Merge branch 'main' of github.com:activeloopai/deeplake

khustup2 · khustup2 · commit d5f1741dbc7e · 2026-02-10T23:45:35.000Z
diff --git a/cpp/deeplake_pg/extension_init.cpp b/cpp/deeplake_pg/extension_init.cpp
@@ -238,8 +238,8 @@ void initialize_guc_parameters()
                              "allowing multiple PostgreSQL instances to share the same tables. "
                              "This adds latency for remote storage (S3, GCS) due to catalog sync operations.",
                              &pg::stateless_enabled,
-                             false,
-                             PGC_USERSET,
+                             true,
+                             PGC_POSTMASTER,
                              0,
                              nullptr,
                              nullptr,
diff --git a/postgres/tests/py_tests/test_startup_latency.py b/postgres/tests/py_tests/test_startup_latency.py
@@ -100,7 +100,6 @@ async def measure_connection_latency(
     database: str = "postgres",
     with_extension: bool = True,
     root_path: Optional[str] = None,
-    stateless_enabled: bool = False,
     run_first_query: bool = True,
     create_table: bool = False,
     table_name: str = "latency_test",
@@ -113,7 +112,6 @@ async def measure_connection_latency(
         database: Database to connect to
         with_extension: Whether to load pg_deeplake extension
         root_path: If set, configure deeplake.root_path
-        stateless_enabled: Whether to enable stateless mode
         run_first_query: Whether to measure first query time
         create_table: Whether to measure table creation time
         table_name: Name for test table
@@ -144,10 +142,6 @@ async def measure_connection_latency(
             await conn.execute("CREATE EXTENSION pg_deeplake")
             metrics.extension_load_time_ms = (time.perf_counter() - ext_start) * 1000
 
-            # Set stateless mode if requested
-            if stateless_enabled:
-                await conn.execute("SET deeplake.stateless_enabled = true")
-
         # 3. Measure root_path set time (triggers catalog loading in stateless mode)
         if root_path:
             root_start = time.perf_counter()
@@ -188,7 +182,6 @@ async def measure_catalog_discovery_latency(
     port: int,
     root_path: str,
     num_tables: int,
-    stateless_enabled: bool = True,
 ) -> LatencyMetrics:
     """
     Measure time to discover existing tables from catalog.
@@ -214,8 +207,6 @@ async def measure_catalog_discovery_latency(
         ext_start = time.perf_counter()
         await conn.execute("DROP EXTENSION IF EXISTS pg_deeplake CASCADE")
         await conn.execute("CREATE EXTENSION pg_deeplake")
-        if stateless_enabled:
-            await conn.execute("SET deeplake.stateless_enabled = true")
         metrics.extension_load_time_ms = (time.perf_counter() - ext_start) * 1000
 
         # Set root_path - this triggers catalog discovery
@@ -349,7 +340,6 @@ async def test_stateless_catalog_loading_latency(pg_server, temp_root_path):
     try:
         await setup_conn.execute("DROP EXTENSION IF EXISTS pg_deeplake CASCADE")
         await setup_conn.execute("CREATE EXTENSION pg_deeplake")
-        await setup_conn.execute("SET deeplake.stateless_enabled = true")
         await setup_conn.execute(f"SET deeplake.root_path = '{temp_root_path}'")
 
         # Create multiple tables to populate the catalog
@@ -383,7 +373,6 @@ async def test_stateless_catalog_loading_latency(pg_server, temp_root_path):
             port=5432,
             root_path=temp_root_path,
             num_tables=num_tables,
-            stateless_enabled=True,
         )
         report.add(metrics)
         print(f"Run {i+1}:")
@@ -425,7 +414,6 @@ async def test_stateless_vs_nonstateless_comparison(pg_server, temp_root_path):
         metrics = await measure_connection_latency(
             with_extension=True,
             root_path=temp_root_path,
-            stateless_enabled=False,
             run_first_query=True,
             create_table=True,
             table_name=f"nonstateless_test_{i}",
@@ -441,7 +429,6 @@ async def test_stateless_vs_nonstateless_comparison(pg_server, temp_root_path):
         metrics = await measure_connection_latency(
             with_extension=True,
             root_path=temp_root_path,
-            stateless_enabled=True,
             run_first_query=True,
             create_table=True,
             table_name=f"stateless_test_{i}",
@@ -492,7 +479,6 @@ async def test_multi_table_catalog_scaling(pg_server, temp_root_path):
         try:
             await setup_conn.execute("DROP EXTENSION IF EXISTS pg_deeplake CASCADE")
             await setup_conn.execute("CREATE EXTENSION pg_deeplake")
-            await setup_conn.execute("SET deeplake.stateless_enabled = true")
             await setup_conn.execute(f"SET deeplake.root_path = '{temp_root_path}'")
 
             # Create tables
@@ -512,7 +498,6 @@ async def test_multi_table_catalog_scaling(pg_server, temp_root_path):
                 port=5432,
                 root_path=temp_root_path,
                 num_tables=num_tables,
-                stateless_enabled=True,
             )
             report.add(metrics)
 
@@ -593,7 +578,6 @@ async def test_cold_start_simulation(pg_server, temp_root_path):
     try:
         await setup_conn.execute("DROP EXTENSION IF EXISTS pg_deeplake CASCADE")
         await setup_conn.execute("CREATE EXTENSION pg_deeplake")
-        await setup_conn.execute("SET deeplake.stateless_enabled = true")
         await setup_conn.execute(f"SET deeplake.root_path = '{temp_root_path}'")
 
         await setup_conn.execute("""
@@ -633,10 +617,6 @@ async def test_cold_start_simulation(pg_server, temp_root_path):
         try:
             # Extension is already loaded via shared_preload_libraries
             # Just configure the session (simulating a new backend)
-            ext_start = time.perf_counter()
-            await conn.execute("SET deeplake.stateless_enabled = true")
-            ext_time = (time.perf_counter() - ext_start) * 1000
-
             root_start = time.perf_counter()
             await conn.execute(f"SET deeplake.root_path = '{temp_root_path}'")
             root_time = (time.perf_counter() - root_start) * 1000
@@ -650,7 +630,6 @@ async def test_cold_start_simulation(pg_server, temp_root_path):
 
             print(f"\nRun {run + 1}:")
             print(f"  Connection:       {conn_time:8.2f} ms")
-            print(f"  Session config:   {ext_time:8.2f} ms")
             print(f"  Root path set:    {root_time:8.2f} ms")
             print(f"  First query:      {query_time:8.2f} ms (count={count})")
             print(f"  TOTAL COLD START: {total_time:8.2f} ms")
diff --git a/postgres/tests/py_tests/test_stateless_catalog_resilience.py b/postgres/tests/py_tests/test_stateless_catalog_resilience.py
@@ -28,8 +28,6 @@ async def test_stateless_bootstrap_permission_error_keeps_backend_alive(db_conn:
     - SET deeplake.root_path fails with a PostgreSQL error
     - Same connection remains usable afterwards
     """
-    await db_conn.execute("SET deeplake.stateless_enabled = true")
-
     readonly_root = Path(temp_dir_for_postgres) / "readonly_root"
     readonly_root.mkdir(parents=True, exist_ok=True)
     os.chmod(readonly_root, 0o555)
diff --git a/postgres/tests/py_tests/test_stateless_multi_instance.py b/postgres/tests/py_tests/test_stateless_multi_instance.py
@@ -213,7 +213,6 @@ async def primary_conn(pg_server):
         # Setup: Clean extension state
         await conn.execute("DROP EXTENSION IF EXISTS pg_deeplake CASCADE")
         await conn.execute("CREATE EXTENSION pg_deeplake")
-        await conn.execute("SET deeplake.stateless_enabled = true")
         yield conn
     finally:
         await conn.close()
@@ -323,7 +322,7 @@ async def test_stateless_data_sync_between_instances(
     try:
         # Setup extension (create if not exists for session-scoped instance reuse)
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
 
         # Setting root_path should automatically discover and register tables from catalog
         await conn_b.execute(f"SET deeplake.root_path = '{shared_root_path}'")
@@ -412,7 +411,7 @@ async def test_stateless_concurrent_writes(
     conn_b = await second_instance.connect()
     try:
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
 
         # Setting root_path should auto-discover tables from deeplake catalog
         await conn_b.execute(f"SET deeplake.root_path = '{shared_root_path}'")
@@ -515,7 +514,7 @@ async def test_stateless_multiple_tables_discovery(
     conn_b = await second_instance.connect()
     try:
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
 
         # Setting root_path should auto-discover ALL tables from deeplake catalog
         await conn_b.execute(f"SET deeplake.root_path = '{shared_root_path}'")
@@ -683,7 +682,7 @@ async def test_stateless_varchar1_catalog_sync(
     conn_b = await second_instance.connect()
     try:
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
         await conn_b.execute(f"SET deeplake.root_path = '{shared_root_path}'")
 
         # Verify table was auto-discovered
diff --git a/postgres/tests/py_tests/test_stateless_reserved_schema.py b/postgres/tests/py_tests/test_stateless_reserved_schema.py
@@ -75,7 +75,6 @@ async def primary_conn(pg_server):
     try:
         await conn.execute("DROP EXTENSION IF EXISTS pg_deeplake CASCADE")
         await conn.execute("CREATE EXTENSION pg_deeplake")
-        await conn.execute("SET deeplake.stateless_enabled = true")
         yield conn
     finally:
         await conn.close()
@@ -144,7 +143,7 @@ async def test_catalog_sync_default_schema(
 
     try:
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
 
         # This is the critical part - setting root_path triggers catalog sync
         # which should properly quote "default" schema name in generated DDL
@@ -236,7 +235,7 @@ async def test_catalog_sync_multiple_reserved_schemas(
 
     try:
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
         await conn_b.execute(f"SET deeplake.root_path = '{shared_root_path}'")
 
         # Verify all tables discovered
@@ -322,7 +321,7 @@ async def test_catalog_sync_default_schema_with_indexes(
 
     try:
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
         await conn_b.execute(f"SET deeplake.root_path = '{shared_root_path}'")
 
         # Verify table discovered
@@ -378,7 +377,7 @@ async def test_catalog_sync_default_schema_write_from_secondary(
 
     try:
         await conn_b.execute("CREATE EXTENSION IF NOT EXISTS pg_deeplake")
-        await conn_b.execute("SET deeplake.stateless_enabled = true")
+
         await conn_b.execute(f"SET deeplake.root_path = '{shared_root_path}'")
 
         # Insert from Instance B
