[CI] Reduce CI_SERIAL_LIST list (#124085)

Add serial marker for individual tests so the test file can be removed from the ci serial list
Run serial marked tests first in serial
Run all other tests afterwards in parallel

Slowly reduce list and mark individual tests as serial instead

Hope # of serial tests is small so sharding evenness doesn't get too messed up

Hopefully can do 3 procs for sm86 and cpu?

serial no longer looks like a real word to me

Pull Request resolved: https://github.com/pytorch/pytorch/pull/124085
Approved by: https://github.com/seemethere, https://github.com/malfet

pytest.ini

@@ -19,3 +19,6 @@ filterwarnings =
     ignore:Module already imported so cannot be rewritten.*hypothesis:pytest.PytestAssertRewriteWarning
 
 xfail_strict = True
+
+markers =
+    serial: marks tests as needs to be run serially (deselect with '-m "not serial"')

test/inductor/test_torchinductor.py

@@ -70,6 +70,7 @@ from torch.testing._internal.common_utils import (
     IS_WINDOWS,
     IS_X86,
     parametrize,
+    serialTest,
     skipIfRocm,
     subtest,
     TEST_WITH_ASAN,
@@ -9278,6 +9279,7 @@ class CommonTemplate:
     @config.patch(
         "triton.autotune_pointwise", True
     )  # needed to introduce config that exceed max shared memory usage
+    @serialTest()
     def test_large_block_sizes(self):
         """
         Inductor will try triton configs like x = 64 and y = 1024 which will

test/run_test.py

@@ -246,9 +246,6 @@ CI_SERIAL_LIST = [
     "test_module_hooks",  # OOM
     "inductor/test_max_autotune",
     "inductor/test_cutlass_backend",  # slow due to many nvcc compilation steps
-    "inductor/test_torchinductor",  # OOM on test_large_block_sizes
-    "inductor/test_torchinductor_dynamic_shapes",  # OOM on test_large_block_sizes
-    "inductor/test_torchinductor_codegen_dynamic_shapes",  # OOM on test_large_block_sizes
     "test_profiler",  # test_source_multithreaded is probably not compatible with parallelism
 ]
 # A subset of onnx tests that cannot run in parallel due to high memory usage.
@@ -1591,6 +1588,11 @@ def run_tests(
         ):
             pool.terminate()
 
+    keep_going_message = (
+        "\n\nTip: You can keep running tests even on failure by passing --keep-going to run_test.py.\n"
+        "If running on CI, add the 'keep-going' label to your PR and rerun your jobs."
+    )
+
     try:
         for test in selected_tests_serial:
             options_clone = copy.deepcopy(options)
@@ -1603,19 +1605,29 @@ def run_tests(
                 and not options.continue_through_error
                 and not RERUN_DISABLED_TESTS
             ):
-                raise RuntimeError(
-                    failure.message
-                    + "\n\nTip: You can keep running tests even on failure by "
-                    "passing --keep-going to run_test.py.\n"
-                    "If running on CI, add the 'keep-going' label to "
-                    "your PR and rerun your jobs."
-                )
+                raise RuntimeError(failure.message + keep_going_message)
+
+        # Run tests marked as serial first
+        for test in selected_tests_parallel:
+            options_clone = copy.deepcopy(options)
+            if can_run_in_pytest(test):
+                options_clone.pytest = True
+            options_clone.additional_unittest_args.extend(["-m", "serial"])
+            failure = run_test_module(test, test_directory, options_clone)
+            test_failed = handle_error_messages(failure)
+            if (
+                test_failed
+                and not options.continue_through_error
+                and not RERUN_DISABLED_TESTS
+            ):
+                raise RuntimeError(failure.message + keep_going_message)
 
         os.environ["NUM_PARALLEL_PROCS"] = str(NUM_PROCS)
         for test in selected_tests_parallel:
             options_clone = copy.deepcopy(options)
             if can_run_in_pytest(test):
                 options_clone.pytest = True
+            options_clone.additional_unittest_args.extend(["-m", "not serial"])
             pool.apply_async(
                 run_test_module,
                 args=(test, test_directory, options_clone),
@@ -1718,6 +1730,7 @@ def main():
     if IS_CI:
         gen_ci_artifact([x.to_json() for x in include], [x.to_json() for x in exclude])
 
+    print_to_stderr(f"Running parallel tests on {NUM_PROCS} processes")
     print_to_stderr(test_batch)
     print_to_stderr(test_batch_exclude)
 

torch/testing/_internal/common_utils.py

@@ -97,6 +97,11 @@ from torch.utils._import_utils import _check_module_exists
 import torch.utils._pytree as pytree
 
 from .composite_compliance import no_dispatch
+try:
+    import pytest
+    has_pytest = True
+except ImportError:
+    has_pytest = False
 
 
 # Class to keep track of test flags configurable by environment variables.
@@ -1384,6 +1389,15 @@ def skipIfTorchInductor(msg="test doesn't currently work with torchinductor",
 
     return decorator
 
+def serialTest(condition=True):
+    """
+    Decorator for running tests serially.  Requires pytest
+    """
+    def decorator(fn):
+        if has_pytest and condition:
+            return pytest.mark.serial(fn)
+        return fn
+    return decorator
 
 def unMarkDynamoStrictTest(cls=None):
     def decorator(cls):