Fix bug 5811173: update cufile tests and configuration

cuda_bindings/tests/cufile.json

@@ -3,6 +3,10 @@
     // e.g : export CUFILE_ENV_PATH_JSON="/home/<xxx>/cufile.json"
 
 
+            "properties" : {
+                    "allow_compat_mode" : true
+            },
+
             "execution" : {
                     // max number of workitems in the queue;
                     "max_io_queue_depth": 128,

cuda_bindings/tests/test_cufile.py

@@ -9,7 +9,7 @@
 import platform
 import subprocess
 import tempfile
-from contextlib import suppress
+from contextlib import contextmanager, suppress
 from functools import cache
 
 import pytest
@@ -28,6 +28,16 @@
 cufile = pytest.importorskip("cuda.bindings.cufile", reason="skipping tests on Windows")
 
 
+@contextmanager
+def _cufile_driver_session():
+    """Open the cuFile driver for a block; always close in a finally (mirrors try/finally)."""
+    cufile.driver_open()
+    try:
+        yield
+    finally:
+        cufile.driver_close()
+
+
 @pytest.fixture
 def cufile_env_json(monkeypatch):
     """Set CUFILE_ENV_PATH_JSON environment variable for async tests."""
@@ -1394,8 +1404,13 @@ def test_set_get_parameter_size_t():
         (cufile.SizeTConfigParameter.EXECUTION_MAX_REQUEST_PARALLELISM, 4),  # Max 4 parallel requests
     )
 
+    # Snapshot baselines after driver_open so getters reflect merged config (defaults + JSON),
+    # not pre-open pending state that could restore invalid values (e.g. 0 for per-buffer cache).
+    with _cufile_driver_session():
+        originals = {param: cufile.get_parameter_size_t(param) for param, _ in param_val_pairs}
+
     def test_param(param, val):
-        orig_val = cufile.get_parameter_size_t(param)
+        orig_val = originals[param]
         cufile.set_parameter_size_t(param, val)
         retrieved_val = cufile.get_parameter_size_t(param)
         assert retrieved_val == val
@@ -1412,6 +1427,14 @@ def test_param(param, val):
 @pytest.mark.usefixtures("ctx")
 def test_set_get_parameter_bool():
     """Test setting and getting boolean parameters with cuFile validation."""
+    # Do not exercise allow/force compat via set_parameter_bool before any driver_open:
+    # pending API values are applied after JSON load on first open and can overwrite
+    # cufile.json (e.g. allow_compat_mode: true), causing DRIVER_NOT_INITIALIZED when
+    # nvidia-fs is not loaded. Other tests cover compat behavior where appropriate.
+    _COMPAT_PARAMS = (
+        cufile.BoolConfigParameter.PROPERTIES_ALLOW_COMPAT_MODE,
+        cufile.BoolConfigParameter.FORCE_COMPAT_MODE,
+    )
     param_val_pairs = (
         (cufile.BoolConfigParameter.PROPERTIES_USE_POLL_MODE, True),
         (cufile.BoolConfigParameter.PROPERTIES_ALLOW_COMPAT_MODE, False),
@@ -1426,22 +1449,26 @@ def test_set_get_parameter_bool():
         (cufile.BoolConfigParameter.SKIP_TOPOLOGY_DETECTION, False),
         (cufile.BoolConfigParameter.STREAM_MEMOPS_BYPASS, True),
     )
+    param_val_pairs = tuple((p, v) for p, v in param_val_pairs if p not in _COMPAT_PARAMS)
+    # PROFILE_NVTX is deprecated (CTK 13.1.0+); cuFile >= 1.16 rejects bool getters for it.
+    if cufile.get_version() >= 1160:
+        param_val_pairs = tuple(
+            (p, v) for p, v in param_val_pairs if p is not cufile.BoolConfigParameter.PROFILE_NVTX
+        )
+
+    with _cufile_driver_session():
+        originals = {param: cufile.get_parameter_bool(param) for param, _ in param_val_pairs}
 
     def test_param(param, val):
-        orig_val = cufile.get_parameter_bool(param)
+        orig_val = originals[param]
         cufile.set_parameter_bool(param, val)
         retrieved_val = cufile.get_parameter_bool(param)
         assert retrieved_val is val
         cufile.set_parameter_bool(param, orig_val)
 
-    try:
-        # Test setting and getting various boolean parameters
-        for param, val in param_val_pairs:
-            test_param(param, val)
-    except cufile.cuFileError:
-        if cufile.get_version() < 1160:
-            raise
-        assert param is cufile.BoolConfigParameter.PROFILE_NVTX  # Deprecated in CTK 13.1.0
+    # Test setting and getting various boolean parameters
+    for param, val in param_val_pairs:
+        test_param(param, val)
 
 
 @pytest.mark.skipif(
@@ -1468,8 +1495,11 @@ def test_set_get_parameter_string(tmp_path):
         ),  # Test log directory
     )
 
+    with _cufile_driver_session():
+        originals = {param: cufile.get_parameter_string(param, 256) for param, _, _ in param_val_pairs}
+
     def test_param(param, val, default_val):
-        orig_val = cufile.get_parameter_string(param, 256)
+        orig_val = originals[param]
 
         val_b = val.encode("utf-8")
         val_buf = ctypes.create_string_buffer(val_b)
@@ -1908,11 +1938,8 @@ def test_set_parameter_posix_pool_slab_array(slab_sizes, slab_counts, driver_con
     retrieved_counts_addr = ctypes.addressof(retrieved_counts)
 
     # Open cuFile driver AFTER setting parameters
-    cufile.driver_open()
-    try:
+    with _cufile_driver_session():
         cufile.get_parameter_posix_pool_slab_array(retrieved_sizes_addr, retrieved_counts_addr, n_slab_sizes)
-    finally:
-        cufile.driver_close()
 
     # Verify they match what we set
     assert list(retrieved_sizes) == slab_sizes