PYTHON-6135 Fix perf regression by explicitly choosing the python binary

.evergreen/generated_configs/functions.yml

@@ -102,6 +102,7 @@ functions:
           - SSL
           - ORCHESTRATION_FILE
           - UV_PYTHON
+          - UV_PYTHON_PREFERENCE
           - STORAGE_ENGINE
           - REQUIRE_API_VERSION
           - DRIVERS_TOOLS
@@ -137,6 +138,7 @@ functions:
           - AWS_SESSION_TOKEN
           - COVERAGE
           - UV_PYTHON
+          - UV_PYTHON_PREFERENCE
           - LIBMONGOCRYPT_URL
           - MONGODB_URI
           - DISABLE_TEST_COMMANDS

.evergreen/generated_configs/tasks.yml

@@ -2784,10 +2784,14 @@ tasks:
         vars:
           VERSION: v8.0-perf
           SSL: ssl
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: run tests
         vars:
           TEST_NAME: perf
           SUB_TEST_NAME: sync
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: attach benchmark test results
       - func: send dashboard data
     tags: [perf]
@@ -2797,10 +2801,14 @@ tasks:
         vars:
           VERSION: v8.0-perf
           SSL: ssl
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: run tests
         vars:
           TEST_NAME: perf
           SUB_TEST_NAME: async
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: attach benchmark test results
       - func: send dashboard data
     tags: [perf]
@@ -2810,10 +2818,14 @@ tasks:
         vars:
           VERSION: v8.0-perf
           SSL: nossl
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: run tests
         vars:
           TEST_NAME: perf
           SUB_TEST_NAME: sync
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: attach benchmark test results
       - func: send dashboard data
     tags: [perf]
@@ -2823,10 +2835,14 @@ tasks:
         vars:
           VERSION: v8.0-perf
           SSL: nossl
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: run tests
         vars:
           TEST_NAME: perf
           SUB_TEST_NAME: async
+          UV_PYTHON: 3.10.11
+          UV_PYTHON_PREFERENCE: only-managed
       - func: attach benchmark test results
       - func: send dashboard data
     tags: [perf]

.evergreen/run-tests.sh

@@ -26,6 +26,9 @@ else
   exit 1
 fi
 
+# No-op on Evergreen (just.sh already ran this); needed for direct/local runs (PYTHON-6135).
+. $SCRIPT_DIR/scripts/setup-uv-python.sh
+
 # Start the test runner.
 echo "Running tests with UV_PYTHON=${UV_PYTHON:-} UV_PYTHON_SEARCH_PATH=${UV_PYTHON_SEARCH_PATH:-}..."
 echo "UV_ARGS=${UV_ARGS}"

.evergreen/scripts/generate_config.py

@@ -43,6 +43,7 @@
     git_get_project,
 )
 from shrub.v3.evg_task import EvgTask, EvgTaskDependency, EvgTaskRef
+from utils import PERF_PYTHON_VERSION
 
 ##############
 # Variants
@@ -1092,9 +1093,12 @@ def create_search_index_tasks():
 def create_perf_tasks():
     tasks = []
     for version, ssl, sync in product(["8.0"], ["ssl", "nossl"], ["sync", "async"]):
-        vars = dict(VERSION=f"v{version}-perf", SSL=ssl)
+        # The toolchain build may be this same patch version but is not
+        # optimized, which regressed CPU-bound benchmarks (PYTHON-6135).
+        perf_vars = dict(UV_PYTHON=PERF_PYTHON_VERSION, UV_PYTHON_PREFERENCE="only-managed")
+        vars = dict(VERSION=f"v{version}-perf", SSL=ssl, **perf_vars)
         server_func = FunctionCall(func="run server", vars=vars)
-        vars = dict(TEST_NAME="perf", SUB_TEST_NAME=sync)
+        vars = dict(TEST_NAME="perf", SUB_TEST_NAME=sync, **perf_vars)
         test_func = FunctionCall(func="run tests", vars=vars)
         attach_func = FunctionCall(func="attach benchmark test results")
         send_func = FunctionCall(func="send dashboard data")
@@ -1313,6 +1317,7 @@ def create_run_server_func():
         "SSL",
         "ORCHESTRATION_FILE",
         "UV_PYTHON",
+        "UV_PYTHON_PREFERENCE",
         "STORAGE_ENGINE",
         "REQUIRE_API_VERSION",
         "DRIVERS_TOOLS",
@@ -1342,6 +1347,7 @@ def create_run_tests_func():
         "AWS_SESSION_TOKEN",
         "COVERAGE",
         "UV_PYTHON",
+        "UV_PYTHON_PREFERENCE",
         "LIBMONGOCRYPT_URL",
         "MONGODB_URI",
         "DISABLE_TEST_COMMANDS",

.evergreen/scripts/setup-uv-python.sh

@@ -9,6 +9,7 @@
 #   UV_PYTHON_SEARCH_PATH - the Python toolchain bin dir for the request, so uv
 #                           uses the toolchain Python instead of downloading one.
 #   UV_PYTHON_PREFERENCE  - "system" so the toolchain wins over managed installs.
+#                           A preference set by the task is left alone.
 #   UV_PYTHON             - the Python interpreter uv uses; defaults to CPython
 #                           3.10 when the task does not set one.
 set -euo pipefail
@@ -85,11 +86,11 @@ elif _dir=$(_toolchain_dir "$UV_PYTHON"); then
   _search_path="$_dir"
 fi
 
-# Point uv at the toolchain Python when there is one. On CI the toolchain dir
-# is already first on PATH (configure-env.sh), so this mainly benefits local
-# hosts and later steps, keeping `uv sync` and `uv tool install` on the
-# toolchain interpreter instead of downloading a managed one.
-if [ -n "$_search_path" ]; then
+# Point uv at the toolchain Python when there is one and the task did not set
+# its own preference: a task-level UV_PYTHON_PREFERENCE wins (PYTHON-6135).
+# On CI the toolchain dir is already first on PATH (configure-env.sh), so this
+# mainly benefits local hosts and later steps.
+if [ -n "$_search_path" ] && [ -z "${UV_PYTHON_PREFERENCE:-}" ]; then
   export UV_PYTHON_SEARCH_PATH="$_search_path"
   export UV_PYTHON_PREFERENCE="system"
 fi

.evergreen/scripts/setup_tests.py

@@ -13,6 +13,7 @@
     ENV_FILE,
     HERE,
     LOGGER,
+    PERF_PYTHON_VERSION,
     PLATFORM,
     ROOT,
     TEST_SUITE_MAP,
@@ -51,9 +52,6 @@
 # Map the test name to test group.
 GROUP_MAP = dict(mockupdb="mockupdb", perf="perf")
 
-# The python version used for perf tests.
-PERF_PYTHON_VERSION = "3.10.11"
-
 # The libmongocrypt release used when LIBMONGOCRYPT_URL is not set. Must be at
 # least 1.20.0 for the GA "substring" query type.
 LIBMONGOCRYPT_VERSION = "1.20.2"
@@ -515,13 +513,12 @@ def handle_test_env() -> None:
             run_command("tar xf single_and_multi_document.tgz", cwd=data_dir)
         write_env("TEST_PATH", str(data_dir))
         write_env("OUTPUT_FILE", str(ROOT / "results.json"))
-        # Overwrite the UV_PYTHON value from env.sh, and unset the toolchain
-        # search-path variables: an empty value would make uv reject the request,
-        # and a toolchain path would miss the exact patch version requested.
-        write_env("UV_PYTHON", "")
+        # The toolchain build may be this same patch version but is not
+        # optimized, which regressed CPU-bound benchmarks (PYTHON-6135).
+        write_env("UV_PYTHON", PERF_PYTHON_VERSION)
+        write_env("UV_PYTHON_PREFERENCE", "only-managed")
         with ENV_FILE.open("a", newline="\n") as fid:
             fid.write("unset UV_PYTHON_SEARCH_PATH\n")
-            fid.write("unset UV_PYTHON_PREFERENCE\n")
 
         UV_ARGS.append(f"--python={PERF_PYTHON_VERSION}")
 

.evergreen/scripts/utils.py

@@ -81,6 +81,9 @@ class Distro:
 
 EXTRA_TESTS = ["aws_lambda", "doctest"]
 
+# The managed Python version used for perf tests (PYTHON-6135).
+PERF_PYTHON_VERSION = "3.10.11"
+
 # Tests that do not use run-mongodb directly.
 NO_RUN_ORCHESTRATION = [
     "auth_oidc",