runner: 진행 상황 출력 — 프로그램별 호출 건수와 호출 하나 끝날 때마다 [i/n]·경과·예상 남은 시간 (stderr)

고객사 실측: unit 100+ 인 프로그램은 15분 넘게 아무것도 안 찍혀 멈춘 것으로 보였다.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

summarize/runner.py

@@ -17,6 +17,7 @@ import argparse
 import json
 import re
 import sqlite3
+import sys
 import time
 from concurrent.futures import ThreadPoolExecutor, as_completed
 from datetime import datetime, timezone
@@ -428,18 +429,29 @@ def _llm_call(llm, system: str, prompt: str) -> dict:
         return {"error": e}
 
 
+def _progress(i: int, n: int, t0: float, c: dict) -> None:
+    """호출 하나가 끝날 때마다 한 줄 — 긴 프로그램에서 멈춘 것처럼 보이지 않게 (stderr)."""
+    state = "실패" if "error" in c else ("대기" if c.get("pending") else "ok")
+    elapsed = time.time() - t0
+    eta = elapsed / i * (n - i) if i else 0
+    print(f"  [{i}/{n}] {state}  {elapsed:.0f}s 경과, 약 {eta:.0f}s 남음", file=sys.stderr, flush=True)
+
+
 def _run_calls_parallel(llm, calls: list[dict], concurrency: int) -> None:
     """calls 각 원소의 'prompt' 를 호출하고 결과를 그 자리에 채운다 (in-place)."""
     if not calls:
         return
+    t0 = time.time()
     if concurrency <= 1 or len(calls) == 1:
-        for c in calls:
+        for i, c in enumerate(calls, 1):
             c.update(_llm_call(llm, SYSTEM_PROMPT, c["prompt"]))
+            _progress(i, len(calls), t0, c)
         return
     with ThreadPoolExecutor(max_workers=min(concurrency, len(calls))) as ex:
         futures = {ex.submit(_llm_call, llm, SYSTEM_PROMPT, c["prompt"]): c for c in calls}
-        for fut in as_completed(futures):
+        for i, fut in enumerate(as_completed(futures), 1):
             futures[fut].update(fut.result())
+            _progress(i, len(futures), t0, futures[fut])
 
 
 def _dedupe_plan(con: sqlite3.Connection, pending: list[sqlite3.Row]) -> tuple[list[sqlite3.Row], dict]:
@@ -569,7 +581,10 @@ def summarize_units(program: str | None, limit: int | None, fake: bool = False,
                 jobs.append((u, [dict(x) for x in prompts]))
 
             # 2) LLM 호출 (병렬, DB 접근 없음) — 수정사항 10번
-            _run_calls_parallel(llm, [c for _, calls in jobs for c in calls], conc)
+            all_calls = [c for _, calls in jobs for c in calls]
+            print(f"[{p['name']}] unit {len(units)}건 → LLM 호출 {len(all_calls)}건 (동시 {conc})",
+                  file=sys.stderr, flush=True)
+            _run_calls_parallel(llm, all_calls, conc)
 
             # 3) 검증·저장 (DB 쓰기, 메인 스레드)
             for u, calls in jobs: