Sync 2026-04-14 08:03

sync.sh

@@ -8,24 +8,32 @@ cd "$SYNC_DIR"
 
 echo "[$(date '+%H:%M:%S')] Sync from $HOSTNAME..."
 
-# ── Step 1: Export local state.db → state_<hostname>.db ──────────────────
+# ── Step 1: Export local state.db (via temp dir to avoid lock) ────────────
 python3 << PYEOF
-import sqlite3, os, shutil
+import sqlite3, os, shutil, tempfile
 
 local_db = os.path.join(os.environ['HERMES_HOME'], 'state.db')
 export_db = os.path.join(os.environ['SYNC_DIR'], f"state_{os.environ['HOSTNAME']}.db")
+tmpdir = tempfile.mkdtemp(prefix='hs_exp_')
 
-# Checkpoint WAL
 try:
     conn = sqlite3.connect(local_db)
     conn.execute('PRAGMA wal_checkpoint(TRUNCATE)')
     conn.execute('PRAGMA optimize')
     conn.close()
-except Exception as e:
-    print(f'Checkpoint: {e}', file=__import__('sys').stderr)
 
-shutil.copy2(local_db, export_db)
+    tmp_db = os.path.join(tmpdir, 'db')
-print(f'Exported: {export_db}')
+    shutil.copy2(local_db, tmp_db)
+
+    test = sqlite3.connect(tmp_db)
+    r = test.execute('SELECT COUNT(*) FROM sessions').fetchone()[0]
+    m = test.execute('SELECT COUNT(*) FROM messages').fetchone()[0]
+    test.close()
+
+    shutil.copy2(tmp_db, export_db)
+    print(f'Exported: {r}s/{m}m')
+finally:
+    shutil.rmtree(tmpdir, ignore_errors=True)
 PYEOF
 
 # ── Step 2: Git stage ────────────────────────────────────────────────────
@@ -65,42 +73,26 @@ fi
 
 # ── Step 4: Merge all state_*.db → state_merged.db ──────────────────────
 python3 << 'PYEOF'
-import sqlite3, os, glob, shutil
+import sqlite3, os, glob, shutil, tempfile
 
 sync_dir = os.environ['SYNC_DIR']
-hermes_home = os.environ['HERMES_HOME']
 merged_path = os.path.join(sync_dir, 'state_merged.db')
 db_files = sorted(glob.glob(os.path.join(sync_dir, 'state_*.db')))
 db_files = [f for f in db_files if not f.endswith('_merged.db')]
-print(f'DBs to merge: {[os.path.basename(f) for f in db_files]}')
+print(f'Merging {len(db_files)} DBs')
 
-if os.path.exists(merged_path):
+tmpdir = tempfile.mkdtemp(prefix='hs_merge_')
-    os.remove(merged_path)
+tmp_merged = os.path.join(tmpdir, 'merged.db')
 
-def get_schema(conn, table):
+try:
-    cols = conn.execute(f'PRAGMA table_info("{table}")').fetchall()
+    # Create merged DB with proper schema
-    return [c[1] for c in cols]  # column names
+    conn = sqlite3.connect(tmp_merged)
+    conn.execute('PRAGMA journal_mode=DELETE')
+    conn.execute('PRAGMA locking_mode=NORMAL')
+    conn.execute('PRAGMA synchronous=FULL')
 
-def copy_table(src_conn, dst_conn, table):
+    conn.execute('''
-    cols = get_schema(src_conn, table)
+    CREATE TABLE sessions (
-    placeholders = ','.join(['?'] * len(cols))
-    col_names = ','.join(f'"{c}"' for c in cols)
-    
-    rows = src_conn.execute(f'SELECT {col_names} FROM "{table}"').fetchall()
-    for row in rows:
-        dst_conn.execute(f'INSERT OR REPLACE INTO "{table}" ({col_names}) VALUES ({placeholders})', row)
-    return len(rows)
-
-# Open merged DB and rebuild from scratch
-conn_merged = sqlite3.connect(merged_path)
-conn_merged.execute('PRAGMA journal_mode=WAL')
-conn_merged.execute('PRAGMA synchronous=NORMAL')
-
-# We need at least one source to initialize schema
-if not db_files:
-    print('No state DBs found, creating empty merged DB')
-    conn_merged.execute('''
-    CREATE TABLE IF NOT EXISTS sessions (
         id TEXT PRIMARY KEY, source TEXT NOT NULL, user_id TEXT,
         model TEXT, model_config TEXT, system_prompt TEXT,
         parent_session_id TEXT, started_at REAL, ended_at REAL,
@@ -112,8 +104,9 @@ if not db_files:
         estimated_cost_usd REAL, actual_cost_usd REAL,
         cost_status TEXT, cost_source TEXT, pricing_version TEXT, title TEXT
     )''')
-    conn_merged.execute('''
+
-    CREATE TABLE IF NOT EXISTS messages (
+    conn.execute('''
+    CREATE TABLE messages (
         id INTEGER PRIMARY KEY,
         session_id TEXT NOT NULL, role TEXT NOT NULL, content TEXT,
         tool_call_id TEXT, tool_calls TEXT, tool_name TEXT,
@@ -121,40 +114,50 @@ if not db_files:
         finish_reason TEXT, reasoning TEXT, reasoning_details TEXT,
         codex_reasoning_items TEXT
     )''')
-    conn_merged.commit()
-else:
-    first_db = db_files[0]
-    src_conn = sqlite3.connect(first_db)
-    src_conn.execute('PRAGMA wal_checkpoint(TRUNCATE)')
 
-    for table in ['sessions', 'messages']:
+    conn.execute('CREATE INDEX idx_msg_session ON messages(session_id)')
-        cols = get_schema(src_conn, table)
+    conn.execute('CREATE INDEX idx_msg_ts ON messages(timestamp)')
-        col_defs = []
-        for c in cols:
-            if c == 'id' and table == 'messages':
-                col_defs.append('id INTEGER PRIMARY KEY')
-            elif c == 'id':
-                col_defs.append(f'{c} TEXT PRIMARY KEY')
-            else:
-                col_defs.append(f'{c} TEXT')
 
-        conn_merged.execute(f'DROP TABLE IF EXISTS "{table}"')
+    for db_file in db_files:
-        conn_merged.execute(f'CREATE TABLE "{table}" ({", ".join(col_defs)})')
+        name = os.path.basename(db_file)
+        # Copy to temp file first (avoids WAL lock issues with open connections)
+        tmp_copy = os.path.join(tmpdir, name)
+        shutil.copy2(db_file, tmp_copy)
+        # Also copy WAL if exists
+        if os.path.exists(db_file + '-wal'):
+            shutil.copy2(db_file + '-wal', tmp_copy + '-wal')
 
-        placeholders = ','.join(['?'] * len(cols))
+        src = sqlite3.connect(tmp_copy)
-        col_names = ','.join(f'"{c}"' for c in cols)
-        rows = src_conn.execute(f'SELECT {col_names} FROM "{table}"').fetchall()
-        for row in rows:
-            conn_merged.execute(f'INSERT OR REPLACE INTO "{table}" ({col_names}) VALUES ({placeholders})', row)
-        print(f'  {os.path.basename(first_db)}.{table}: {len(rows)} rows')
 
-    src_conn.close()
+        s_cnt = src.execute('SELECT COUNT(*) FROM sessions').fetchone()[0]
+        m_cnt = src.execute('SELECT COUNT(*) FROM messages').fetchone()[0]
+        print(f'  {name}: {s_cnt}s/{m_cnt}m')
 
-conn_merged.execute('PRAGMA wal_checkpoint(TRUNCATE)')
+        sess_rows = src.execute('SELECT * FROM sessions').fetchall()
-conn_merged.commit()
+        sess_cols = len(src.execute('PRAGMA table_info(sessions)').fetchall())
-conn_merged.close()
+        for row in sess_rows:
+            conn.execute(f'INSERT OR REPLACE INTO sessions VALUES ({",".join(["?"]*sess_cols)})', row)
 
-print(f'Merged: {merged_path} ({os.path.getsize(merged_path)/1024:.0f} KB)')
+        msg_rows = src.execute('SELECT * FROM messages').fetchall()
+        msg_cols = len(src.execute('PRAGMA table_info(messages)').fetchall())
+        for row in msg_rows:
+            conn.execute(f'INSERT OR IGNORE INTO messages VALUES ({",".join(["?"]*msg_cols)})', row)
+
+        src.close()
+        os.remove(tmp_copy)
+        if os.path.exists(tmp_copy + '-wal'):
+            os.remove(tmp_copy + '-wal')
+
+    conn.commit()
+    conn.close()
+
+    if os.path.exists(merged_path):
+        os.remove(merged_path)
+    shutil.copy2(tmp_merged, merged_path)
+    print(f'Merged: {os.path.getsize(merged_path)/1024:.0f} KB')
+
+finally:
+    shutil.rmtree(tmpdir, ignore_errors=True)
 PYEOF
 
 # ── Step 5: Push ─────────────────────────────────────────────────────────
@@ -171,36 +174,36 @@ fi
 
 # ── Step 6: Restore merged state to local hermes ─────────────────────────
 python3 << 'PYEOF'
-import sqlite3, os, shutil
+import sqlite3, os, shutil, tempfile
 
-sync_dir = os.environ['SYNC_DIR']
 hermes_home = os.environ['HERMES_HOME']
-merged_path = os.path.join(sync_dir, 'state_merged.db')
+merged_path = os.path.join(os.environ['SYNC_DIR'], 'state_merged.db')
 local_db = os.path.join(hermes_home, 'state.db')
 
 if not os.path.exists(merged_path):
     print('No merged DB, skipping restore')
 else:
-    # Checkpoint + close local
+    tmpdir = tempfile.mkdtemp(prefix='hs_rest_')
     try:
         conn = sqlite3.connect(local_db)
         conn.execute('PRAGMA wal_checkpoint(TRUNCATE)')
         conn.execute('PRAGMA optimize')
         conn.close()
-    except Exception as e:
-        print(f'Local checkpoint: {e}')
 
-    # Backup
+        tmp_db = os.path.join(tmpdir, 'db')
-    shutil.copy2(local_db, local_db + '.pre_sync_bak')
+        shutil.copy2(merged_path, tmp_db)
 
-    # Replace with merged
+        test = sqlite3.connect(tmp_db)
-    shutil.copy2(merged_path, local_db)
+        r = test.execute('SELECT COUNT(*) FROM sessions').fetchone()[0]
-    print(f'Restored merged state to {local_db}')
+        m = test.execute('SELECT COUNT(*) FROM messages').fetchone()[0]
+        test.close()
 
-    # Remove backup after successful restore
+        shutil.copy2(local_db, local_db + '.bak')
-    bak = local_db + '.pre_sync_bak'
+        shutil.copy2(tmp_db, local_db)
-    if os.path.exists(bak):
+        os.remove(local_db + '.bak')
-        os.remove(bak)
+        print(f'Restored: {r}s/{m}m')
+    finally:
+        shutil.rmtree(tmpdir, ignore_errors=True)
 PYEOF
 
 # ── Step 7: Sync memories + skills (additive) ────────────────────────────