Update scripts/sync_hf.py

scripts/sync_hf.py

@@ -4,10 +4,12 @@ OpenClaw HF Spaces Persistence — Full Directory Sync
 =====================================================
 
 Fixes for 2026.9.7 gateway startup hang:
-- Purge legacy Telegram thread-bindings before doctor
-- Clear stale gateway owner leases from state/openclaw.sqlite
-- Remove whatsapp/perplexity plugin entries that expand plugins.allow
-- Re-enforce plugins.allow AFTER doctor --fix
+- Park all .migrated archives (doctor refuses if these exist)
+- Park inconsistent SQLite databases (state DB caused the 303s read-only timeout)
+- Park empty telegram/thread-bindings-default.json (blocks migration)
+- Remove whatsapp/perplexity plugin + channel entries that expand plugins.allow
+- Add SQLite + .migrated patterns to upload ignore list
+- Re-enforce plugins.allow AFTER doctor --fix (doctor re-expands it)
 - Use 'gateway run' (foreground) instead of 'gateway' (supervised)
 - On health failure, run 'openclaw gateway status' for diagnostics
 """
@@ -45,6 +47,7 @@ from huggingface_hub import HfApi, snapshot_download
 # ── Logging helper ──────────────────────────────────────────────────────────
 
 class TeeLogger:
+    """Duplicate output to stream and file."""
     def __init__(self, filename, stream):
         self.stream = stream
         self.file = open(filename, "a", encoding="utf-8")
@@ -65,21 +68,30 @@ OPENCLAW_HOME = Path.home() / ".openclaw"
 APP_DIR       = Path("/app/openclaw")
 DATASET_PATH  = ".openclaw"
 
+# Z.AI API key (optional)
 ZAI_API_KEY   = os.environ.get("ZAI_API_KEY", "")
+
+# Gateway token
 GATEWAY_TOKEN = os.environ.get("GATEWAY_TOKEN", "huggingclaw")
+
+# A2A configuration (optional)
 AGENT_NAME    = os.environ.get("AGENT_NAME", "HuggingClaw")
 A2A_PEERS     = os.environ.get("A2A_PEERS", "")
+
+# Default model (optional)
 OPENCLAW_DEFAULT_MODEL = os.environ.get("OPENCLAW_DEFAULT_MODEL", "")
 
+# HF Spaces built-in env vars
 SPACE_HOST = os.environ.get("SPACE_HOST", "")
 SPACE_ID   = os.environ.get("SPACE_ID", "")
 
-SYNC_INTERVAL = int(os.environ.get("SYNC_INTERVAL", "600"))
-MIN_UPLOAD_GAP = int(os.environ.get("MIN_UPLOAD_GAP", "300"))
+SYNC_INTERVAL          = int(os.environ.get("SYNC_INTERVAL", "600"))
+MIN_UPLOAD_GAP         = int(os.environ.get("MIN_UPLOAD_GAP", "300"))
 STARTUP_HEALTH_TIMEOUT = int(os.environ.get("STARTUP_HEALTH_TIMEOUT", "900"))
 
 AUTO_CREATE_DATASET = os.environ.get("AUTO_CREATE_DATASET", "false").lower() in ("true", "1", "yes")
 
+# Dataset repo: auto-derive from SPACE_ID when not explicitly set
 HF_REPO_ID = os.environ.get("OPENCLAW_DATASET_REPO", "")
 if not HF_REPO_ID and SPACE_ID:
     HF_REPO_ID = f"{SPACE_ID}-data"
@@ -95,6 +107,7 @@ elif not HF_REPO_ID and HF_TOKEN:
         print(f"[SYNC] WARNING: Could not derive username from HF_TOKEN: {e}")
         HF_REPO_ID = ""
 
+# Setup logging
 log_dir = OPENCLAW_HOME / "workspace"
 log_dir.mkdir(parents=True, exist_ok=True)
 sys.stdout = TeeLogger(log_dir / "sync.log", sys.stdout)
@@ -109,7 +122,9 @@ TELEGRAM_API_BASES = [
     "https://telegram-api-proxy-anonymous.pages.dev/api",
 ]
 
+
 def probe_telegram_api(timeout: int = 8) -> str:
+    """Probe Telegram API endpoints and return the first reachable one."""
     ctx = ssl.create_default_context()
     for base in TELEGRAM_API_BASES:
         url = base.rstrip("/") + "/"
@@ -127,9 +142,12 @@ def probe_telegram_api(timeout: int = 8) -> str:
     print("[TELEGRAM] WARNING: All API endpoints unreachable!")
     return ""
 
+
 # ── Sync Manager ────────────────────────────────────────────────────────────
 
 class OpenClawFullSync:
+    """Upload/download the entire ~/.openclaw directory to HF Dataset."""
+
     def __init__(self):
         self.enabled = False
         self.dataset_exists = False
@@ -209,6 +227,8 @@ class OpenClawFullSync:
                             dest.parent.mkdir(parents=True, exist_ok=True)
                             shutil.copy2(str(item), str(dest))
                     print("[SYNC] ✓ Restore completed.")
+                else:
+                    print("[SYNC] Downloaded snapshot but dir not found. Starting fresh.")
 
         except Exception as e:
             print(f"[SYNC] ✗ Restore failed: {e}")
@@ -303,12 +323,27 @@ class OpenClawFullSync:
                 token=HF_TOKEN,
                 commit_message=f"Sync .openclaw — {datetime.now().isoformat()}",
                 ignore_patterns=[
-                    "*.log", "*.lock", "*.tmp", "*.pid", "__pycache__", "**/__pycache__/**",
-                    "*.bak", "*.migrated", "*.migrated.*", "**/*.migrated", "**/*.migrated.*",
-                    "*.pre-startup-migration-*", "**/*.pre-startup-migration-*", "**/*.corrupt_*",
-                    "**/*.sqlite-wal", "**/*.sqlite-shm", "**/*.sqlite-journal",
-                    "npm/**", "**/node_modules/**", "**/.npm/**", "**/.cache/**", "**/.pnpm-store/**",
+                    # Logs & runtime junk
+                    "*.log", "*.lock", "*.tmp", "*.pid",
+                    "__pycache__", "**/__pycache__/**",
+                    # Doctor-generated archives & backups
+                    "*.bak",
+                    "**/*.migrated",
+                    "**/*.migrated.*",
+                    "*.pre-startup-migration-*",
+                    "**/*.pre-startup-migration-*",
+                    "**/*.corrupt_*",
+                    # SQLite databases: MUST NOT be synced (inconsistent if written during upload)
+                    "**/*.sqlite",
+                    "**/*.sqlite-wal",
+                    "**/*.sqlite-shm",
+                    "**/*.sqlite-journal",
+                    # Plugin / npm caches
+                    "npm/**", "**/node_modules/**", "**/.npm/**",
+                    "**/.cache/**", "**/.pnpm-store/**",
+                    # Extensions shadow copies
                     "extensions/**", "**/extensions/**",
+                    # Transient runtime state
                     "**/sessions/*.tmp", "**/tmp/**", "**/.DS_Store",
                 ],
             )
@@ -316,33 +351,86 @@ class OpenClawFullSync:
             self._last_upload_time = time.time()
             self._last_dir_signature = sig
 
+            try:
+                files = self.api.list_repo_files(repo_id=HF_REPO_ID, repo_type="dataset")
+                oc_files = [f for f in files if f.startswith(f"{DATASET_PATH}/")]
+                print(f"[SYNC] Dataset now has {len(oc_files)} files under {DATASET_PATH}/")
+            except Exception:
+                pass
+
         except Exception as e:
             print(f"[SYNC] ✗ Upload failed: {str(e)[:500]}")
             if "429" in str(e) or "rate limit" in str(e).lower():
                 self._cooldown_until = time.time() + 65 * 60
                 print("[SYNC] Hub rate limit hit — pausing uploads for 65 min")
 
-    # ── Pre-gateway legacy cleanup (NEW) ──────────────────────────────
+    # ── Pre-gateway legacy cleanup (CRITICAL) ─────────────────────────
 
     def _purge_legacy_state(self):
-        """Remove files and SQLite rows that block 2026.9.7 gateway startup."""
+        """Remove files and SQLite databases that block gateway startup."""
         print("[SYNC] ── Purging legacy state that blocks gateway startup ──")
 
-        # 1. Empty Telegram thread-bindings file (issue #161795)
+        park_dir = Path("/tmp/openclaw-parked-migrated")
+        park_dir.mkdir(parents=True, exist_ok=True)
+
+        # 1. Park all .migrated archive files (doctor refuses if these exist)
+        parked = 0
+        for pattern in ("**/*.migrated", "**/*.migrated.*"):
+            for f in OPENCLAW_HOME.glob(pattern):
+                if not f.is_file():
+                    continue
+                try:
+                    dest = park_dir / (f.name + f".{int(time.time())}")
+                    shutil.move(str(f), str(dest))
+                    parked += 1
+                except Exception as e:
+                    print(f"[SYNC] Could not park {f.name}: {e}")
+        if parked:
+            print(f"[SYNC] Parked {parked} stale .migrated archive(s)")
+
+        # 2. Park inconsistent SQLite databases (caused the 303s read-only timeout)
+        sqlite_parked = 0
+        for rel in ("state/openclaw.sqlite",
+                    "state/openclaw.sqlite-wal",
+                    "state/openclaw.sqlite-shm",
+                    "state/openclaw.sqlite-journal"):
+            src = OPENCLAW_HOME / rel
+            if src.exists():
+                try:
+                    dest = park_dir / (rel.replace("/", "_") + f".{int(time.time())}")
+                    shutil.move(str(src), str(dest))
+                    sqlite_parked += 1
+                    print(f"[SYNC] Parked {rel}")
+                except Exception as e:
+                    print(f"[SYNC] Could not park {rel}: {e}")
+        if sqlite_parked:
+            print(f"[SYNC] Parked {sqlite_parked} SQLite file(s) — gateway will create fresh DB")
+
+        # 3. Empty Telegram thread-bindings file (issue #161795)
         tb = OPENCLAW_HOME / "telegram" / "thread-bindings-default.json"
         if tb.exists():
             try:
                 content = tb.read_text().strip()
-                if content in ('{"version":1,"bindings":[]}', '{"version":1,"bindings":[]}\n'):
-                    backup = tb.with_suffix(f".blocked_{int(time.time())}")
-                    shutil.move(str(tb), str(backup))
+                if content in ('{"version":1,"bindings":[]}',
+                               '{"version":1,"bindings":[]}\n'):
+                    shutil.move(str(tb), str(park_dir / (tb.name + f".{int(time.time())}")))
                     print(f"[SYNC] Removed blocking empty thread-bindings file")
-                else:
-                    print(f"[SYNC] thread-bindings exists but has content — leaving")
             except Exception as e:
                 print(f"[SYNC] Could not process thread-bindings: {e}")
 
-        # 2. Stale gateway owner lease in state/openclaw.sqlite
+        # 4. Also park retired Telegram JSON state (doctor demands 2026.9.5 upgrade)
+        for rel in ("telegram/update-offset-default.json",
+                    "agents/main/sessions/sessions.json.telegram-sent-messages.json"):
+            src = OPENCLAW_HOME / rel
+            if src.exists():
+                try:
+                    dest = park_dir / (rel.replace("/", "_") + f".{int(time.time())}")
+                    shutil.move(str(src), str(dest))
+                    print(f"[SYNC] Parked retired Telegram state: {rel}")
+                except Exception as e:
+                    print(f"[SYNC] Could not park {rel}: {e}")
+
+        # 5. Stale gateway owner lease in state/openclaw.sqlite (if not parked)
         db_path = OPENCLAW_HOME / "state" / "openclaw.sqlite"
         if db_path.exists():
             try:
@@ -355,12 +443,12 @@ class OpenClawFullSync:
                     if count > 0:
                         cur.execute("DELETE FROM state_leases")
                         conn.commit()
-                        print(f"[SYNC] Cleared {count} stale gateway lease row(s) from state_leases")
+                        print(f"[SYNC] Cleared {count} stale gateway lease row(s)")
                 conn.close()
             except Exception as e:
                 print(f"[SYNC] Could not purge state_leases: {e}")
 
-        # 3. Remove whatsapp/perplexity plugin entries that expand plugins.allow
+        # 6. Remove plugin entries that expand plugins.allow
         cfg_path = OPENCLAW_HOME / "openclaw.json"
         if cfg_path.exists():
             try:
@@ -374,11 +462,18 @@ class OpenClawFullSync:
                         (pid == "a2a-gateway" and A2A_PEERS)
                     ):
                         entries.pop(pid)
-                        removed.append(pid)
+                        removed.append(f"plugins.entries.{pid}")
+
+                # Remove whatsapp channel config so doctor doesn't re-add the plugin
+                chans = data.get("channels", {})
+                if "whatsapp" in chans:
+                    chans.pop("whatsapp")
+                    removed.append("channels.whatsapp")
+
                 if removed:
                     with open(cfg_path, "w") as f:
                         json.dump(data, f, indent=2)
-                    print(f"[SYNC] Removed plugin entries that expand allowlist: {removed}")
+                    print(f"[SYNC] Removed entries that expand allowlist: {removed}")
             except Exception as e:
                 print(f"[SYNC] Could not clean plugin entries: {e}")
 
@@ -533,13 +628,13 @@ class OpenClawFullSync:
                 "config": {"permissionMode": "approve-all"}
             }
 
-            # Telegram
+            # Telegram plugin
             if "telegram" not in data["plugins"]["entries"]:
                 data["plugins"]["entries"]["telegram"] = {"enabled": True}
             elif isinstance(data["plugins"]["entries"]["telegram"], dict):
                 data["plugins"]["entries"]["telegram"]["enabled"] = True
 
-            # A2A
+            # A2A plugin
             if A2A_PEERS:
                 peers = []
                 for peer_url in A2A_PEERS.split(","):
@@ -642,7 +737,7 @@ class OpenClawFullSync:
 
     def run_doctor_fix(self):
         if os.environ.get("SKIP_DOCTOR") == "1":
-            print("[SYNC] SKIP_DOCTOR=1 — skipping doctor")
+            print("[SYNC] SKIP_DOCTOR=1 — skipping doctor entirely")
             return True
         if not (APP_DIR / "dist" / "entry.js").exists():
             return True
@@ -840,6 +935,7 @@ class OpenClawFullSync:
             traceback.print_exc()
             return None
 
+
 # ── Main ────────────────────────────────────────────────────────────────────
 
 def main():
@@ -855,14 +951,15 @@ def main():
         sync.load_from_repo()
         print(f"[TIMER] load_from_repo (restore): {time.time() - t0:.1f}s")
 
-        # 2. Purge legacy state BEFORE doctor
+        # 2. Purge legacy state BEFORE doctor (park .migrated + SQLite + telegram state)
         t0 = time.time()
         sync._purge_legacy_state()
-        print(f"[TIMER] purge_legacy_state: {time.time() - t0:.1f}s")
+        print(f"[TIMER] pre-doctor purge: {time.time() - t0:.1f}s")
 
-        # 3. Run doctor
+        # 3. Run doctor (skipped if SKIP_DOCTOR=1)
         t0 = time.time()
         sync.run_doctor_fix()
+        print(f"[TIMER] doctor: {time.time() - t0:.1f}s")
 
         # 4. Re-enforce config AFTER doctor (doctor may have expanded allowlist)
         t0 = time.time()
@@ -882,7 +979,7 @@ def main():
         print(f"[TIMER] run_openclaw launch: {time.time() - t0:.1f}s")
         print(f"[TIMER] Total startup (init → app launched): {time.time() - t_main_start:.1f}s")
 
-        # 7. Optional conversation-loop
+        # 7. Optional conversation-loop (RUN_ORCHESTRATOR=1)
         conv_loop_proc = None
         if os.environ.get("RUN_ORCHESTRATOR") == "1":
             def run_conversation_loop_forever():