Fix: db_sync.py - smart restore + WAL-aware backup

db_sync.py

@@ -10,7 +10,9 @@ Usage:
 
 import os
 import shutil
+import sqlite3
 import sys
+import tempfile
 import time
 
 HF_TOKEN   = os.environ.get("HF_TOKEN", "")
@@ -27,6 +29,13 @@ def _api():
     return HfApi(token=HF_TOKEN)
 
 
+def _safe_copy(src, dst):
+    """Copy a SQLite DB using the backup API so WAL data is included."""
+    with sqlite3.connect(src) as src_conn:
+        with sqlite3.connect(dst) as dst_conn:
+            src_conn.backup(dst_conn)
+
+
 def restore():
     """Download DB from HF backup repo if it exists. Returns True if restored."""
     try:
@@ -59,26 +68,40 @@ def restore():
 
 
 def backup():
-    """Upload current DB to HF backup repo."""
+    """Upload current DB to HF backup repo using SQLite backup API (handles WAL)."""
     if not os.path.exists(DB_PATH):
         print(f"[db_sync] No DB at {DB_PATH} — skipping backup.", flush=True)
         return False
+    tmp = None
     try:
         api = _api()
-        size = os.path.getsize(DB_PATH)
-        print(f"[db_sync] Backing up DB ({size:,} bytes) → {BACKUP_REPO}...", flush=True)
+        # Use SQLite backup API to create a consistent snapshot that includes WAL data.
+        # Uploading the raw .sqlite3 file misses uncommitted WAL transactions.
+        tmp = DB_PATH + ".upload_tmp"
+        _safe_copy(DB_PATH, tmp)
+        size = os.path.getsize(tmp)
+        # Verify the copy has data
+        with sqlite3.connect(tmp) as check:
+            n = check.execute("SELECT COUNT(*) FROM htx_user").fetchone()[0]
+        print(f"[db_sync] Backing up DB ({size:,} bytes, {n} users) → {BACKUP_REPO}...", flush=True)
         api.upload_file(
-            path_or_fileobj=DB_PATH,
+            path_or_fileobj=tmp,
             path_in_repo=REMOTE_FILE,
             repo_id=BACKUP_REPO,
             repo_type="dataset",
-            commit_message="Auto-backup from HF Space",
+            commit_message=f"Auto-backup from HF Space ({n} users)",
         )
-        print(f"[db_sync] Backup complete.", flush=True)
+        print(f"[db_sync] Backup complete ({n} users).", flush=True)
         return True
     except Exception as e:
         print(f"[db_sync] Backup failed: {e}", flush=True)
         return False
+    finally:
+        if tmp and os.path.exists(tmp):
+            try:
+                os.unlink(tmp)
+            except Exception:
+                pass
 
 
 def watch():