Update app.py

app.py

@@ -4,27 +4,92 @@ from datasets import Dataset
 from datetime import datetime, timedelta
 import pandas as pd
 import time, os, random, tempfile, json, glob
+import re
+import threading
 
 # --- API / HF 設定 ---
 API_KEY = os.getenv("API_KEY")
 BASE_URL = "https://openrouter.ai/api/v1"
 HF_TOKEN = os.getenv("HF_TOKEN")
 DATASET_REPO = "Toya0421/reading_exercise_logging"
-LOG_FILE = "reading_logs.csv"
+
+# ログは Spaces の永続ストレージ(= Files)へ
+LOG_FILE = "/data/log.csv"
 
 client = OpenAI(base_url=BASE_URL, api_key=API_KEY)
 
 # --- passage_information.xlsx 読み込み (Text# と flesch_score 使用) ---
 passage_info_df = pd.read_excel("passage_information.xlsx")
 
-# --- 状態変数 ---
-used_passages = set()
-current_user_id = None
-current_level = None
+# ======================================================
+#  (③) 重い処理の同時実行制限：rewrite API を最大N並列に制限
+# ======================================================
+REWRITE_CONCURRENCY = int(os.getenv("REWRITE_CONCURRENCY", "3"))  # 5,6人想定なら 2〜3 推奨
+_rewrite_sem = threading.Semaphore(REWRITE_CONCURRENCY)
+
+# ======================================================
+#  (①) ログ：毎アクションはFiles(/data/log.csv)へ追記のみ
+#      + 5分ごとに dataset.push_to_hub だけ実行
+# ======================================================
+_log_lock = threading.Lock()
+PUSH_INTERVAL_SEC = int(os.getenv("PUSH_INTERVAL_SEC", "300"))  # 5分
+
+def save_log(entry):
+    """
+    毎アクション：/data/log.csv に追記するだけ（軽量）
+    """
+    os.makedirs(os.path.dirname(LOG_FILE), exist_ok=True)
+    df = pd.DataFrame([entry])
+
+    with _log_lock:
+        if os.path.exists(LOG_FILE):
+            df.to_csv(LOG_FILE, mode="a", index=False, header=False, encoding="utf-8")
+        else:
+            df.to_csv(LOG_FILE, index=False, header=True, encoding="utf-8")
+
+def _push_logs_to_hub_once():
+    """
+    5分ごと：/data/log.csv を読み込み、parquet化して Dataset として Hub にpush
+    """
+    if not HF_TOKEN:
+        print("[WARN] HF_TOKEN is not set. Skip push.")
+        return
+    if not os.path.exists(LOG_FILE):
+        return
+
+    # CSV追記と競合しないようロック
+    with _log_lock:
+        all_logs = pd.read_csv(LOG_FILE)
 
+    tmp_dir = tempfile.mkdtemp()
+    tmp_path = os.path.join(tmp_dir, "data.parquet")
+    all_logs.to_parquet(tmp_path)
+
+    dataset = Dataset.from_parquet(tmp_path)
+    dataset.push_to_hub(DATASET_REPO, token=HF_TOKEN)
+
+def _start_periodic_push_thread():
+    """
+    アプリ起動時に1回だけ呼ぶ。以後はdaemonスレッドで5分ごとpush。
+    """
+    def _worker():
+        while True:
+            time.sleep(PUSH_INTERVAL_SEC)
+            try:
+                _push_logs_to_hub_once()
+            except Exception as e:
+                # push失敗しても全体を落とさない
+                print(f"[WARN] periodic push failed: {e}")
+
+    t = threading.Thread(target=_worker, daemon=True)
+    t.start()
+
+# 起動（1回だけ）
+_start_periodic_push_thread()
 
 # ======================================================
 #  新しい教材管理：passages フォルダからランダム選択
+#  ※ used_passages は session_state に保持（グローバル禁止）
 # ======================================================
 
 def load_passage_file(text_id):
@@ -37,19 +102,15 @@ def load_passage_file(text_id):
     with open(path, "r", encoding="utf-8") as f:
         return f.read()
 
-def get_new_passage_random():
+def get_new_passage_random(used_passages_set):
     """
     passages フォルダからランダムに教材を選び（pg◯.txt）、
     passage_information.xlsx の Text# の flesch_score を original_level として返す。
     """
-    global used_passages
-
-    # --- pg*.txt を取得 ---
     files = glob.glob("passages/pg*.txt")
     if not files:
-        return None, None, None
+        return None, None, None, used_passages_set
 
-    # --- ファイル名から Text# (整数) を抽出 ---
     all_ids = []
     for f in files:
         name = os.path.basename(f)
@@ -57,33 +118,28 @@ def get_new_passage_random():
         if num.isdigit():
             all_ids.append(int(num))
 
-    # --- 未使用の ID を優先 ---
-    available = [pid for pid in all_ids if pid not in used_passages]
+    available = [pid for pid in all_ids if pid not in used_passages_set]
     if not available:
-        used_passages.clear()
+        used_passages_set = set()
         available = list(all_ids)
 
-    # --- ランダムに選択 ---
     text_id = random.choice(available)
-    used_passages.add(text_id)
+    used_passages_set.add(text_id)
 
-    # --- テキスト読み込み ---
     text = load_passage_file(text_id)
     if text is None:
-        return None, None, None
+        return None, None, None, used_passages_set
 
-    # --- Excel から original_level (flesch_score) を取得 ---
     row = passage_info_df[passage_info_df["Text#"] == text_id]
     if len(row) == 0:
         orig_level = None
     else:
         orig_level = row.iloc[0]["flesch_score"]
 
-    return text_id, text, orig_level
-
+    return text_id, text, orig_level, used_passages_set
 
 # ======================================================
-#  Rewrite
+#  Rewrite（同時実行制限付き）
 # ======================================================
 
 def rewrite_level(text, target_level):
@@ -108,24 +164,22 @@ excluding the title, author name, source information, chapter number, annotation
 {text}
 """
 
-    resp = client.chat.completions.create(
-        model="google/gemini-2.5-flash",
-        messages=[{"role": "user", "content": prompt}],
-        temperature=0.4,
-        max_tokens=5000
-    )
+    # (③) rewrite API 同時実行制限
+    with _rewrite_sem:
+        resp = client.chat.completions.create(
+            model="google/gemini-2.5-flash",
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0.4,
+            max_tokens=5000
+        )
     return resp.choices[0].message.content.strip()
 
-
-import re
-
 def split_pages(text, max_words=300):
     """
     文単位でページを分割する。
     - 文の途中でページを分割しない
     - max_words の上限を超えないようにする
     """
-    # 文に分割（. ? ! のあとに改行やスペースが続くパターン）
     sentences = re.split(r'(?<=[.!?])\s+', text.strip())
     pages = []
     current_page = []
@@ -135,7 +189,6 @@ def split_pages(text, max_words=300):
         words = sentence.split()
         sentence_len = len(words)
 
-        # 次の文を追加すると max_words を超える場合 → 新しいページを作る
         if current_word_count + sentence_len > max_words:
             if current_page:
                 pages.append(" ".join(current_page))
@@ -145,46 +198,50 @@ def split_pages(text, max_words=300):
             current_page.append(sentence)
             current_word_count += sentence_len
 
-    # 最後のページを追加
     if current_page:
         pages.append(" ".join(current_page))
 
     return pages or [text]
 
-
 # ======================================================
-#  Save Log
+#  Start（session_stateでユーザー状態管理）
 # ======================================================
 
-def save_log(entry):
-    df = pd.DataFrame([entry])
-    if os.path.exists(LOG_FILE):
-        df.to_csv(LOG_FILE, mode="a", index=False, header=False)
-    else:
-        df.to_csv(LOG_FILE, index=False)
-
-    all_logs = pd.read_csv(LOG_FILE)
-    tmp_dir = tempfile.mkdtemp()
-    tmp_path = os.path.join(tmp_dir, "data.parquet")
-    all_logs.to_parquet(tmp_path)
-    dataset = Dataset.from_parquet(tmp_path)
-    dataset.push_to_hub(DATASET_REPO, token=HF_TOKEN)
+def start_test(student_id, level_input, session_state):
+    action = "start_pushed"
+    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
 
+    # student_id 未入力でも「押した」ログは残す（元意図に近い）
+    if not student_id or str(student_id).strip() == "":
+        entry = {
+            "user_id": None,
+            "assigned_level": None,
+            "passage_id": None,
+            "original_level": None,
+            "action_time": now,
+            "action_type": action,
+            "page_text": None
+        }
+        save_log(entry)
 
-# ======================================================
-#  Start
-# ======================================================
+        return (
+            "", "", json.dumps([]), 0, "",
+            0, "", None, None,
+            gr.update(interactive=False, visible=False),
+            gr.update(interactive=False, visible=True),
+            gr.update(interactive=False, visible=False),
+            session_state
+        )
 
-def start_test(student_id, level_input):
-    global current_user_id, current_level, used_passages
-    used_passages = set()
+    user_id = str(student_id).strip()
+    level = int(level_input)
 
-    action = "start_pushed"
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
+    # startでリセット（元コード踏襲）
+    used_passages_set = set()
 
     entry = {
-        "user_id": student_id,
-        "assigned_level": current_level,
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": None,
         "original_level": None,
         "action_time": now,
@@ -193,29 +250,18 @@ def start_test(student_id, level_input):
     }
     save_log(entry)
 
-    if not student_id or str(student_id).strip() == "":
-        return (
-            "", "", json.dumps([]), 0, "",
-            0, "", None, None,
-            gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=True),
-            gr.update(interactive=False, visible=False)
-        )
-
-    current_user_id = str(student_id).strip()
-    current_level = int(level_input)
-
-    pid, text, orig_lev = get_new_passage_random()
+    pid, text, orig_lev, used_passages_set = get_new_passage_random(used_passages_set)
     if text is None:
         return (
             "教材が見つかりません", "", json.dumps([]), 0, "",
             0, "", None, None,
             gr.update(interactive=False, visible=False),
             gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=False)
+            gr.update(interactive=False, visible=False),
+            session_state
         )
 
-    rewritten = rewrite_level(text, current_level)
+    rewritten = rewrite_level(text, level)
     pages = split_pages(rewritten)
     total = len(pages)
 
@@ -229,18 +275,25 @@ def start_test(student_id, level_input):
         finish_upd = gr.update(interactive=False, visible=False)
 
     page_num = 1
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
+    now2 = (datetime.utcnow() + timedelta(hours=9)).isoformat()
 
-    entry = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
+    entry2 = {
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": pid,
         "original_level": orig_lev,
-        "action_time": now,
+        "action_time": now2,
         "action_type": f"page_displayed_{page_num}",
         "page_text": pages[0]
     }
-    save_log(entry)
+    save_log(entry2)
+
+    # session_state 更新
+    session_state = {
+        "user_id": user_id,
+        "level": level,
+        "used_passages": list(used_passages_set)
+    }
 
     return (
         pages[0],
@@ -250,23 +303,25 @@ def start_test(student_id, level_input):
         total,
         pid,
         orig_lev,
-        current_level,
+        level,
         prev_upd,
         next_upd,
-        finish_upd
+        finish_upd,
+        session_state
     )
 
-
 # ======================================================
-#  Next / Prev / Finish（以下は元コードのまま）
+#  Next / Prev / Finish（元コードのままの構造 + state参照）
 # ======================================================
 
-def next_page(pages_json, current_page, total_pages, pid, orig_lev):
-    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
+def next_page(pages_json, current_page, total_pages, pid, orig_lev, session_state):
+    user_id = session_state.get("user_id")
+    level = session_state.get("level")
 
+    now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
     entry = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": pid,
         "original_level": orig_lev,
         "action_time": now,
@@ -280,14 +335,15 @@ def next_page(pages_json, current_page, total_pages, pid, orig_lev):
         return ("", "", json.dumps([]), 0,
                 gr.update(interactive=False, visible=False),
                 gr.update(interactive=False, visible=False),
-                gr.update(interactive=False, visible=False))
+                gr.update(interactive=False, visible=False),
+                session_state)
 
     new_page = min(current_page + 1, total_pages - 1)
 
     now2 = (datetime.utcnow() + timedelta(hours=9)).isoformat()
     entry2 = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": pid,
         "original_level": orig_lev,
         "action_time": now2,
@@ -304,7 +360,8 @@ def next_page(pages_json, current_page, total_pages, pid, orig_lev):
             new_page,
             gr.update(interactive=True, visible=True),
             gr.update(interactive=False, visible=False),
-            gr.update(interactive=True, visible=True)
+            gr.update(interactive=True, visible=True),
+            session_state
         )
 
     return (
@@ -314,16 +371,18 @@ def next_page(pages_json, current_page, total_pages, pid, orig_lev):
         new_page,
         gr.update(interactive=(new_page > 0), visible=(new_page > 0)),
         gr.update(interactive=True, visible=True),
-        gr.update(interactive=False, visible=False)
+        gr.update(interactive=False, visible=False),
+        session_state
     )
 
+def prev_page(pages_json, current_page, total_pages, pid, orig_lev, session_state):
+    user_id = session_state.get("user_id")
+    level = session_state.get("level")
 
-def prev_page(pages_json, current_page, total_pages, pid, orig_lev):
     now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
-
     entry = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": pid,
         "original_level": orig_lev,
         "action_time": now,
@@ -337,7 +396,8 @@ def prev_page(pages_json, current_page, total_pages, pid, orig_lev):
         return ("", "", json.dumps([]), 0,
                 gr.update(interactive=False, visible=False),
                 gr.update(interactive=False, visible=False),
-                gr.update(interactive=False, visible=False))
+                gr.update(interactive=False, visible=False),
+                session_state)
 
     new_page = max(current_page - 1, 0)
 
@@ -348,8 +408,8 @@ def prev_page(pages_json, current_page, total_pages, pid, orig_lev):
 
     now2 = (datetime.utcnow() + timedelta(hours=9)).isoformat()
     entry2 = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": pid,
         "original_level": orig_lev,
         "action_time": now2,
@@ -365,17 +425,21 @@ def prev_page(pages_json, current_page, total_pages, pid, orig_lev):
         new_page,
         prev_upd,
         next_upd,
-        finish_upd
+        finish_upd,
+        session_state
     )
 
+def finish_or_retire(pages_json, current_page, pid, orig_lev, action, session_state):
+    user_id = session_state.get("user_id")
+    level = session_state.get("level")
+    used_passages_set = set(session_state.get("used_passages", []))
 
-def finish_or_retire(pages_json, current_page, pid, orig_lev, action):
     pages = json.loads(pages_json)
     now = (datetime.utcnow() + timedelta(hours=9)).isoformat()
 
     entry = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": pid,
         "original_level": orig_lev,
         "action_time": now,
@@ -384,17 +448,18 @@ def finish_or_retire(pages_json, current_page, pid, orig_lev, action):
     }
     save_log(entry)
 
-    new_pid, new_text, new_orig_lev = get_new_passage_random()
+    new_pid, new_text, new_orig_lev, used_passages_set = get_new_passage_random(used_passages_set)
     if new_text is None:
         return (
             "教材がありません", "", json.dumps([]), 0, "",
             0, "", None, None,
             gr.update(interactive=False, visible=False),
             gr.update(interactive=False, visible=False),
-            gr.update(interactive=False, visible=False)
+            gr.update(interactive=False, visible=False),
+            session_state
         )
 
-    rewritten = rewrite_level(new_text, current_level)
+    rewritten = rewrite_level(new_text, level)
     new_pages = split_pages(rewritten)
     total = len(new_pages)
 
@@ -409,8 +474,8 @@ def finish_or_retire(pages_json, current_page, pid, orig_lev, action):
 
     now2 = (datetime.utcnow() + timedelta(hours=9)).isoformat()
     entry2 = {
-        "user_id": current_user_id,
-        "assigned_level": current_level,
+        "user_id": user_id,
+        "assigned_level": level,
         "passage_id": new_pid,
         "original_level": new_orig_lev,
         "action_time": now2,
@@ -419,6 +484,12 @@ def finish_or_retire(pages_json, current_page, pid, orig_lev, action):
     }
     save_log(entry2)
 
+    session_state = {
+        "user_id": user_id,
+        "level": level,
+        "used_passages": list(used_passages_set)
+    }
+
     return (
         new_pages[0],
         f"1 / {total}",
@@ -427,13 +498,13 @@ def finish_or_retire(pages_json, current_page, pid, orig_lev, action):
         total,
         new_pid,
         new_orig_lev,
-        current_level,
+        level,
         prev_upd,
         next_upd,
-        finish_upd
+        finish_upd,
+        session_state
     )
 
-
 # ======================================================
 #  UI
 # ======================================================
@@ -507,10 +578,12 @@ custom_css = """
 }
 """
 
-
 with gr.Blocks(css=custom_css) as demo:
     gr.Markdown("# 📚 Reading Exercise")
 
+    # セッションごとの状態（グローバル禁止）
+    session_state = gr.State({"user_id": None, "level": None, "used_passages": []})
+
     student_id_input = gr.Textbox(label="学生番号（必須）")
     level_input = gr.Dropdown(
         choices=[1,2,3,4,5],
@@ -544,13 +617,14 @@ with gr.Blocks(css=custom_css) as demo:
 
     start_btn.click(
         fn=start_test,
-        inputs=[student_id_input, level_input],
+        inputs=[student_id_input, level_input, session_state],
         outputs=[
             text_display, page_display,
             hidden_pages, hidden_page_index,
             hidden_total_pages, hidden_passage_id,
             hidden_orig_lev, hidden_assigned_lev,
-            prev_btn, next_btn, finish_btn
+            prev_btn, next_btn, finish_btn,
+            session_state
         ]
     )
 
@@ -559,12 +633,13 @@ with gr.Blocks(css=custom_css) as demo:
         inputs=[
             hidden_pages, hidden_page_index,
             hidden_total_pages, hidden_passage_id,
-            hidden_orig_lev
+            hidden_orig_lev, session_state
         ],
         outputs=[
             text_display, page_display,
             hidden_pages, hidden_page_index,
-            prev_btn, next_btn, finish_btn
+            prev_btn, next_btn, finish_btn,
+            session_state
         ]
     )
 
@@ -573,40 +648,47 @@ with gr.Blocks(css=custom_css) as demo:
         inputs=[
             hidden_pages, hidden_page_index,
             hidden_total_pages, hidden_passage_id,
-            hidden_orig_lev
+            hidden_orig_lev, session_state
         ],
         outputs=[
             text_display, page_display,
             hidden_pages, hidden_page_index,
-            prev_btn, next_btn, finish_btn
+            prev_btn, next_btn, finish_btn,
+            session_state
         ]
     )
 
     finish_btn.click(
-        fn=lambda p, i, pid, o: finish_or_retire(p, i, pid, o, "finished"),
-        inputs=[hidden_pages, hidden_page_index, hidden_passage_id, hidden_orig_lev],
+        fn=lambda p, i, pid, o, st: finish_or_retire(p, i, pid, o, "finished", st),
+        inputs=[hidden_pages, hidden_page_index, hidden_passage_id, hidden_orig_lev, session_state],
         outputs=[
             text_display, page_display,
             hidden_pages, hidden_page_index,
             hidden_total_pages, hidden_passage_id,
             hidden_orig_lev, hidden_assigned_lev,
-            prev_btn, next_btn, finish_btn
+            prev_btn, next_btn, finish_btn,
+            session_state
         ]
     )
 
     retire_btn.click(
-        fn=lambda p, i, pid, o: finish_or_retire(p, i, pid, o, "retire"),
+        fn=lambda p, i, pid, o, st: finish_or_retire(p, i, pid, o, "retire", st),
         inputs=[
             hidden_pages, hidden_page_index,
-            hidden_passage_id, hidden_orig_lev
+            hidden_passage_id, hidden_orig_lev,
+            session_state
         ],
         outputs=[
             text_display, page_display,
             hidden_pages, hidden_page_index,
             hidden_total_pages, hidden_passage_id,
             hidden_orig_lev, hidden_assigned_lev,
-            prev_btn, next_btn, finish_btn
+            prev_btn, next_btn, finish_btn,
+            session_state
         ]
     )
 
+# Gradio側のキューもON（HF環境差分で壊れにくい設定に寄せる）
+demo.queue(max_size=64)
+
 demo.launch()