Upload sync_hf.py

scripts/sync_hf.py

@@ -621,24 +621,60 @@ class OpenClawFullSync:
     # ── Application runner ─────────────────────────────────────────────
 
     def run_doctor_fix(self):
-        """Run `openclaw doctor --fix` so pending state-DB schema migrations
-        (e.g. audit-events-v2) are applied BEFORE the gateway starts.
-        Exit code 78 = gateway.maintenance_required."""
+        """Apply pending OpenClaw state-DB migrations (e.g. audit-events-v2)
+        BEFORE the gateway starts. Output is streamed live so a hang is visible.
+
+        Env knobs: DOCTOR_TIMEOUT (seconds, default 900), SKIP_DOCTOR=1."""
+        if os.environ.get("SKIP_DOCTOR") == "1":
+            print("[SYNC] SKIP_DOCTOR=1 — skipping doctor")
+            return True
         if not (APP_DIR / "dist" / "entry.js").exists():
-            return
-        print("[SYNC] Running `openclaw doctor --fix` (schema migrations)...")
+            return True
+        timeout = int(os.environ.get("DOCTOR_TIMEOUT", "900"))
+
+        env = os.environ.copy()
+        env["CI"] = "1"
+        env["NO_COLOR"] = "1"
+        # Drop --require/-r preloads (telegram-proxy, token-redirect, ...): they
+        # keep timers/sockets open and can stop a short-lived CLI from exiting.
+        no = env.get("NODE_OPTIONS", "")
+        no = re.sub(r"(--require[= ]\S+|-r \S+)", "", no).strip()
+        if no:
+            env["NODE_OPTIONS"] = no
+        else:
+            env.pop("NODE_OPTIONS", None)
+
+        cmd = ["node", "dist/entry.js", "doctor", "--fix", "--non-interactive"]
+        print(f"[SYNC] Running: {' '.join(cmd)} (timeout={timeout}s)")
+        t0 = time.time()
         try:
-            r = subprocess.run(
-                ["node", "dist/entry.js", "doctor", "--fix"],
-                cwd=str(APP_DIR), env=os.environ.copy(),
-                capture_output=True, text=True, timeout=300,
+            proc = subprocess.Popen(
+                cmd, cwd=str(APP_DIR), env=env,
+                stdin=subprocess.DEVNULL,
+                stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
+                text=True, bufsize=1, start_new_session=True,
             )
-            tail = (r.stdout + r.stderr).strip().splitlines()[-15:]
-            for line in tail:
-                print(f"[DOCTOR] {line}")
-            print(f"[SYNC] doctor --fix exit code: {r.returncode}")
         except Exception as e:
-            print(f"[SYNC] doctor --fix failed: {e}")
+            print(f"[SYNC] could not start doctor: {e}")
+            return False
+
+        def pump():
+            for line in proc.stdout:
+                print(f"[DOCTOR +{time.time() - t0:.0f}s] {line.rstrip()}", flush=True)
+        th = threading.Thread(target=pump, daemon=True)
+        th.start()
+        try:
+            rc = proc.wait(timeout=timeout)
+        except subprocess.TimeoutExpired:
+            print(f"[SYNC] doctor exceeded {timeout}s — killing it")
+            try:
+                os.killpg(proc.pid, signal.SIGKILL)
+            except Exception:
+                proc.kill()
+            return False
+        th.join(timeout=5)
+        print(f"[SYNC] doctor --fix exit code: {rc} ({time.time() - t0:.0f}s)")
+        return rc == 0
 
     def run_openclaw(self):
         log_file = OPENCLAW_HOME / "workspace" / "startup.log"