[feat] solution/backend,frontend,site: 수집·사진 분석·발행 진행을 Teams 로 발송 — 수집 중 배포에도 사진 목록 복구

컨테이너를 재생성하면 docker logs 가 사라져 크롤링이 잘 됐는지 확인할 방법이 없었다.
실측(09-30 버터브루): 수집 도중 배포로 API 가 80초 끊기자 화면이 사진 분석 폴링을
3회 실패 후 포기해 사진 10장이 DB 에 있는데도 0장으로 보였다.

- services/activity_feed.py: 기존 alert_outbox·장애 채널로 kind=activity 이벤트 적재(중복 억제 없음)
- collect_service: 수집 시작 · 채널 크롤링 실패 즉시 · 완료 요약(채널별·fact·사진·누락 필수항목·소요초) · 실패
- vision_service: 사진 분석 결과
- build_service · rollback_service: 첫 발행/재발행/빌드만/되돌리기 시작·끝 — URL · 굽기 사진 미러링 수
- site/prerender.ts: 렌더 보고서에 사진 미러링 수(media) 추가
- frontend pollJob: 연속 3회 실패여도 3분간 무응답일 때만 unreachable
- frontend collectJobs: 사진 분석 폴링을 못 끝내도 사진 목록을 다시 읽고 경고
- test_build_publish: 게이트 반려 검사에서 activity 이벤트는 제외

관련 테스트 17개 파일 295 passed · 2 failed(test_search_console_service — 변경 전에도 실패)
frontend tsc·eslint, site tsc·eslint 통과

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs/ALERTS.md

@@ -13,6 +13,7 @@
 | `partial_failure` | 노래 등 곁가지 생성 실패(발행 자체는 계속) | `song_failed:{place_id}` |
 | `queue_stuck` | dead-letter 누적·좀비 실행·PENDING 30분 이상 정체 | `queue_health` |
 | `recovery` | 위 dedupe_key 가 다음 정상 상태에서 풀릴 때 한 번 | 없음(매번 새 행) |
+| `activity` | 수집 시작·채널 실패·완료 · 사진 분석 · 첫 발행/재발행/빌드/되돌리기 시작·끝(`services/activity_feed.py`) | 없음(매번 보낸다) |
 
 ★ **게이트 반려는 알리지 않는다.** 사장님이 fact 를 안 채웠거나 고유 콘텐츠가 없어서 막힌 건
 운영자가 손댈 일이 아니다 — `build_service._fail(reason, gate=None)` 일 때만 `build_failed`.

solution/backend/services/build_service.py

@@ -1,5 +1,6 @@
 """정적 빌드 + 발행 — BUILD 잡이 하는 일."""
 import os
+import time
 import uuid
 
 from sqlalchemy import select
@@ -25,6 +26,7 @@ from crud.site_crud import SiteCRUD
 from crud.place_crud import PlaceCRUD
 from crud.post_crud import PostCRUD
 from services import (
+    activity_feed,
     alert_service,
     azure_static,
     indexnow,
@@ -118,6 +120,45 @@ async def _log(site_id, version_id, action: PublishAction, result: PublishResult
 
 async def run_build(job: dict) -> dict:
     """BUILD 잡 핸들러."""
+    started = time.monotonic()
+    try:
+        result = await _run_build(job)
+    except Exception as ex:
+        await activity_feed.post(
+            "❌ 빌드 실패",
+            [f"place_id={job['payload'].get('place_id')}", f"{type(ex).__name__}: {ex}",
+             f"{time.monotonic() - started:.0f}초"],
+        )
+        raise
+    await post_build_summary(result, time.monotonic() - started)
+    return result
+
+
+async def post_build_summary(result: dict, elapsed: float) -> None:
+    version = f"v{result.get('version')}"
+    mode = result.get("mode") or "빌드"
+    lines = [f"{result.get('label') or result.get('place_id')} {version}"]
+    if result.get("build_status") == "BUILT":
+        title = f"✅ {mode} 완료"
+        if result.get("url"):
+            lines.append(result["url"])
+    elif result.get("gate", {}).get("passed") is False:
+        title = f"🚫 {mode} 반려(게이트)"
+        lines.append(str(result.get("error") or ""))
+    else:
+        title = f"❌ {mode} 실패"
+        lines.append(str(result.get("error") or ""))
+    media = result.get("render_media")
+    if media:
+        lines.append(
+            f"굽기 — 사진 {media.get('total', 0)}장: 새로 받음 {media.get('fetched', 0)} · "
+            f"기존 {media.get('cached', 0)} · 실패 {media.get('failed', 0)}"
+        )
+    lines.append(f"{elapsed:.0f}초")
+    await activity_feed.post(title, lines)
+
+
+async def _run_build(job: dict) -> dict:
     payload = job["payload"]
     place_id = payload["place_id"]
     owner_user_id = payload["owner_user_id"]
@@ -132,6 +173,7 @@ async def run_build(job: dict) -> dict:
         raise BuildAborted(f"사업장을 찾을 수 없다: {place_id}")
 
     site = await ensure_site(place_id)
+    mode = ("재발행" if site.published_at is not None else "첫 발행") if want_publish else "빌드만"
 
     # 주변 정보(맛집·관광지·축제·코스)는 빌드 시점에 업장 좌표로 새로 받는다 — 발행본은 정적이라 이때 받은 값이 실린다.
     try:
@@ -198,7 +240,12 @@ async def run_build(job: dict) -> dict:
         raise BuildAborted(f"버전 생성 실패: {add_err.name}")
 
     result = {"place_id": place_id, "site_id": str(site.site_id), "version": version_no,
-              "site_version_id": str(version.site_version_id)}
+              "site_version_id": str(version.site_version_id), "label": activity_feed.place_label(place),
+              "mode": mode}
+    await activity_feed.post(
+        f"🏗 {mode} 시작",
+        [f"{result['label']} v{version_no}", f"템플릿 {site.template_id or '기본'}"],
+    )
     # 잡 결과에 남긴다 — "노래가 왜 없나" 를 잡 하나만 열어 보면 알 수 있어야 한다.
     if song_result is not None:
         result["song"] = song_result
@@ -267,6 +314,8 @@ async def run_build(job: dict) -> dict:
         # 발행하지 않는다.
         return await _fail(str(ex))
 
+    if report.get("media"):
+        result["render_media"] = report["media"]
     mismatches = list(report.get("mismatches") or [])
     unique_count_raw = report.get("uniqueContentCount")
     unique_count = unique_count_raw or 0
@@ -366,6 +415,7 @@ async def run_build(job: dict) -> dict:
         if thumbnail_url:
             site.thumbnail_url = thumbnail_url
         result["published"] = True
+        result["url"] = site_payload.publish_url(place, site)
         LOG.i(f"[build] place={place_id} v{version_no} 발행 완료 "
               f"(고유 콘텐츠 {unique_count}건 · {report.get('routes')} 페이지)")
     else:

solution/backend/services/collect_service.py

@@ -1,5 +1,6 @@
 """채널 발견 → 확정 URL 크롤링 → fact·사진 후보 저장."""
 import re
+import time
 import uuid
 
 from common import collect_diagnostics
@@ -12,6 +13,7 @@ from common.utils.gtime import GTime
 from crud.fact_crud import FactCRUD
 from crud.place_crud import PlaceCRUD
 from common.category_schema import get_schema
+from services import activity_feed
 from services.collector import AdapterDisabled, AdapterNotFound, REGISTRY
 from services.collector import yanolja_adapter
 from services.external import naver_place_lookup, perplexity, tour_lookup
@@ -456,14 +458,47 @@ async def store_media(place_id: str, sources: list, unit_map: dict) -> dict:
 # 오케스트레이션
 async def run_collect(job: dict) -> dict:
     """COLLECT 잡 핸들러."""
+    started = time.monotonic()
     with collect_diagnostics.collecting():
-        result = await _run_collect(job)
+        try:
+            result = await _run_collect(job)
+        except Exception as ex:
+            await activity_feed.post(
+                "❌ 수집 실패",
+                [f"place_id={job['payload'].get('place_id')}", f"{type(ex).__name__}: {ex}",
+                 f"{time.monotonic() - started:.0f}초"],
+            )
+            raise
         issues = collect_diagnostics.snapshot()
         if issues:
             result["issues"] = issues
+        await _post_collect_summary(result, issues, time.monotonic() - started)
         return result
 
 
+async def _post_collect_summary(result: dict, issues: list[dict], elapsed: float) -> None:
+    lines = [result.get("label") or f"place_id={result.get('place_id')}"]
+    if result.get("note"):
+        lines.append(result["note"])
+    for row in result.get("channels") or []:
+        lines.append(row)
+    facts = result.get("facts")
+    if facts:
+        reasons = ", ".join(f"{k} {v}" for k, v in facts["by_reason"].items())
+        lines.append(f"fact 저장 {facts['stored']} · 반려 {facts['rejected']}" + (f" ({reasons})" if reasons else ""))
+    media = result.get("media")
+    if media:
+        lines.append(f"사진 {media['stored']}장 (중복 {media['skipped_duplicate']})")
+    cov = result.get("coverage")
+    if cov and cov["missing"]:
+        lines.append(f"필수 항목 {cov['covered']}/{cov['total']} — 누락 {', '.join(cov['missing'])}")
+    other = [i for i in issues if i["stage"] != "fetch"]
+    for issue in other:
+        lines.append(f"⚠️ {issue['stage']} 실패: {issue['error_type']}: {issue['message'][:200]}")
+    lines.append(f"{elapsed:.0f}초")
+    await activity_feed.post("✅ 수집 완료", lines)
+
+
 async def _run_collect(job: dict) -> dict:
     payload = job["payload"]
     place_id = payload["place_id"]
@@ -480,7 +515,7 @@ async def _run_collect(job: dict) -> dict:
     if place.verified_at is None:
         raise CollectAborted("동일 업소 검증(verify) 전에는 수집하지 않는다 — 남의 가게가 섞인다")
 
-    result: dict = {"place_id": place_id}
+    result: dict = {"place_id": place_id, "label": activity_feed.place_label(place)}
     result["discover"] = await discover_links(
         place,
         place_id,
@@ -493,6 +528,12 @@ async def _run_collect(job: dict) -> dict:
         link.channel == LinkChannel.NAVER_PLACE.value for link in targets
     )
 
+    await activity_feed.post(
+        "🔎 수집 시작",
+        [result["label"], f"place_id={place_id}",
+         f"크롤링 대상 {len(targets)}건: " + (", ".join(activity_feed.channel_name(l.channel) for l in targets) or "없음")],
+    )
+
     if not targets:
         result["note"] = "크롤링 대상이 없다(어댑터가 처리할 수 있는 확정 URL 0건)"
         await _finish(place_id, owner_user_id, PlaceStatus.REVIEW)
@@ -521,6 +562,8 @@ async def _run_collect(job: dict) -> dict:
     facts_stat = {"stored": 0, "refreshed": 0, "candidate": 0, "rejected": 0, "by_reason": {}}
     media_stat = {"stored": 0, "skipped_duplicate": 0}
     unit_total = 0
+    channel_rows: list[str] = []
+    result["channels"] = channel_rows
 
     for index, link in enumerate(targets):
         if index > 0:
@@ -530,11 +573,23 @@ async def _run_collect(job: dict) -> dict:
                 fetch_stat["stopped_early"] = True
                 LOG.i(f"[collect] 필수 항목 {cov['covered']}/{cov['total']} 확보 — "
                       f"남은 링크 {fetch_stat['skipped_enough']}건 크롤링 생략")
+                channel_rows.append(f"필수 항목 확보 — 남은 {fetch_stat['skipped_enough']}건 생략")
                 break
 
+        name = activity_feed.channel_name(link.channel)
+        fetch_started = time.monotonic()
+        issues_before = len(collect_diagnostics.snapshot())
         source, outcome = await fetch_one(link, PlaceCategory(place.category))
         fetch_stat[outcome] += 1
         if source is None:
+            fresh = collect_diagnostics.snapshot()[issues_before:]
+            reason = f"{fresh[-1]['error_type']}: {fresh[-1]['message'][:300]}" if fresh else outcome
+            channel_rows.append(f"❌ {name} — {reason}")
+            if outcome == "failed":
+                await activity_feed.post(
+                    "⚠️ 채널 크롤링 실패",
+                    [result["label"], f"{name} {link.url}", reason],
+                )
             continue
 
         # 수집 원문을 링크에 남긴다 — fact 가 아니라 '생성 근거' 자리다.
@@ -558,6 +613,9 @@ async def _run_collect(job: dict) -> dict:
             facts_stat["by_reason"][reason] = facts_stat["by_reason"].get(reason, 0) + n
         media_stat["stored"] += m["stored"]
         media_stat["skipped_duplicate"] += m["skipped_duplicate"]
+        channel_rows.append(
+            f"✓ {name} — fact {f['stored']} · 사진 {m['stored']} · {time.monotonic() - fetch_started:.1f}초"
+        )
 
     result["fetch"] = fetch_stat
     result["units"] = unit_total

solution/backend/services/rollback_service.py

@@ -1,5 +1,6 @@
 """대상 버전 스냅샷 로드 → payload 재조립 → 렌더 → 게이트 → 공개 주소 전환"""
 import os
+import time
 import uuid
 
 from common.database.db_session_manager import DB_SESSION_MNG
@@ -17,8 +18,8 @@ from common.logger import LOG
 from common.utils.gtime import GTime
 from crud.place_crud import PlaceCRUD
 from crud.site_crud import SiteCRUD
-from services import alert_service, azure_static, indexnow, publish_gate, render_service, site_payload
+from services import activity_feed, alert_service, azure_static, indexnow, publish_gate, render_service, site_payload
-from services.build_service import ensure_site, load_channel_links
+from services.build_service import post_build_summary, ensure_site, load_channel_links
 
 _site_crud = SiteCRUD()
 _place_crud = PlaceCRUD()
@@ -45,6 +46,25 @@ async def _log(site_id, version_id, result: PublishResult, gate=None, actor=None
 
 async def run_rollback(job: dict) -> dict:
     """ROLLBACK 잡 핸들러."""
+    started = time.monotonic()
+    try:
+        result = await _run_rollback(job)
+    except Exception as ex:
+        await activity_feed.post(
+            "❌ 되돌리기 실패",
+            [f"place_id={job['payload'].get('place_id')} → v{job['payload'].get('target_version')}",
+             f"{type(ex).__name__}: {ex}", f"{time.monotonic() - started:.0f}초"],
+        )
+        raise
+    await post_build_summary(
+        {**result, "version": result.get("target_version"), "mode": "되돌리기",
+         "build_status": "BUILT" if result.get("rolled_back") else "FAILED"},
+        time.monotonic() - started,
+    )
+    return result
+
+
+async def _run_rollback(job: dict) -> dict:
     payload = job["payload"]
     place_id = payload["place_id"]
     owner_user_id = payload["owner_user_id"]
@@ -76,7 +96,9 @@ async def run_rollback(job: dict) -> dict:
     result = {
         "place_id": place_id, "site_id": str(site.site_id),
         "target_version": target_version, "site_version_id": str(version.site_version_id),
+        "label": activity_feed.place_label(place),
     }
+    await activity_feed.post("🏗 되돌리기 시작", [f"{result['label']} → v{target_version}"])
 
     async def _fail(reason: str, gate: publish_gate.GateResult | None = None):
         await _log(site.site_id, version.site_version_id, PublishResult.REJECTED if gate else PublishResult.FAILED,
@@ -107,6 +129,8 @@ async def run_rollback(job: dict) -> dict:
     except render_service.RenderFailed as ex:
         return await _fail(str(ex))
 
+    if report.get("media"):
+        result["render_media"] = report["media"]
     mismatches = list(report.get("mismatches") or [])
     unique_count_raw = report.get("uniqueContentCount")
 
@@ -141,5 +165,6 @@ async def run_rollback(job: dict) -> dict:
     await alert_service.resolve_alert(f"build_failed:{place_id}", f"롤백 성공 — {place_id} → v{target_version}")
 
     result["rolled_back"] = True
+    result["url"] = site_payload.publish_url(place, site)
     LOG.i(f"[rollback] place={place_id} v{target_version} 로 되돌림")
     return result

solution/backend/services/vision_service.py

@@ -9,6 +9,7 @@ from common.utils.gtime import GTime
 from config.server_configs import external_api_config
 from crud.media_crud import MediaCRUD
 from crud.place_crud import PlaceCRUD
+from services import activity_feed
 from services.external import gemini
 from services.llm import provider
 from common.job_errors import PermanentJobError
@@ -112,4 +113,9 @@ async def run_vision(job: dict) -> dict:
 
     LOG.i(f"[vision] place={place_id} {stat['analyzed']}장 → 자동반영 {stat['approved']} · "
           f"확인필요 {stat['needs_review']} · 실패 {stat['failed']} (임계값 {threshold})")
+    await activity_feed.post(
+        "🖼 사진 분석 완료" if not stat["failed"] else "⚠️ 사진 분석 일부 실패",
+        [activity_feed.place_label(place),
+         f"분석 {stat['analyzed']} · 승인 {stat['approved']} · 확인대기 {stat['needs_review']} · 실패 {stat['failed']}"],
+    )
     return stat

solution/backend/tests/test_build_publish.py

@@ -309,7 +309,7 @@ async def test_구조화데이터가_화면과_다르면_발행하지_않는다(
 
     # 게이트 반려는 알리지 않는다 — 사장님 쪽 문제를 운영자에게 알리면 안 된다.
     async with db_engine.begin() as c:
-        rows = (await c.execute(text("SELECT kind FROM alert_outbox"))).all()
+        rows = (await c.execute(text("SELECT kind FROM alert_outbox WHERE kind <> 'activity'"))).all()
     assert rows == []
 
 

solution/frontend/src/api/pollJob.ts

@@ -12,6 +12,9 @@ export const JOB_POLL_TIMEOUT_MS = 10 * 60 * 1000;
 /** 연속 실패 허용치. */
 export const JOB_MAX_POLL_ERRORS = 3;
 
+/** 응답이 한 번도 없을 때 포기하기까지의 시간. */
+export const JOB_UNREACHABLE_MS = 3 * 60 * 1000;
+
 /** 폴링이 끝난 이유. */
 export type JobOutcome =
   | {kind: 'done'; job: JobData}
@@ -26,6 +29,7 @@ export interface PollJobOptions {
   intervalMs?: number;
   timeoutMs?: number;
   maxErrors?: number;
+  unreachableMs?: number;
   /** 폴링 1회마다 호출. */
   onTick?: (job: JobData) => void;
 }
@@ -52,12 +56,14 @@ export async function pollJob(jobId: string, options: PollJobOptions): Promise<J
     intervalMs = JOB_POLL_MS,
     timeoutMs = JOB_POLL_TIMEOUT_MS,
     maxErrors = JOB_MAX_POLL_ERRORS,
+    unreachableMs = JOB_UNREACHABLE_MS,
     onTick,
   } = options;
 
   const deadline = Date.now() + timeoutMs;
   let errors = 0;
   let lastError: unknown;
+  let lastReachedAt = Date.now();
 
   while (Date.now() < deadline) {
     await delay(intervalMs, signal);
@@ -74,7 +80,10 @@ export async function pollJob(jobId: string, options: PollJobOptions): Promise<J
 
     // 한 번이라도 잡을 받아오면 카운터를 되돌린다 — 잠깐 끊긴 네트워크로 몇 분짜리 잡을 접으면 안 된다.
     errors = job ? 0 : errors + 1;
-    if (errors >= maxErrors) return {kind: 'unreachable', error: lastError};
+    if (job) lastReachedAt = Date.now();
+    if (errors >= maxErrors && Date.now() - lastReachedAt >= unreachableMs) {
+      return {kind: 'unreachable', error: lastError};
+    }
     if (!job) continue;
 
     onTick?.(job);

solution/frontend/src/features/onboarding/collectJobs.ts

@@ -79,7 +79,11 @@ export async function waitForVision(placeId: string, job: JobData, signal: Abort
   const visionJobId = (job.result as {vision_job_id?: string} | undefined)?.vision_job_id;
   if (!visionJobId) return;
   const outcome = await pollJob(visionJobId, {signal});
-  if (outcome.kind === 'done') await refreshPlaceCaches(placeId);
+  if (outcome.kind === 'aborted') return;
+  await refreshPlaceCaches(placeId);
+  if (outcome.kind !== 'done') {
+    notify.warn('사진 분석 결과를 아직 받지 못했습니다.', '잠시 후 새로고침하면 사진이 보입니다.');
+  }
 }
 
 /** 수집이 끝난 뒤 서버가 진실인 것들 — place(주소·갱신시각) · 링크 · fact · **사진**. */

solution/site/scripts/prerender.ts

@@ -255,9 +255,17 @@ const MEDIA_EXT: Record<string, string> = {
 };
 
 /** 사진을 **우리 오리진으로 옮긴다.** */
-async function mirrorMedia(payload: SitePayload, siteDir: string) {
+interface MediaMirrorStats {
+  total: number;
+  fetched: number;
+  cached: number;
+  failed: number;
+}
+
+async function mirrorMedia(payload: SitePayload, siteDir: string): Promise<MediaMirrorStats> {
   const items = (payload.media ?? []).filter((item) => /^https?:\/\//i.test(item.url ?? ''));
-  if (items.length === 0) return;
+  const stats: MediaMirrorStats = {total: items.length, fetched: 0, cached: 0, failed: 0};
+  if (items.length === 0) return stats;
 
   const dir = join(siteDir, MEDIA_DIR);
   // 절대 주소로 바꾼다.
@@ -275,6 +283,7 @@ async function mirrorMedia(payload: SitePayload, siteDir: string) {
     if (hit) {
       wanted.add(hit);
       item.url = joinUrl(publicBase, `${MEDIA_DIR}/${hit}`);
+      stats.cached += 1;
       continue;
     }
 
@@ -297,6 +306,7 @@ async function mirrorMedia(payload: SitePayload, siteDir: string) {
       item.url = joinUrl(publicBase, `${MEDIA_DIR}/${name}`);
       fetched += 1;
     } catch (ex) {
+      stats.failed += 1;
       console.warn(`  ! 사진을 못 받았습니다(원래 주소를 씁니다): ${origin} — ${ex}`);
     }
   }
@@ -306,6 +316,8 @@ async function mirrorMedia(payload: SitePayload, siteDir: string) {
     if (!wanted.has(name)) rmSync(join(dir, name), {force: true});
   }
   if (fetched > 0) console.log(`    사진 ${fetched}장을 내려받았습니다 (총 ${wanted.size}장)`);
+  stats.fetched = fetched;
+  return stats;
 }
 
 /** 이 슬러그·버전이 실제로 구워지는 자리 — `out/versions/<slug>/<version>/`. */
@@ -573,6 +585,7 @@ interface RenderReport {
   /** 절대규칙 3 위반 목록. */
   mismatches: string[];
   error: string | null;
+  media?: MediaMirrorStats;
 }
 
 function writeReport(payloadFile: string, report: RenderReport) {
@@ -922,7 +935,7 @@ async function main() {
         }
         continue;
       }
-      await mirrorMedia(entry.payload, stagingDir);
+      const media = await mirrorMedia(entry.payload, stagingDir);
       const result = prerenderSite(entry.payload, args.out, assets, referenced);
       const payload = result.payload;
       origin = origin || payload.site.origin;
@@ -955,6 +968,7 @@ async function main() {
           jsonld: result.jsonld,
           mismatches: [],
           error: null,
+          media,
         };
         writeFileSync(savedReportPath, JSON.stringify({...report, origin: payload.site.origin}));
         writeReport(entry.file, report);