## Why
fast-mvp loop 三段 archive 收尾後驗證,`POST /api/external/ifc-ready` → coordinator dispatch → streaming-server 真實轉檔 succeeded(40s)— **但 coordinator 端 `conversion_status` 永遠停在 `queued`**:既有 spec `Coordinator ingests host-native conversion result into callback outbox` 已寫「SHALL ingest through polling, an internal result loop, or an equivalent internal callback」,但 polling / internal result loop 都未實作。`fetchConversionResult` 只在 manual `POST /api/internal/conversions/<id>/ingest` 端點被呼叫;沒有自動 caller,viewer_url 永不出現。
## What changed(10 files / +968 / -50)
### Coordinator src(3 files)
- `services/streamingConversionClient.ts`:加 module-level `isTerminalConversionResult` helper(terminal 判定抽出來避免雙處硬編)+ class method `pollConversionResult(jobId, options): PollerHandle`:setTimeout chain(避免 setInterval overlap)、cancel()、max attempts → poll_timeout 走 fake failed result 餵 onTerminal
- `config.ts`:加 `conversionPollEnabled` / `conversionPollIntervalSeconds`(default 5)/ `conversionPollMaxAttempts`(default 60 = 5 min)+ `parseBooleanEnv` helper
- `app.ts`:
- `CoordinatorApp` interface 加 `dispose: () => void`
- 加 module-scope `pollerRegistry: Map<conversion_job_id, PollerHandle>`
- refactor 既有 manual ingest handler 內邏輯抽成共用 helper `ingestStreamingConversionResult(jobId, { result?, source })`(manual + auto-poll 共用)
- 加 `schedulePollerForConversion(jobId)`:closure wrapper for `streamingConversionClient.pollConversionResult` + onTerminal=ingest helper + registry cleanup
- dispatch path:`markDispatched` 之後若 `config.conversionPollEnabled && !pollerRegistry.has(id)` → `schedulePollerForConversion(id)`
- manual ingest handler 開頭 cancel + delete registry(避免雙 ingest)
- return 加 `dispose`:遍歷 registry cancel 所有 timer
### Coordinator tests(2 files)
- `unit_kitpool.test.ts`:fixture 加 3 個新 config field(default `false` 避免 unit test 啟 timer)
- `auto-poll-conversion.test.ts`(新):5 case
- dispatch → poller queued/queued/ready 序列 → 自動 ingest 出 viewer_url
- dispatch → poller queued/failed → 自動 ingest 為 failed,viewer_url 不出現
- 重複 idempotent dispatch 不雙起 poller(stub /result 呼叫次數有 budget)
- manual ingest 觸發 cancel auto poller(stub /result 只被打 1 次)
- `conversionPollEnabled: false` fixture 不啟 poller
### OpenSpec change(5 files)
- proposal / design / tasks / acceptance / `conversion-webhook-lifecycle` MODIFIED requirement + 3 新 Scenario
## GitNexus blast radius = LOW
| symbol | risk | d=1 |
|---|---|---|
| `fetchConversionResult` | LOW | `createCoordinatorApp`(同 module) |
| `createCoordinatorApp` | LOW | 無 upstream(main entry) |
| `loadConfig` | LOW | (PR #94 同樣 path,LOW) |
## Verification
| Level | Result |
|---|---|
| L1 coordinator `npm run verify` | 12 files / **173 tests passed**(168 既有 + 5 新 auto-poll) |
| L1 streaming-server `pytest tests -q` | **31 passed**(不動,regression OK) |
| L1 root `pytest tests` | **9 passed** |
| L2 `openspec validate coordinator-auto-poll-streaming-conversion --strict` | **valid** |
| L2 `openspec validate --specs --strict` | **26 passed / 0 failed** |
| L3 GitNexus pre-impact | LOW for `fetchConversionResult` / `createCoordinatorApp` |
| **L4 真實 runtime** | **跳過** — 留 merge 後 docker compose recreate coordinator(讀新 code)+ 跑 Postman ① + 等(不手動 POST ingest)→ ② Poll 預期 5-90 秒內 `viewer_url` 自動出現 |
## Predecessor / Follow-up
✓ Predecessor:`streaming-server-prefer-local-ifc-path`(PR #96 / archive PR #97)+ PR #94 / PR #95 hotfix bundle
本 change 是 fast-mvp loop 自動化的最後一片拼圖,merge 後外部 caller 不再需要任何手動 trigger 就能拿到 viewer_url。
不解 / 排除:
- 不持久化 poller state(coordinator restart in-memory timers lost;手動 endpoint 仍可救)
- 不解 cloud callback outbox retry / dead-letter(另一條既有 capability,本 change 不動)
- 不引入 streaming-server push callback(本 change 走 coordinator pull)
- 不引入第三方 scheduler library
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Contributor
📝 WalkthroughWalkthroughThis PR implements coordinator-side auto-polling for streaming IFC→USDC conversion results. After a successful dispatch, the coordinator automatically polls ChangesCoordinator Auto-Poll Streaming Conversion
Sequence DiagramsequenceDiagram
participant Client
participant Coordinator
participant StreamingServer as bim-streaming-server
participant Outbox
Client->>Coordinator: POST /api/external/ifc-ready
Coordinator->>StreamingServer: POST /api/conversions/ifc-to-usdc (dispatch)
StreamingServer-->>Coordinator: 202 {conversion_job_id}
Coordinator->>Coordinator: Schedule auto-poller (setTimeout chain)
loop Poll until terminal (max 60 attempts @ 5s)
Coordinator->>StreamingServer: GET /api/conversions/{id}/result
alt Not ready
StreamingServer-->>Coordinator: {status: "queued"}
Coordinator->>Coordinator: Schedule next tick
else Ready
StreamingServer-->>Coordinator: {status: "ready", model_status: "ready"}
Coordinator->>Coordinator: ingestStreamingConversionResult
Coordinator->>Outbox: Enqueue conversion_result_ready callback
Coordinator->>Coordinator: Cancel poller, clear registry
else Failed
StreamingServer-->>Coordinator: {status: "failed"}
Coordinator->>Outbox: Enqueue conversion_failed callback
Coordinator->>Coordinator: Cancel poller, clear registry
end
end
alt Manual ingest before poller completes
Client->>Coordinator: POST /api/internal/conversions/{id}/ingest
Coordinator->>Coordinator: Cancel & delete active poller
Coordinator->>Outbox: Enqueue callback once
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~22 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 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 |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (5)
bim-review-coordinator/tests/auto-poll-conversion.test.ts (1)
29-40: ⚡ Quick winClean up per-test temporary directories created in
makeApp.Line 130 creates a new temp root per test, but teardown never removes it. Over repeated local/CI runs this can accumulate stale data under tmp.
♻️ Proposed fix
let active: CoordinatorApp | null = null; let activeStub: http.Server | null = null; +const tempRoots: string[] = []; afterEach(async () => { if (active) active.dispose(); @@ if (active) { active.io.close(); await new Promise<void>((resolve) => active?.server.close(() => resolve())); active = null; } + for (const root of tempRoots.splice(0)) { + fs.rmSync(root, { recursive: true, force: true }); + } }); @@ function makeApp(streamingBase: string, overrides: Partial<CoordinatorConfig> = {}): CoordinatorApp { const root = fs.mkdtempSync(path.join(os.tmpdir(), "bim-coord-auto-poll-test-")); + tempRoots.push(root); active = createCoordinatorApp({Also applies to: 129-147
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@bim-review-coordinator/tests/auto-poll-conversion.test.ts` around lines 29 - 40, The test teardown currently disposes network stubs (active, activeStub) but forgets to remove the per-test temporary root created by makeApp, causing stale tmp accumulation; update the afterEach in auto-poll-conversion.test.ts to capture the temp root returned by makeApp (or access the tempRoot property on the app instance) and remove it during teardown using fs.promises.rm or fs.rmSync with recursive/force; ensure you reference the same identifier used when creating the app (e.g., tempRoot or app.tempRoot) and call await fs.promises.rm(tempRoot, { recursive: true, force: true }) (or equivalent) before nulling the variable so every test cleans up its temp directory.openspec/changes/coordinator-auto-poll-streaming-conversion/design.md (2)
7-13: ⚡ Quick winAdd language identifier to fenced code block.
The code block lacks a language identifier. Consider adding
```textto improve markdown parsing and rendering consistency.📝 Proposed fix
-``` +```text [POST /api/external/ifc-ready] → coordinator dispatch streaming-server `POST /api/conversions/ifc-to-usdc`🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@openspec/changes/coordinator-auto-poll-streaming-conversion/design.md` around lines 7 - 13, The fenced code block that begins with "[POST /api/external/ifc-ready]" in design.md lacks a language identifier; update the opening triple-backtick to include a language (e.g., replace ``` with ```text) so the block becomes ```text and preserves formatting/rendering consistently for the sequence showing coordinator → streaming-server → conversion_job_id/status flow.
27-72: ⚡ Quick winAdd language identifier to fenced code block.
This large component layout block should have a language identifier for better markdown rendering. Consider
```textor```typescriptfor the TypeScript portions.📝 Proposed fix
-``` +```text streamingConversionClient: - fetchConversionResult(id) (既有,不改)🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@openspec/changes/coordinator-auto-poll-streaming-conversion/design.md` around lines 27 - 72, The fenced code block starting with "streamingConversionClient:" is missing a language identifier; update that code fence in design.md to include a language (e.g., ```text or ```typescript) so markdown renders it correctly — locate the block beginning with streamingConversionClient and the closing ``` and change the opening fence from ``` to something like ```text (or ```typescript if you prefer) to apply the identifier.openspec/changes/coordinator-auto-poll-streaming-conversion/tasks.md (2)
71-71: 💤 Low valueClarify test count reference.
Line 71 mentions "11 files + 新 case 全綠" while
acceptance.mdline 5 specifies "168 既有 + ~6 新 case 全綠". The "11 files" likely refers to test file count, not test case count, but this could be clarified for consistency with the acceptance plan.💡 Suggested clarification
-- [ ] 7.1 `cd bim-review-coordinator && npm run verify`(11 files + 新 case 全綠) +- [ ] 7.1 `cd bim-review-coordinator && npm run verify`(168 既有 + ~6 新 case 全綠)This aligns with the more specific count in
acceptance.md.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@openspec/changes/coordinator-auto-poll-streaming-conversion/tasks.md` at line 71, Update the checklist entry for 7.1 (`cd bim-review-coordinator && npm run verify`) to clarify that "11 files" refers to test files rather than test cases and align the wording with acceptance.md's test-case counts (e.g., change "11 files + 新 case 全綠" to "11 test files (covers ~6 new test cases) — all green" or similar); edit the line in tasks.md (the `- [ ] 7.1` checklist item) so the terminology matches acceptance.md's "168 既有 + ~6 新 case 全綠" for consistency.
84-84: ⚡ Quick winEnvironment-specific PID reference.
Line 84 references "PID 7488" which is specific to a particular runtime environment and will not be applicable to other environments. Consider rephrasing to indicate this is an example.
📝 Suggested clarification
-- [ ] 9.2 streaming-server 仍跑(PID 7488,STORAGE_ROOT absolute) +- [ ] 9.2 streaming-server 仍跑(例如 PID 7488,STORAGE_ROOT absolute)This makes it clear the PID is just an example from a specific test run.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@openspec/changes/coordinator-auto-poll-streaming-conversion/tasks.md` at line 84, Line 84 currently embeds an environment-specific PID ("PID 7488") in the checklist item "9.3 跑 Postman ① POST /api/external/ifc-ready"; remove the hardcoded PID and rephrase to indicate it is an example (e.g., "示例 PID（例如 7488）" or "PID shown is from a specific run and will vary by environment") so readers know it's not a required value; update the text in tasks.md for that checklist entry accordingly.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@bim-review-coordinator/src/config.ts`:
- Around line 229-230: The config currently assigns
conversionPollIntervalSeconds and conversionPollMaxAttempts from numberFromEnv
which may return 0 or negative values; clamp both assignments to a minimum of 1
so polling cannot be instant or disabled by misconfigured envs (e.g. replace the
raw calls with Math.max(1, numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5))
and Math.max(1, numberFromEnv("CONVERSION_POLL_MAX_ATTEMPTS", 60)) or implement
equivalent min logic inside numberFromEnv), referring to the
conversionPollIntervalSeconds and conversionPollMaxAttempts variables and the
numberFromEnv helper.
In `@openspec/changes/coordinator-auto-poll-streaming-conversion/design.md`:
- Around line 95-100: The design needs poll_timeout to be normalized to failed
everywhere: update the onTerminal handling so when onTerminal({ status:
"poll_timeout", conversion_job_id }) is triggered it calls
ingestStreamingConversionResult with a failed-equivalent report; ensure
externalIfcReadyStore.recordConversionOutcome(conversionStatus: "ready" |
"failed") is invoked with "failed" (not a new enum) for that case; make the
callback outbox enqueue event "conversion_failed" with payload.status="failed"
and payload.reason="poll_timeout" (so report.status remains failed and reason
carries the timeout), and confirm API surface code only exposes
conversion_status as "ready" or "failed" (no poll_timeout).
In
`@openspec/changes/coordinator-auto-poll-streaming-conversion/specs/conversion-webhook-lifecycle/spec.md`:
- Around line 30-49: Add a new test in
bim-review-coordinator/tests/auto-poll-conversion.test.ts that exercises the
"Poll timeout yields a failed-equivalent terminal state" scenario: simulate
dispatching a conversion via the coordinator (POST /api/external/ifc-ready) and
stub GET /api/conversions/<id>/result to return non-terminal responses for
conversionPollMaxAttempts (or CONVERSION_POLL_MAX_ATTEMPTS) cycles, assert the
auto-poller de-registers and the coordinator marks the conversion terminally
failed with reason "poll_timeout", and also assert the manual ingest endpoint
(/api/internal/conversions/<conversion_job_id>/ingest) MAY still be accepted
later (no duplicate callback/ingest produced); ensure assertions mirror existing
tests' patterns for ready→auto-ingest and failed→failed ingest to keep
idempotency and de-registration checks consistent with current helpers.
---
Nitpick comments:
In `@bim-review-coordinator/tests/auto-poll-conversion.test.ts`:
- Around line 29-40: The test teardown currently disposes network stubs (active,
activeStub) but forgets to remove the per-test temporary root created by
makeApp, causing stale tmp accumulation; update the afterEach in
auto-poll-conversion.test.ts to capture the temp root returned by makeApp (or
access the tempRoot property on the app instance) and remove it during teardown
using fs.promises.rm or fs.rmSync with recursive/force; ensure you reference the
same identifier used when creating the app (e.g., tempRoot or app.tempRoot) and
call await fs.promises.rm(tempRoot, { recursive: true, force: true }) (or
equivalent) before nulling the variable so every test cleans up its temp
directory.
In `@openspec/changes/coordinator-auto-poll-streaming-conversion/design.md`:
- Around line 7-13: The fenced code block that begins with "[POST
/api/external/ifc-ready]" in design.md lacks a language identifier; update the
opening triple-backtick to include a language (e.g., replace ``` with ```text)
so the block becomes ```text and preserves formatting/rendering consistently for
the sequence showing coordinator → streaming-server → conversion_job_id/status
flow.
- Around line 27-72: The fenced code block starting with
"streamingConversionClient:" is missing a language identifier; update that code
fence in design.md to include a language (e.g., ```text or ```typescript) so
markdown renders it correctly — locate the block beginning with
streamingConversionClient and the closing ``` and change the opening fence from
``` to something like ```text (or ```typescript if you prefer) to apply the
identifier.
In `@openspec/changes/coordinator-auto-poll-streaming-conversion/tasks.md`:
- Line 71: Update the checklist entry for 7.1 (`cd bim-review-coordinator && npm
run verify`) to clarify that "11 files" refers to test files rather than test
cases and align the wording with acceptance.md's test-case counts (e.g., change
"11 files + 新 case 全綠" to "11 test files (covers ~6 new test cases) — all green"
or similar); edit the line in tasks.md (the `- [ ] 7.1` checklist item) so the
terminology matches acceptance.md's "168 既有 + ~6 新 case 全綠" for consistency.
- Line 84: Line 84 currently embeds an environment-specific PID ("PID 7488") in
the checklist item "9.3 跑 Postman ① POST /api/external/ifc-ready"; remove the
hardcoded PID and rephrase to indicate it is an example (e.g., "示例 PID（例如 7488）"
or "PID shown is from a specific run and will vary by environment") so readers
know it's not a required value; update the text in tasks.md for that checklist
entry accordingly.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 32a4163c-e5c2-4261-9824-e28cdf493ac7
📒 Files selected for processing (10)
bim-review-coordinator/src/app.tsbim-review-coordinator/src/config.tsbim-review-coordinator/src/services/streamingConversionClient.tsbim-review-coordinator/tests/auto-poll-conversion.test.tsbim-review-coordinator/tests/unit_kitpool.test.tsopenspec/changes/coordinator-auto-poll-streaming-conversion/acceptance.mdopenspec/changes/coordinator-auto-poll-streaming-conversion/design.mdopenspec/changes/coordinator-auto-poll-streaming-conversion/proposal.mdopenspec/changes/coordinator-auto-poll-streaming-conversion/specs/conversion-webhook-lifecycle/spec.mdopenspec/changes/coordinator-auto-poll-streaming-conversion/tasks.md
Comment on lines
+229
to
+230
| conversionPollIntervalSeconds: numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5), | ||
| conversionPollMaxAttempts: numberFromEnv("CONVERSION_POLL_MAX_ATTEMPTS", 60), |
Contributor
There was a problem hiding this comment.
Guard polling interval/attempts against non-positive env values
At Line 229 and Line 230, numberFromEnv allows 0 or negative values. That can cause immediate/tight polling behavior or instant timeout semantics under misconfiguration. Please clamp both to a minimum of 1.
💡 Suggested patch
- conversionPollIntervalSeconds: numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5),
- conversionPollMaxAttempts: numberFromEnv("CONVERSION_POLL_MAX_ATTEMPTS", 60),
+ conversionPollIntervalSeconds: Math.max(1, numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5)),
+ conversionPollMaxAttempts: Math.max(1, numberFromEnv("CONVERSION_POLL_MAX_ATTEMPTS", 60)),📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| conversionPollIntervalSeconds: numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5), | |
| conversionPollMaxAttempts: numberFromEnv("CONVERSION_POLL_MAX_ATTEMPTS", 60), | |
| conversionPollIntervalSeconds: Math.max(1, numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5)), | |
| conversionPollMaxAttempts: Math.max(1, numberFromEnv("CONVERSION_POLL_MAX_ATTEMPTS", 60)), |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@bim-review-coordinator/src/config.ts` around lines 229 - 230, The config
currently assigns conversionPollIntervalSeconds and conversionPollMaxAttempts
from numberFromEnv which may return 0 or negative values; clamp both assignments
to a minimum of 1 so polling cannot be instant or disabled by misconfigured envs
(e.g. replace the raw calls with Math.max(1,
numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5)) and Math.max(1,
numberFromEnv("CONVERSION_POLL_MAX_ATTEMPTS", 60)) or implement equivalent min
logic inside numberFromEnv), referring to the conversionPollIntervalSeconds and
conversionPollMaxAttempts variables and the numberFromEnv helper.
Comment on lines
+95
to
+100
| `attempts >= maxAttempts` 觸發 `onTerminal({ status: "poll_timeout", conversion_job_id })`: | ||
|
|
||
| - ingestStreamingConversionResult 收到後視為 failed-equivalent | ||
| - externalIfcReadyStore 標 conversion_status `poll_timeout`(新 enum 值或映射 failed) | ||
| - callback outbox enqueue `conversion_failed` with reason="poll_timeout" | ||
| - 不自動重啟 poller(留給後續 ops decision;手動 endpoint 仍可救) |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Description: Check for poll_timeout status handling in stores and schemas
# Search for conversion_status enum definitions
rg -nP -C5 'conversion_status|ConversionStatus' \
bim-review-coordinator/src/ \
-g '!*.test.{ts,js}'
# Search for poll_timeout references
rg -nP -C5 'poll_timeout' \
bim-review-coordinator/src/
# Search for callback outbox event type mapping
rg -nP -C8 'conversion_failed|enqueue' \
bim-review-coordinator/src/ \
-g '!*.test.{ts,js}'Repository: monkey1sai/AI-BIM-governance
Length of output: 50384
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Focused search for poll_timeout wiring
rg -n --hidden --no-ignore -C3 'poll_timeout' bim-review-coordinator/src/app.ts bim-review-coordinator/src/services bim-review-coordinator/src/public/dev-console.html || true
# Confirm what externalIfcReadyStore allows for conversion_status
rg -n 'recordConversionOutcome|conversionStatus:|conversion_status' bim-review-coordinator/src/services/externalIfcReadyStore.ts bim-review-coordinator/src/types.ts bim-review-coordinator/src/app.ts || true
# Confirm callbackOutbox event mapping for failures (including reason)
rg -n 'enqueue\(|conversion_failed|reason:' bim-review-coordinator/src/app.ts bim-review-coordinator/src/services/callbackOutbox.ts || true
# Show the exact onTerminal fake timeout payload mapping
rg -n -C3 'fakeTimeoutResult|onTerminal\(fakeTimeoutResult\)|model_status: "failed"|reason: "poll_timeout"' bim-review-coordinator/src/services/streamingConversionClient.ts || true
# Show what API returns for conversions/:id and external/ifc-ready/:jobId
rg -n -C3 '/api/dev/conversions|/api/internal/conversion-result|/api/external/ifc-ready|conversions/:id/ingest|conversion_status' bim-review-coordinator/src/app.ts || trueRepository: monkey1sai/AI-BIM-governance
Length of output: 16254
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Print relevant sections (small ranges) to see reason->callback payload mapping
sed -n '840,930p' bim-review-coordinator/src/app.ts | nl -ba | head -n 120
echo "----"
sed -n '1,220p' bim-review-coordinator/src/services/callbackOutbox.ts | nl -ba | sed -n '1,220p' | head -n 220
echo "----"
sed -n '150,220p' bim-review-coordinator/src/services/externalIfcReadyStore.ts | nl -ba | head -n 120
echo "----"
sed -n '200,260p' bim-review-coordinator/src/services/streamingConversionClient.ts | nl -ba | head -n 120Repository: monkey1sai/AI-BIM-governance
Length of output: 115
🏁 Script executed:
#!/bin/bash
set -euo pipefail
echo "---- app.ts (callback outbox enqueue payload area)"
sed -n '830,930p' bim-review-coordinator/src/app.ts | cat -n
echo "---- callbackOutbox.ts (enqueue + payload shape)"
sed -n '1,240p' bim-review-coordinator/src/services/callbackOutbox.ts | cat -n
echo "---- externalIfcReadyStore.ts (recordConversionOutcome contract)"
sed -n '140,220p' bim-review-coordinator/src/services/externalIfcReadyStore.ts | cat -n
echo "---- streamingConversionClient.ts (fake poll_timeout terminal result)"
sed -n '200,280p' bim-review-coordinator/src/services/streamingConversionClient.ts | cat -nRepository: monkey1sai/AI-BIM-governance
Length of output: 20829
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Show the ingestStreamingConversionResult failure handling and report construction
sed -n '760,840p' bim-review-coordinator/src/app.ts | cat -n
# Show normalizeConversionReportStatus + conversionResultReportSchema (closest small ranges)
rg -n 'normalizeConversionReportStatus|conversionResultReportSchema' bim-review-coordinator/src/app.ts
sed -n '520,640p' bim-review-coordinator/src/app.ts | cat -nRepository: monkey1sai/AI-BIM-governance
Length of output: 11111
Align design “poll_timeout” status with implementation (mapped to failed + callback reason).
externalIfcReadyStoreonly recordsconversion_statusas"ready" | "failed"(recordConversionOutcome(conversionStatus: "ready" | "failed")); the auto-pollpoll_timeoutterminal is normalized tostatus: "failed", soconversion_statusbecomes"failed"(not a distinctpoll_timeoutvalue).- The callback outbox enqueues
event: "conversion_failed"whenreport.statusis failed, and carriespayload.status: "failed"pluspayload.reason: "poll_timeout"(fake timeout setsreason: "poll_timeout", then ingest forwards it toreason). - API responses surface
conversion_statusas"ready"/"failed"(nopoll_timeoutstatus is exposed via the API);poll_timeoutappears only in the cloud callback payloadreason.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@openspec/changes/coordinator-auto-poll-streaming-conversion/design.md` around
lines 95 - 100, The design needs poll_timeout to be normalized to failed
everywhere: update the onTerminal handling so when onTerminal({ status:
"poll_timeout", conversion_job_id }) is triggered it calls
ingestStreamingConversionResult with a failed-equivalent report; ensure
externalIfcReadyStore.recordConversionOutcome(conversionStatus: "ready" |
"failed") is invoked with "failed" (not a new enum) for that case; make the
callback outbox enqueue event "conversion_failed" with payload.status="failed"
and payload.reason="poll_timeout" (so report.status remains failed and reason
carries the timeout), and confirm API surface code only exposes
conversion_status as "ready" or "failed" (no poll_timeout).
Comment on lines
+30
to
+49
| #### Scenario: Dispatch auto-schedules a poller that drives ingestion to terminal | ||
|
|
||
| - **WHEN** coordinator returns 202 from `POST /api/external/ifc-ready` with a dispatched `conversion_job_id` and `CONVERSION_POLL_ENABLED` is unset or `true` | ||
| - **THEN** coordinator schedules an in-process polling task keyed by that `conversion_job_id` | ||
| - **AND** the polling task fetches `GET /api/conversions/<id>/result` every `CONVERSION_POLL_INTERVAL_SECONDS` | ||
| - **AND** when the result reaches a terminal state the polling task triggers the same ingestion path as the manual endpoint(callback outbox + local session handoff per existing requirements) | ||
| - **AND** the polling task de-registers itself after terminal ingestion or after `CONVERSION_POLL_MAX_ATTEMPTS` poll-timeout | ||
|
|
||
| #### Scenario: Auto-poll de-duplicates with manual ingest endpoint | ||
|
|
||
| - **WHEN** a poller is active for `conversion_job_id` and the operator(or other internal caller)POSTs `/api/internal/conversions/<conversion_job_id>/ingest` with a valid internal token | ||
| - **THEN** coordinator cancels and de-registers the auto-poller for that `conversion_job_id` | ||
| - **AND** the manual ingest path runs exactly once | ||
| - **AND** no duplicate `conversion_result_ready` / `conversion_failed` callback is enqueued for the same `conversion_job_id` | ||
|
|
||
| #### Scenario: Poll timeout yields a failed-equivalent terminal state | ||
|
|
||
| - **WHEN** the auto-poller reaches `CONVERSION_POLL_MAX_ATTEMPTS` without observing a terminal result from `bim-streaming-server` | ||
| - **THEN** coordinator treats the local conversion job as terminally failed with reason `poll_timeout` | ||
| - **AND** the manual ingest endpoint MAY still be invoked later to re-ingest if the streaming-server eventually reaches terminal |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Description: Check that new test file covers all three scenarios
# Search for auto-poll test cases
rg -nP -C3 'test|it|describe' \
bim-review-coordinator/tests/auto-poll-conversion.test.ts
# Look for specific test coverage:
# - dispatch auto-schedules poller (ready path)
# - manual ingest cancels auto-poller
# - max attempts / poll timeout
rg -nP 'dispatch.*poll|auto.*poll.*start|schedule.*poll' \
bim-review-coordinator/tests/auto-poll-conversion.test.ts
rg -nP 'manual.*cancel|ingest.*cancel' \
bim-review-coordinator/tests/auto-poll-conversion.test.ts
rg -nP 'timeout|max.*attempt' \
bim-review-coordinator/tests/auto-poll-conversion.test.tsRepository: monkey1sai/AI-BIM-governance
Length of output: 11073
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# 1) Search for any explicit poll-timeout / max-attempts test coverage
rg -nP 'timeout|poll_timeout|maxAttempts|max attempts|conversionPollMaxAttempts|failed-equivalent' \
bim-review-coordinator/tests/auto-poll-conversion.test.ts || true
# 2) Quick check if there's any additional auto-poll test file that might cover timeout
fd -e 'ts' -a bim-review-coordinator/tests -t f . | true
fd -a -e 'test.ts' -e 'spec.ts' bim-review-coordinator/tests || true
# 3) Find tasks.md and show section 5 context
fd -e 'md' -a . | rg -n 'tasks\.md$' || true
TASKS_FILE=$(fd -e 'md' -a . | rg -m1 'tasks\.md$' || true)
echo "TASKS_FILE=$TASKS_FILE"
if [ -n "${TASKS_FILE:-}" ]; then
# Print around "section 5" (best-effort: look for "5." or "## 5" patterns)
rg -nP '(^|\n)#{1,6}\s*5\b|(^|\n)\b5\.\s|\bSection\s*5\b' "$TASKS_FILE" -n || true
# Show the first matching header section
rg -nP '(^|\n)#{1,6}\s*5\b|(^|\n)\b5\.\s|\bSection\s*5\b' "$TASKS_FILE" -n | head -n 5
# Show ~120 lines around the first match
LINE=$(rg -nP '(^|\n)#{1,6}\s*5\b|(^|\n)\b5\.\s|\bSection\s*5\b' "$TASKS_FILE" -n | head -n 1 | sed -E 's/^(.*):([0-9]+):.*$/\2/')
if [ -n "${LINE:-}" ]; then
start=$((LINE-40))
end=$((LINE+120))
sed -n "${start},${end}p" "$TASKS_FILE"
fi
fiRepository: monkey1sai/AI-BIM-governance
Length of output: 927
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# A) Locate the relevant test file(s)
fd -t f 'auto-poll' bim-review-coordinator/tests -a || true
fd -t f 'auto.*poll.*conversion' bim-review-coordinator/tests -a || true
fd -t f 'conversion.*poll' bim-review-coordinator/tests -a || true
# B) Search across all tests for poll-timeout / max-attempts related assertions
rg -nP 'poll_timeout|poll time(out)?|CONVERSION_POLL_MAX_ATTEMPTS|conversionPollMaxAttempts|max.*attempt|failed-equivalent|timeout' \
bim-review-coordinator/tests || true
# C) Locate tasks.md and print the section 5 area (best effort)
TASKS_MD=$(fd -t f -e md --full-path 'tasks\.md$' . | head -n 1 || true)
echo "TASKS_MD=$TASKS_MD"
if [ -n "${TASKS_MD:-}" ]; then
rg -nP '(^|\n)#{1,6}\s*5\b|(^|\n)\b5\.\s|Section\s*5' "$TASKS_MD" -n || true
# print around first match
MATCH_LINE=$(rg -nP '(^|\n)#{1,6}\s*5\b|(^|\n)\b5\.\s|Section\s*5' "$TASKS_MD" -n | head -n 1 | sed -E 's/^(.*):([0-9]+):.*$/\2/')
if [ -n "${MATCH_LINE:-}" ]; then
start=$((MATCH_LINE-60))
end=$((MATCH_LINE+220))
sed -n "${start},${end}p" "$TASKS_MD"
else
# fallback: show any "tasks" headings
head -n 220 "$TASKS_MD"
fi
fiRepository: monkey1sai/AI-BIM-governance
Length of output: 1565
🏁 Script executed:
#!/bin/bash
set -euo pipefail
TEST_FILE="bim-review-coordinator/tests/auto-poll-conversion.test.ts"
echo "== $TEST_FILE =="
wc -l "$TEST_FILE"
echo "--- tail ---"
tail -n 120 "$TEST_FILE"
echo "--- lines 250-380 (if exists) ---"
sed -n '240,380p' "$TEST_FILE" || true
TASKS_MD="./openspec/changes/archive/2026-05-07-add-dev-ifc-source-selection-flow/tasks.md"
echo "== $TASKS_MD =="
wc -l "$TASKS_MD"
# show section 5 content (from the header line to next header)
# first find line number of the header
SECTION_LINE=$(rg -nP '^##\s*5\.' "$TASKS_MD" | head -n 1 | cut -d: -f1 || true)
echo "SECTION_LINE=$SECTION_LINE"
if [ -n "${SECTION_LINE:-}" ]; then
# print from section line to next '## ' header
awk -v start="$SECTION_LINE" 'NR>=start{print} NR>start && /^##\s*[0-9]+(\.|$)/{exit}' "$TASKS_MD"
fiRepository: monkey1sai/AI-BIM-governance
Length of output: 8990
🏁 Script executed:
#!/bin/bash
set -euo pipefail
# Search across all coordinator tests for explicit timeout/max-attempts coverage
rg -nP 'poll_timeout|poll time(out)?|CONVERSION_POLL_MAX_ATTEMPTS|conversionPollMaxAttempts|max.*attempt' \
bim-review-coordinator/tests -SRepository: monkey1sai/AI-BIM-governance
Length of output: 379
🏁 Script executed:
#!/bin/bash
set -euo pipefail
TEST_FILE="bim-review-coordinator/tests/auto-poll-conversion.test.ts"
echo "== Head =="
sed -n '1,240p' "$TEST_FILE"
echo "== Search within file =="
rg -nP 'it\(|timeout|poll_timeout|maxAttempts|CONVERSION_POLL_MAX_ATTEMPTS|max attempts|failed-equivalent' "$TEST_FILE"Repository: monkey1sai/AI-BIM-governance
Length of output: 9690
Add test coverage for the poll-timeout (max-attempts) scenario
bim-review-coordinator/tests/auto-poll-conversion.test.tscovers: ready → auto-ingest (incl. viewer_url/callback), failed → failed ingest (no viewer_url), idempotent duplicate dispatch (no double polling), and manual ingest canceling the auto-poller (no double ingest).- The spec scenario “Poll timeout yields a failed-equivalent terminal state” is not actually tested: there’s only a comment mentioning
poll_timeout, but no test forces the auto-poller to hitconversionPollMaxAttemptswithout reaching a terminal result. tasks.md“## 5.” in the located file is unrelated to coordinator auto-poll conversion test coverage, so it doesn’t support the mapping.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In
`@openspec/changes/coordinator-auto-poll-streaming-conversion/specs/conversion-webhook-lifecycle/spec.md`
around lines 30 - 49, Add a new test in
bim-review-coordinator/tests/auto-poll-conversion.test.ts that exercises the
"Poll timeout yields a failed-equivalent terminal state" scenario: simulate
dispatching a conversion via the coordinator (POST /api/external/ifc-ready) and
stub GET /api/conversions/<id>/result to return non-terminal responses for
conversionPollMaxAttempts (or CONVERSION_POLL_MAX_ATTEMPTS) cycles, assert the
auto-poller de-registers and the coordinator marks the conversion terminally
failed with reason "poll_timeout", and also assert the manual ingest endpoint
(/api/internal/conversions/<conversion_job_id>/ingest) MAY still be accepted
later (no duplicate callback/ingest produced); ensure assertions mirror existing
tests' patterns for ready→auto-ingest and failed→failed ingest to keep
idempotency and de-registration checks consistent with current helpers.
Contributor
There was a problem hiding this comment.
Pull request overview
This PR completes the coordinator-side “internal result loop” gap by adding an in-process auto-poller that starts after a successful streaming conversion dispatch and automatically ingests terminal conversion results to produce viewer_url without requiring a manual /api/internal/conversions/:id/ingest call.
Changes:
- Add
pollConversionResult+ shared terminal-detection helper inStreamingConversionClient. - Add polling config knobs (
CONVERSION_POLL_ENABLED, interval, max attempts) and integrate an auto-poller registry + shared ingest helper intocreateCoordinatorApp. - Add Vitest coverage for auto-poll happy/failed paths, dedupe behavior, and manual-ingest cancellation; plus OpenSpec change scaffold/spec delta for the new behavior.
Reviewed changes
Copilot reviewed 10 out of 10 changed files in this pull request and generated 5 comments.
Show a summary per file
| File | Description |
|---|---|
| openspec/changes/coordinator-auto-poll-streaming-conversion/tasks.md | OpenSpec task breakdown for implementing coordinator auto-poll + ingest loop |
| openspec/changes/coordinator-auto-poll-streaming-conversion/specs/conversion-webhook-lifecycle/spec.md | Spec delta documenting auto-poll scheduling, dedupe, and timeout behavior |
| openspec/changes/coordinator-auto-poll-streaming-conversion/proposal.md | Motivation and scope constraints for choosing polling over push callbacks |
| openspec/changes/coordinator-auto-poll-streaming-conversion/design.md | Design details for setTimeout polling chain, registry, terminal detection |
| openspec/changes/coordinator-auto-poll-streaming-conversion/acceptance.md | Acceptance criteria for tests/spec validation and runtime expectations |
| bim-review-coordinator/tests/unit_kitpool.test.ts | Extend test config fixture with new polling config fields |
| bim-review-coordinator/tests/auto-poll-conversion.test.ts | New unit tests validating dispatch-triggered polling and manual cancel behavior |
| bim-review-coordinator/src/services/streamingConversionClient.ts | Add terminal helper + poller implementation with timeout handling |
| bim-review-coordinator/src/config.ts | Add polling config fields + boolean env parsing |
| bim-review-coordinator/src/app.ts | Add poller registry, refactor ingest logic, schedule poller after dispatch, cancel on manual ingest, add dispose() |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+228
to
+240
| if (attempts >= options.maxAttempts) { | ||
| const fakeTimeoutResult: StreamingConversionResult = { | ||
| conversion_job_id: conversionJobId, | ||
| status: "failed", | ||
| ready: false, | ||
| model_status: "failed", | ||
| usdc_ref: null, | ||
| element_mapping_ref: null, | ||
| manifest_ref: null, | ||
| reason: "poll_timeout", | ||
| raw: { reason: "poll_timeout", attempts }, | ||
| }; | ||
| await options.onTerminal(fakeTimeoutResult); |
| const result = options.result ?? (await streamingConversionClient.fetchConversionResult(conversionJobId)); | ||
| const correlationId = result.correlation_id; | ||
| if (!correlationId) { | ||
| return { ok: false, status: 422, detail: "streaming conversion result has no correlation_id" }; |
Comment on lines
+830
to
+834
| await ingestStreamingConversionResult(conversionJobId, { result, source: "auto-poll" }); | ||
| } catch (err) { | ||
| console.warn("auto-poll ingest failed", { conversionJobId, err: err instanceof Error ? err.message : String(err) }); | ||
| } finally { | ||
| pollerRegistry.delete(conversionJobId); |
| // - dispatch 後 in-process polling 自動 ingest ready / failed | ||
| // - poller 重複 dispatch 不雙起 | ||
| // - manual ingest endpoint 觸發 cancel auto poller(no double ingest) | ||
| // - max attempts 達到 → poll_timeout 走 failed-equivalent ingest |
Comment on lines
+970
to
976
| // coordinator-auto-poll-streaming-conversion §4.4:manual endpoint 觸發前 cancel | ||
| // 對應 auto-poller,避免後續 onTerminal 觸發第二次 ingest(double callback)。 | ||
| const existing = pollerRegistry.get(conversionJobId); | ||
| if (existing) { | ||
| existing.cancel(); | ||
| pollerRegistry.delete(conversionJobId); | ||
| } |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f59ccb3fd2
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
+229
to
+233
| const fakeTimeoutResult: StreamingConversionResult = { | ||
| conversion_job_id: conversionJobId, | ||
| status: "failed", | ||
| ready: false, | ||
| model_status: "failed", |
There was a problem hiding this comment.
When maxAttempts is reached, the synthesized timeout result omits correlation_id, but ingestStreamingConversionResult requires that field and returns 422 without ingesting. In that timeout path, the poller then de-registers and stops, so the job never transitions to failed, no conversion_failed callback is enqueued, and viewer_url/terminal state can remain stuck indefinitely unless someone manually recovers it.
Useful? React with 👍 / 👎.
Comment on lines
+972
to
976
| const existing = pollerRegistry.get(conversionJobId); | ||
| if (existing) { | ||
| existing.cancel(); | ||
| pollerRegistry.delete(conversionJobId); | ||
| } |
There was a problem hiding this comment.
The manual ingest endpoint now cancels and removes the active poller before checking whether the conversion result is terminal. If an internal caller invokes /api/internal/conversions/:id/ingest while the job is still queued/running (the endpoint returns 409), auto-polling has already been disabled and is not re-registered, so the job can remain permanently stuck in non-terminal state unless another manual ingest is triggered later.
Useful? React with 👍 / 👎.
| // coordinator-auto-poll-streaming-conversion:default 啟用 polling;test fixture | ||
| // 應在 loadConfig overrides 內傳 conversionPollEnabled: false。 | ||
| conversionPollEnabled: parseBooleanEnv("CONVERSION_POLL_ENABLED", true), | ||
| conversionPollIntervalSeconds: numberFromEnv("CONVERSION_POLL_INTERVAL_SECONDS", 5), |
There was a problem hiding this comment.
The new polling interval env is parsed via numberFromEnv, which uses integer parsing; values like CONVERSION_POLL_INTERVAL_SECONDS=0.5 are truncated to 0. That makes setTimeout(..., 0) in the poll chain and can generate very high-frequency polling load against the conversion API. This is especially easy to hit because the test suite uses sub-second intervals (e.g. 0.05) through overrides, so operators may reasonably try fractional seconds in env too.
Useful? React with 👍 / 👎.
monkey1sai
deleted the
codex/openspec/coordinator-auto-poll-streaming-conversion
branch
May 22, 2026 05:28
monkey1sai
added a commit
that referenced
this pull request
May 22, 2026
…步 specs + roadmap (#99) PR #98(implementation,squash `2103bf7`)merged 後的 OpenSpec sync/archive cleanup。 - `git mv openspec/changes/coordinator-auto-poll-streaming-conversion/` → `openspec/changes/archive/2026-05-22-coordinator-auto-poll-streaming-conversion/` - `openspec/specs/conversion-webhook-lifecycle/spec.md` MODIFIED `Coordinator ingests host-native conversion result into callback outbox` requirement:加 SHALL auto-poll 子條款(env config 三項可調)+ 3 新 Scenario(dispatch auto-schedules / manual de-dup / poll timeout failed-equivalent)+ implementation status note - `docs/plans/AI-BIM-governance-saas-roadmap-2026-05.md` 加 archive 摘要(2026-05-22 fast-mvp loop zero-touch automation 完成) `openspec validate --specs --strict` = **26 passed / 0 failed** ## fast-mvp loop zero-touch 收尾完成 4 段 OpenSpec change archive(`remove-conflict-review-from-fast-mvp` + `fast-ifc-link-demo-loop` + `streaming-server-prefer-local-ifc-path` + `coordinator-auto-poll-streaming-conversion`)+ 4 段 hotfix bundle(PR #94 / #95 / #96 / #98)整體把外部 IFC Worker → coordinator → streaming-server → viewer 連結這條 happy path 從 spec 到 code、從 manual 到 zero-touch 完整對齊並驗證。 L4 真實 runtime 第一次 zero-touch end-to-end 跑通(2026-05-22):dispatch → 40 秒內 viewer_url 自動出現,無 manual ingest trigger。 Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
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
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.
Summary
完整 OpenSpec change(
coordinator-auto-poll-streaming-conversion)。補實既有Coordinator ingests host-native conversion result into callback outboxrequirement 已寫但未實作的「polling / internal result loop」gap。fast-mvp loop archive 三段(
remove-conflict-review-from-fast-mvp/fast-ifc-link-demo-loop/streaming-server-prefer-local-ifc-path)+ hotfix 三段(PR #94 / #95 / #96)收尾後驗證,coordinator 仍要外部手動 POST/api/internal/conversions/<id>/ingest才能拿到 viewer_url。本 PR 加 in-process auto-poller 完成自動化,merge 後 fast-mvp happy path 真正 zero-touch。What changed(10 files / +968 / -50)
Coordinator src
streamingConversionClient.ts:isTerminalConversionResulthelper +pollConversionResult(jobId, options): PollerHandle(setTimeout chain、cancel()、max attempts → poll_timeout fake failed result)config.ts:conversionPollEnabled/conversionPollIntervalSeconds(5)/conversionPollMaxAttempts(60)+parseBooleanEnvapp.ts:pollerRegistry、refactor manual ingest 邏輯成共用 helper、dispatch 後 schedule、manual ingest 開頭 cancel、dispose()Coordinator tests
unit_kitpool.test.ts:fixture 加 3 個 config field(test 預設false防 timer 干擾 isolation)auto-poll-conversion.test.ts(新):5 case cover happy path / failed / duplicate dispatch budget / manual cancel auto / disabled configOpenSpec change
conversion-webhook-lifecycleMODIFIED requirement + 3 新 Scenario(dispatch auto-schedules poller / manual de-dup / poll timeout failed-equivalent)GitNexus blast radius = LOW
fetchConversionResultcreateCoordinatorAppcreateCoordinatorAppVerification
npm run verifypytest tests -qpytest testsopenspec validate coordinator-auto-poll-streaming-conversion --strictopenspec validate --specs --strictfetchConversionResult/createCoordinatorAppviewer_url自動出現Design highlights
詳見
openspec/changes/coordinator-auto-poll-streaming-conversion/design.md。重點:poll_timeoutfailed-equivalent,coordinator 端標 failedpollerRegistrykeyed byconversion_job_id,dispatch 前 lookup 已 active 就 reusedispose()清空所有 timer 防止 process exit 卡住Backward compatibility
POST /api/external/ifc-readyresponse 不變(202 dispatched,新增背景 schedule poller)POST /api/internal/conversions/:id/ingestresponse 不變(同樣 callback / session payload,新增 cancel auto poller 動作)unit_kitpoolfixture 加conversionPollEnabled: false防 timer 干擾)Spec / Capability impact
conversion-webhook-lifecycleMODIFIEDCoordinator ingests host-native conversion result into callback outboxrequirement,加 SHALL auto-poll 子條款 + 3 新 Scenario:Test plan
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Tests
Documentation