Extend async tree benchmarks to cover eager task execution (#279)

Co-authored-by: Kumar Aditya <[email protected]>

doc/benchmarks.rst

@@ -66,12 +66,15 @@ Async workload benchmark, which calls ``asyncio.gather()`` on a tree (6 levels d
 
 * ``async_tree``: no actual async work at any leaf node.
 * ``async_tree_io``: all leaf nodes simulate async IO workload (async sleep 50ms).
-* ``async_tree_memoization``: all leaf nodes simulate async IO workload with 90% of 
+* ``async_tree_memoization``: all leaf nodes simulate async IO workload with 90% of
   the data memoized.
 * ``async_tree_cpu_io_mixed``: half of the leaf nodes simulate CPU-bound workload
-  (``math.factorial(500)``) and the other half simulate the same workload as the 
+  (``math.factorial(500)``) and the other half simulate the same workload as the
   ``async_tree_memoization`` variant.
 
+These benchmarks also have an "eager" flavor that uses asyncio eager task factory,
+if available.
+
 
 chameleon
 ---------

pyperformance/data-files/benchmarks/MANIFEST

@@ -7,6 +7,10 @@ async_tree	<local>
 async_tree_cpu_io_mixed	<local:async_tree>
 async_tree_io	<local:async_tree>
 async_tree_memoization	<local:async_tree>
+async_tree_eager	<local:async_tree>
+async_tree_eager_cpu_io_mixed	<local:async_tree>
+async_tree_eager_io	<local:async_tree>
+async_tree_eager_memoization	<local:async_tree>
 asyncio_tcp	<local>
 asyncio_tcp_ssl	<local:asyncio_tcp>
 concurrent_imap	<local>

pyperformance/data-files/benchmarks/bm_async_tree/bm_async_tree_cpu_io_mixed.toml

@@ -1,4 +1,3 @@
 [tool.pyperformance]
 name = "async_tree_cpu_io_mixed"
 extra_opts = ["cpu_io_mixed"]
-

pyperformance/data-files/benchmarks/bm_async_tree/bm_async_tree_eager.toml

@@ -0,0 +1,3 @@
+[tool.pyperformance]
+name = "async_tree_eager"
+extra_opts = ["eager"]

pyperformance/data-files/benchmarks/bm_async_tree/bm_async_tree_eager_cpu_io_mixed.toml

@@ -0,0 +1,3 @@
+[tool.pyperformance]
+name = "async_tree_eager_cpu_io_mixed"
+extra_opts = ["eager_cpu_io_mixed"]

pyperformance/data-files/benchmarks/bm_async_tree/bm_async_tree_eager_io.toml

@@ -0,0 +1,3 @@
+[tool.pyperformance]
+name = "async_tree_eager_io"
+extra_opts = ["eager_io"]

pyperformance/data-files/benchmarks/bm_async_tree/bm_async_tree_eager_memoization.toml

@@ -0,0 +1,3 @@
+[tool.pyperformance]
+name = "async_tree_eager_memoization"
+extra_opts = ["eager_memoization"]

pyperformance/data-files/benchmarks/bm_async_tree/bm_async_tree_io.toml

@@ -1,4 +1,3 @@
 [tool.pyperformance]
 name = "async_tree_io"
 extra_opts = ["io"]
-

pyperformance/data-files/benchmarks/bm_async_tree/bm_async_tree_memoization.toml

@@ -1,4 +1,3 @@
 [tool.pyperformance]
 name = "async_tree_memoization"
 extra_opts = ["memoization"]
-

pyperformance/data-files/benchmarks/bm_async_tree/run_benchmark.py

@@ -6,11 +6,14 @@
 
 1) "none": No actual async work in the async tree.
 2) "io": All leaf nodes simulate async IO workload (async sleep 50ms).
-3) "memoization": All leaf nodes simulate async IO workload with 90% of 
+3) "memoization": All leaf nodes simulate async IO workload with 90% of
                   the data memoized
-4) "cpu_io_mixed": Half of the leaf nodes simulate CPU-bound workload and 
-                   the other half simulate the same workload as the 
+4) "cpu_io_mixed": Half of the leaf nodes simulate CPU-bound workload and
+                   the other half simulate the same workload as the
                    "memoization" variant.
+
+All variants also have an "eager" flavor that uses
+the asyncio eager task factory (if available).
 """
 
 
@@ -57,16 +60,32 @@ async def run(self):
         await self.recurse(NUM_RECURSE_LEVELS)
 
 
+class EagerMixin:
+    async def run(self):
+        loop = asyncio.get_running_loop()
+        if hasattr(asyncio, 'eager_task_factory'):
+            loop.set_task_factory(asyncio.eager_task_factory)
+        return await super().run()
+
+
 class NoneAsyncTree(AsyncTree):
     async def workload_func(self):
         return
 
 
+class EagerAsyncTree(EagerMixin, NoneAsyncTree):
+    pass
+
+
 class IOAsyncTree(AsyncTree):
     async def workload_func(self):
         await self.mock_io_call()
 
 
+class EagerIOAsyncTree(EagerMixin, IOAsyncTree):
+    pass
+
+
 class MemoizationAsyncTree(AsyncTree):
     async def workload_func(self):
         # deterministic random, seed set in AsyncTree.__init__()
@@ -82,6 +101,10 @@ async def workload_func(self):
         return data
 
 
+class EagerMemoizationAsyncTree(EagerMixin, MemoizationAsyncTree):
+    pass
+
+
 class CpuIoMixedAsyncTree(MemoizationAsyncTree):
     async def workload_func(self):
         # deterministic random, seed set in AsyncTree.__init__()
@@ -92,6 +115,10 @@ async def workload_func(self):
             return await MemoizationAsyncTree.workload_func(self)
 
 
+class EagerCpuIoMixedAsyncTree(EagerMixin, CpuIoMixedAsyncTree):
+    pass
+
+
 def add_metadata(runner):
     runner.metadata["description"] = "Async tree workloads."
     runner.metadata["async_tree_recurse_levels"] = NUM_RECURSE_LEVELS
@@ -115,20 +142,24 @@ def add_parser_args(parser):
 Determines which benchmark to run. Options:
 1) "none": No actual async work in the async tree.
 2) "io": All leaf nodes simulate async IO workload (async sleep 50ms).
-3) "memoization": All leaf nodes simulate async IO workload with 90% of 
+3) "memoization": All leaf nodes simulate async IO workload with 90% of
                   the data memoized
-4) "cpu_io_mixed": Half of the leaf nodes simulate CPU-bound workload and 
-                   the other half simulate the same workload as the 
+4) "cpu_io_mixed": Half of the leaf nodes simulate CPU-bound workload and
+                   the other half simulate the same workload as the
                    "memoization" variant.
 """,
     )
 
 
 BENCHMARKS = {
     "none": NoneAsyncTree,
+    "eager": EagerAsyncTree,
     "io": IOAsyncTree,
+    "eager_io": EagerIOAsyncTree,
     "memoization": MemoizationAsyncTree,
+    "eager_memoization": EagerMemoizationAsyncTree,
     "cpu_io_mixed": CpuIoMixedAsyncTree,
+    "eager_cpu_io_mixed": EagerCpuIoMixedAsyncTree,
 }
 
 
@@ -142,4 +173,3 @@ def add_parser_args(parser):
     async_tree_class = BENCHMARKS[benchmark]
     async_tree = async_tree_class()
     runner.bench_async_func(f"async_tree_{benchmark}", async_tree.run)
-