feat(worker-pool): Registry-1C cap 1MB + deterministic compaction

사용자 결정 2026-05-19: 100KB cap 이 운영 7d 데이터 1.36MB 대비 부족 →
cap 상향만으로 raw 비대화 위험. cap 1MB + payload compaction 병행.

fetch_recap_context() 변경:
- memo payload item field 축소 = id/title/ai_tldr/ai_event_kind/created_at (5 필드)
  (ai_bullets/file_type/source_channel/category/extracted_text 등 제외)
- memo top-N = RECAP_MEMO_TOP_N env (default 200) — 초과분은 aggregate 로
- aggregate = memos_by_day + memos_by_kind + omitted_memos
- payload_compacted flag = aggregate fallback 발현 여부
- events 는 raw (운영 7d 데이터에서 통상 0~소량)

internal_worker.py:
- PAYLOAD_MAX_BYTES → _payload_max_bytes() env override
  (WORKER_RECAP_PAYLOAD_MAX_BYTES default 1_000_000)
- JobsRecapResponse 에 payload_compacted / omitted_memos 노출
- 413 detail 에 "after compaction" 명시 + RECAP_MEMO_TOP_N 조정 안내

테스트 3 항목 신규 + 기존 endpoint 413 test 업데이트:
- 700 memo → 200 kept + 500 omitted + compacted=true + < 1MB
- 10 memo → compacted=false + omitted=0
- 비정상 큰 title (compaction 후에도 cap 초과) → 413 유지

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

app/api/internal_worker.py

@@ -15,6 +15,7 @@ worker-pool-policy §B.2 invariant 매핑:
 """
 
 import json
+import os
 from datetime import datetime, timezone
 from typing import Annotated, Any
 
@@ -30,8 +31,11 @@ from models.worker_pool import WorkerCapability, WorkerHeartbeat, WorkerJob
 from services.worker_recap_context import fetch_recap_context
 
 # PR-Worker-Pool-Registry-1C — payload size guard (recap context 가 큰 경우 차단).
-# JSON 직렬화 후 100KB 초과 → 413. memo/event 7d 묶음이 통상 < 30KB.
-PAYLOAD_MAX_BYTES = 100_000
+# 사용자 결정 2026-05-19: cap 1MB 상향 + fetch_recap_context deterministic compaction
+# (top-N memo + daily/kind aggregate). 운영 7d 데이터 ~1.36MB → 100KB 부족 → 1MB.
+# 운영 조정용 env override = `WORKER_RECAP_PAYLOAD_MAX_BYTES`.
+def _payload_max_bytes() -> int:
+    return int(os.getenv("WORKER_RECAP_PAYLOAD_MAX_BYTES", "1000000"))
 
 router = APIRouter()
 
@@ -254,6 +258,8 @@ class JobsRecapResponse(BaseModel):
     memo_count: int
     event_count: int
     payload_bytes: int
+    payload_compacted: bool
+    omitted_memos: int
 
 
 @router.post("/jobs/recap", response_model=JobsRecapResponse)
@@ -269,12 +275,14 @@ async def enqueue_recap(
     """
     context = await fetch_recap_context(session, user_id=user.id, days=body.days)
     payload_bytes = len(json.dumps(context, ensure_ascii=False).encode("utf-8"))
-    if payload_bytes > PAYLOAD_MAX_BYTES:
+    cap = _payload_max_bytes()
+    if payload_bytes > cap:
         raise HTTPException(
             status_code=status.HTTP_413_REQUEST_ENTITY_TOO_LARGE,
             detail=(
-                f"recap context payload {payload_bytes} bytes > {PAYLOAD_MAX_BYTES} bytes. "
-                "days 를 줄여 재시도하거나 운영자에게 cap 조정 요청."
+                f"recap context payload {payload_bytes} bytes > {cap} bytes (after compaction). "
+                f"days 를 줄여 재시도 (현재 {body.days}d) 또는 운영자에게 RECAP_MEMO_TOP_N / "
+                "WORKER_RECAP_PAYLOAD_MAX_BYTES 조정 요청."
             ),
         )
 
@@ -291,6 +299,8 @@ async def enqueue_recap(
         memo_count=context["memo_count"],
         event_count=context["event_count"],
         payload_bytes=payload_bytes,
+        payload_compacted=context["payload_compacted"],
+        omitted_memos=context["summary_stats"]["omitted_memos"],
     )
 
 

app/services/worker_recap_context.py

@@ -1,16 +1,28 @@
 """PR-Worker-Pool-Registry-1C — recap context 조립 service.
 
 memo/event 7d recap context 를 user_id 기준으로 묶어 worker_jobs.payload 로 사용.
-회귀 위험 0 보장:
+
+회귀 위험 0:
   - 새 service 모듈에 격리. memos/events API select 절 touch 0.
   - read-only 쿼리만. INSERT/UPDATE 0.
+
+policy:
   - documents 는 single-user invariant (user_id 컬럼 부재) — file_type='note' 7d 전체.
   - events 는 user_id 매칭 + cancelled 제외.
   - timezone = Asia/Seoul ([[project_voice_memo_pipeline]] + events DEFAULT_TIMEZONE 일관).
+
+Deterministic compaction (cap 정책 후속, 사용자 결정 2026-05-19):
+  - memo payload item field 축소 = `id`, `title`, `ai_tldr`, `ai_event_kind`, `created_at` (5 필드만)
+  - memo top-N = `RECAP_MEMO_TOP_N` (default 200) — 초과분은 aggregate 로 대체
+  - aggregate = `memos_by_day` + `memos_by_kind` + `omitted_memos`
+  - `payload_compacted` flag = aggregate fallback 발현 여부
+  - events 는 raw 그대로 (7d 운영 데이터에서 통상 작음)
 """
 
 from __future__ import annotations
 
+import os
+from collections import Counter, defaultdict
 from datetime import datetime, timedelta, timezone
 from typing import Any
 from zoneinfo import ZoneInfo
@@ -25,17 +37,18 @@ DEFAULT_TIMEZONE = "Asia/Seoul"
 KST = ZoneInfo(DEFAULT_TIMEZONE)
 
 
-def _memo_to_payload_item(doc: Document) -> dict[str, Any]:
+def _memo_top_n() -> int:
+    return int(os.getenv("RECAP_MEMO_TOP_N", "200"))
+
+
+def _compact_memo(doc: Document) -> dict[str, Any]:
+    # 5 필드만 — ai_bullets / file_type / source_channel / category / extracted_text 등 제외.
     return {
         "id": doc.id,
         "title": doc.title,
         "ai_tldr": doc.ai_tldr,
-        "ai_bullets": doc.ai_bullets,
         "ai_event_kind": doc.ai_event_kind,
-        "category": doc.category,
         "created_at": doc.created_at.astimezone(KST).isoformat() if doc.created_at else None,
-        "file_type": doc.file_type,
-        "source_channel": doc.source_channel,
     }
 
 
@@ -54,19 +67,32 @@ def _event_to_payload_item(ev: Event) -> dict[str, Any]:
     }
 
 
+def _aggregate_memos(docs: list[Document]) -> dict[str, Any]:
+    by_day: dict[str, int] = defaultdict(int)
+    by_kind: Counter[str] = Counter()
+    for d in docs:
+        if d.created_at is not None:
+            day = d.created_at.astimezone(KST).date().isoformat()
+            by_day[day] += 1
+        by_kind[d.ai_event_kind or "_unknown"] += 1
+    # by_day → 날짜 정렬, by_kind → 빈도 정렬
+    return {
+        "memos_by_day": dict(sorted(by_day.items())),
+        "memos_by_kind": dict(by_kind.most_common()),
+    }
+
+
 async def fetch_recap_context(
     session: AsyncSession, user_id: int, days: int = 7
 ) -> dict[str, Any]:
-    """user_id 의 최근 N 일 memo + event 묶음 반환.
+    """user_id 의 최근 N 일 memo + event recap context.
 
-    memos: documents WHERE file_type='note' AND deleted_at IS NULL AND created_at >= cutoff
-      (single-user invariant — documents.user_id 부재).
-    events: events WHERE user_id=:user_id AND cancelled_at IS NULL AND updated_at >= cutoff.
-
-    archived 메모 제외 (recap 노이즈 감소). audio voice 메모는 file_type='immutable' 라 자동 제외.
+    memo > top_n 일 때 deterministic compaction (최근 top_n full compact + 나머지 aggregate).
+    payload_compacted flag 로 fallback 발현 여부 노출.
     """
     now = datetime.now(timezone.utc)
     cutoff = now - timedelta(days=days)
+    top_n = _memo_top_n()
 
     memo_res = await session.execute(
         select(Document)
@@ -78,7 +104,21 @@ async def fetch_recap_context(
         )
         .order_by(Document.created_at.desc())
     )
-    memos = [_memo_to_payload_item(d) for d in memo_res.scalars().all()]
+    all_memos = memo_res.scalars().all()
+    total_memos = len(all_memos)
+
+    if total_memos > top_n:
+        kept = all_memos[:top_n]
+        omitted = all_memos[top_n:]
+        memos_payload = [_compact_memo(d) for d in kept]
+        aggregate = _aggregate_memos(omitted)
+        omitted_count = len(omitted)
+        payload_compacted = True
+    else:
+        memos_payload = [_compact_memo(d) for d in all_memos]
+        aggregate = {"memos_by_day": {}, "memos_by_kind": {}}
+        omitted_count = 0
+        payload_compacted = False
 
     event_res = await session.execute(
         select(Event)
@@ -97,8 +137,16 @@ async def fetch_recap_context(
         "period_start": cutoff.astimezone(KST).isoformat(),
         "period_end": now.astimezone(KST).isoformat(),
         "timezone": DEFAULT_TIMEZONE,
-        "memos": memos,
+        "memos": memos_payload,
         "events": events,
-        "memo_count": len(memos),
+        "memo_count": total_memos,
         "event_count": len(events),
+        "summary_stats": {
+            "total_memos": total_memos,
+            "memos_kept": len(memos_payload),
+            "omitted_memos": omitted_count,
+            "top_n": top_n,
+            **aggregate,
+        },
+        "payload_compacted": payload_compacted,
     }

tests/test_worker_recap_compaction.py

@@ -0,0 +1,202 @@
+"""PR-Worker-Pool-Registry-1C compaction — deterministic top-N + aggregate.
+
+3 항목 (사용자 결정 2026-05-19 spec):
+  1. 700+ memo → /jobs/recap 200 + payload_size < 1MB + payload_compacted=true + omitted > 0
+  2. 작은 fixture (10 memos) → compacted=false + omitted=0
+  3. 거대 title (1MB 초과 시뮬레이션) → 413 유지
+
+monkeypatch 로 fetch_recap_context 또는 _memo_top_n 조정.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import sys
+
+import pytest
+import pytest_asyncio
+
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "app"))
+
+from httpx import ASGITransport, AsyncClient
+
+from _worker_pool_helpers import (
+    cleanup_worker_jobs,
+    ensure_user,
+    fetch_worker_job,
+    mint_access_token,
+)
+
+
+@pytest_asyncio.fixture
+async def env_setup(monkeypatch):
+    monkeypatch.setenv("LAPTOP_WORKER_BOT_USERNAME", "laptop-worker-bot")
+
+
+def _make_fake_context(total_memos: int, top_n: int = 200) -> dict:
+    """compaction 모사 — _compact_memo 형식 (5 필드) 의 row."""
+    from datetime import datetime, timezone
+    from zoneinfo import ZoneInfo
+
+    kst = ZoneInfo("Asia/Seoul")
+    now = datetime.now(timezone.utc)
+    kept_count = min(total_memos, top_n)
+    omitted = max(0, total_memos - top_n)
+    memos = [
+        {
+            "id": i,
+            "title": f"memo {i}",
+            "ai_tldr": f"tldr {i}",
+            "ai_event_kind": "note",
+            "created_at": now.astimezone(kst).isoformat(),
+        }
+        for i in range(kept_count)
+    ]
+    return {
+        "user_id": 1,
+        "days": 7,
+        "period_start": now.astimezone(kst).isoformat(),
+        "period_end": now.astimezone(kst).isoformat(),
+        "timezone": "Asia/Seoul",
+        "memos": memos,
+        "events": [],
+        "memo_count": total_memos,
+        "event_count": 0,
+        "summary_stats": {
+            "total_memos": total_memos,
+            "memos_kept": kept_count,
+            "omitted_memos": omitted,
+            "top_n": top_n,
+            "memos_by_day": {"2026-05-19": omitted} if omitted else {},
+            "memos_by_kind": {"note": omitted} if omitted else {},
+        },
+        "payload_compacted": omitted > 0,
+    }
+
+
+@pytest.mark.asyncio
+async def test_compaction_kicks_in_at_top_n(env_setup, monkeypatch):
+    """700 memo → 200 kept + 500 omitted + payload_compacted=true + payload < 1MB."""
+    from api import internal_worker as iw_mod
+    from main import app
+
+    fake = _make_fake_context(total_memos=700, top_n=200)
+
+    async def fake_fetch(session, user_id, days=7):
+        # user_id 채워 줌
+        fake["user_id"] = user_id
+        return fake
+
+    monkeypatch.setattr(iw_mod, "fetch_recap_context", fake_fetch)
+    await ensure_user("test-recap-compact-1c")
+    token = mint_access_token("test-recap-compact-1c")
+
+    try:
+        async with AsyncClient(
+            transport=ASGITransport(app=app), base_url="http://test"
+        ) as c:
+            r = await c.post(
+                "/internal/worker/jobs/recap",
+                json={"days": 7},
+                headers={"Authorization": f"Bearer {token}"},
+            )
+        assert r.status_code == 200, r.text
+        js = r.json()
+        assert js["memo_count"] == 700
+        assert js["payload_compacted"] is True
+        assert js["omitted_memos"] == 500
+        assert js["payload_bytes"] < 1_000_000
+        # DB verify — payload 안에 compacted memos + summary_stats
+        job = await fetch_worker_job(js["job_id"])
+        assert job is not None
+    finally:
+        await cleanup_worker_jobs("recap")
+
+
+@pytest.mark.asyncio
+async def test_no_compaction_when_under_top_n(env_setup, monkeypatch):
+    """10 memo → kept 10 + omitted 0 + payload_compacted=false."""
+    from api import internal_worker as iw_mod
+    from main import app
+
+    fake = _make_fake_context(total_memos=10, top_n=200)
+
+    async def fake_fetch(session, user_id, days=7):
+        fake["user_id"] = user_id
+        return fake
+
+    monkeypatch.setattr(iw_mod, "fetch_recap_context", fake_fetch)
+    await ensure_user("test-recap-uncompact-1c")
+    token = mint_access_token("test-recap-uncompact-1c")
+
+    try:
+        async with AsyncClient(
+            transport=ASGITransport(app=app), base_url="http://test"
+        ) as c:
+            r = await c.post(
+                "/internal/worker/jobs/recap",
+                json={"days": 7},
+                headers={"Authorization": f"Bearer {token}"},
+            )
+        assert r.status_code == 200, r.text
+        js = r.json()
+        assert js["memo_count"] == 10
+        assert js["payload_compacted"] is False
+        assert js["omitted_memos"] == 0
+    finally:
+        await cleanup_worker_jobs("recap")
+
+
+@pytest.mark.asyncio
+async def test_413_when_compacted_still_over_cap(env_setup, monkeypatch):
+    """비정상 큰 title (compaction 후에도 1MB 초과) → 413 유지."""
+    from api import internal_worker as iw_mod
+    from main import app
+
+    async def fake_fetch(session, user_id, days=7):
+        huge_title = "x" * 6000  # per memo
+        return {
+            "user_id": user_id,
+            "days": days,
+            "period_start": "2026-05-12T00:00:00+09:00",
+            "period_end": "2026-05-19T00:00:00+09:00",
+            "timezone": "Asia/Seoul",
+            "memos": [
+                {
+                    "id": i,
+                    "title": huge_title,
+                    "ai_tldr": None,
+                    "ai_event_kind": None,
+                    "created_at": "2026-05-19T00:00:00+09:00",
+                }
+                for i in range(200)
+            ],  # ~1.2MB
+            "events": [],
+            "memo_count": 200,
+            "event_count": 0,
+            "summary_stats": {
+                "total_memos": 200,
+                "memos_kept": 200,
+                "omitted_memos": 0,
+                "top_n": 200,
+                "memos_by_day": {},
+                "memos_by_kind": {},
+            },
+            "payload_compacted": False,
+        }
+
+    monkeypatch.setattr(iw_mod, "fetch_recap_context", fake_fetch)
+    await ensure_user("test-recap-413-large-1c")
+    token = mint_access_token("test-recap-413-large-1c")
+
+    async with AsyncClient(
+        transport=ASGITransport(app=app), base_url="http://test"
+    ) as c:
+        r = await c.post(
+            "/internal/worker/jobs/recap",
+            json={"days": 7},
+            headers={"Authorization": f"Bearer {token}"},
+        )
+    assert r.status_code == 413, r.text
+    assert "after compaction" in r.json()["detail"]

tests/test_worker_recap_endpoint.py

@@ -83,6 +83,8 @@ async def test_recap_endpoint_creates_worker_job(env_setup):
         assert js["memo_count"] >= 0
         assert js["event_count"] >= 0
         assert js["payload_bytes"] > 0
+        assert "payload_compacted" in js
+        assert "omitted_memos" in js
         # DB verify
         job = await fetch_worker_job(js["job_id"])
         assert job is not None
@@ -93,7 +95,7 @@ async def test_recap_endpoint_creates_worker_job(env_setup):
 
 @pytest.mark.asyncio
 async def test_recap_payload_413_when_oversize(env_setup, monkeypatch):
-    """payload 100KB 초과 시 413."""
+    """payload 1MB 초과 시 413 (사용자 결정 2026-05-19 cap 1MB)."""
     from api import internal_worker as iw_mod
     from main import app
 
@@ -107,10 +109,20 @@ async def test_recap_payload_413_when_oversize(env_setup, monkeypatch):
             "period_start": "2026-05-12T00:00:00+09:00",
             "period_end": "2026-05-19T00:00:00+09:00",
             "timezone": "Asia/Seoul",
-            "memos": [{"id": i, "title": "x" * 1000} for i in range(120)],  # ~120KB
+            # ~1.2MB raw payload (compaction 후에도 cap 초과 가정)
+            "memos": [{"id": i, "title": "x" * 6000} for i in range(200)],
             "events": [],
-            "memo_count": 120,
+            "memo_count": 200,
             "event_count": 0,
+            "summary_stats": {
+                "total_memos": 200,
+                "memos_kept": 200,
+                "omitted_memos": 0,
+                "top_n": 200,
+                "memos_by_day": {},
+                "memos_by_kind": {},
+            },
+            "payload_compacted": False,
         }
 
     monkeypatch.setattr(iw_mod, "fetch_recap_context", fake_fetch)
@@ -125,3 +137,4 @@ async def test_recap_payload_413_when_oversize(env_setup, monkeypatch):
         )
     assert r.status_code == 413, r.text
     assert "bytes" in r.json()["detail"]
+    assert "after compaction" in r.json()["detail"]