Randomize test order each run

A couple of dependencies discovered so far, and an apparent cleanup bug that I
don't know how to fix with the `caplog` fixture.  Drop using our `mock_log`
fixture.

[sc-46860]

Author: khk-globus

compute_endpoint/setup.py

@@ -51,6 +51,7 @@
     "coverage>=5.2",
     "pytest-mock==3.2.0",
     "pyfakefs<5.9.2",  # 5.9.2 (Jul 30, 2025), breaks us; retry after 6.0.0 lands?
+    "pytest-random-order",
 ]
 
 

compute_endpoint/tests/unit/test_execute_task.py

@@ -1,6 +1,7 @@
 import logging
 import os
 import random
+from itertools import chain
 from unittest import mock
 
 import pytest
@@ -69,14 +70,13 @@ def test_bad_run_dir(endpoint_uuid, task_uuid, run_dir):
         execute_task(task_uuid, b"", endpoint_uuid, run_dir=None)
 
 
-def test_happy_path(serde, caplog, task_uuid, ez_pack_task, execute_task_runner):
+def test_happy_path(serde, mock_log, task_uuid, ez_pack_task, execute_task_runner):
     out = random.randint(1, 100_000)
     divisor = random.randint(1, 100_000)
 
     task_bytes = ez_pack_task(divide, divisor * out, divisor)
 
-    with caplog.at_level(logging.DEBUG):
-        packed_result = execute_task_runner(task_bytes)
+    packed_result = execute_task_runner(task_bytes)
     assert isinstance(packed_result, bytes)
 
     result = messagepack.unpack(packed_result)
@@ -89,7 +89,12 @@ def test_happy_path(serde, caplog, task_uuid, ez_pack_task, execute_task_runner)
     assert "endpoint_id" in result.details
     assert serde.deserialize(result.data) == out
 
-    log_msgs = "\n".join(r.msg % r.args for r in caplog.records)
+    log_recs = []
+    for a, _ in chain(mock_log.info.call_args_list, mock_log.debug.call_args_list):
+        fmt, *a = a
+        log_recs.append(a and fmt % tuple(a) or fmt)
+    log_msgs = "\n".join(log_recs)
+
     assert "Preparing to execute" in log_msgs, "Expect log clue of progress"
     assert "Unpacking" in log_msgs, "Expect log clue of progress"
     assert "Deserializing function" in log_msgs, "Expect log to clue of progress"
@@ -98,8 +103,8 @@ def test_happy_path(serde, caplog, task_uuid, ez_pack_task, execute_task_runner)
     assert "Task function complete" in log_msgs, "Expect log clue of progress"
     assert "Execution completed" in log_msgs, "Expect log clue of progress"
     assert "Task processing completed in" in log_msgs, "Expect log clue of progress"
-    assert len(caplog.records) == 7, "Time to update test?"
-    assert log_msgs.count(str(task_uuid)) == len(caplog.records), "Expect id prefixed"
+    assert len(log_recs) == 7, "Time to update test?"
+    assert log_msgs.count(str(task_uuid)) == len(log_recs), "Expect id always prefixed"
 
 
 def test_sandbox(task_10_2, execute_task_runner, task_uuid, tmp_path):

compute_endpoint/tox.ini

@@ -10,7 +10,7 @@ extras = test
 usedevelop = true
 commands =
     coverage erase
-    coverage run -m pytest --durations 5 --log-cli-level=ERROR {posargs}
+    coverage run -m pytest --random-order --durations 5 --log-cli-level=ERROR {posargs}
     coverage report
 
 [testenv:mypy]

compute_sdk/globus_compute_sdk/sdk/executor.py

@@ -1447,10 +1447,12 @@ def run(self):
                         fut.set_exception(self._cancellation_reason)
                         log.debug("Cancelled: %s", fut.task_id)
                     self._open_futures_empty.set()
+        self.shutdown()
         log.debug("%r AMQP thread complete.", self)
 
     def shutdown(self, wait=True, *, cancel_futures=False):
         if not self.is_alive():
+            _RESULT_WATCHERS.pop(self.task_group_id, None)
             return
 
         self._closed = True  # No more futures will be accepted

compute_sdk/setup.py

@@ -36,6 +36,7 @@
     "flake8==3.8.0",
     "pytest>=7.2",
     "pytest-mock",
+    "pytest-random-order",
     "pyfakefs",
     "coverage",
     # easy mocking of the `requests` library

compute_sdk/tests/unit/test_executor.py

@@ -69,7 +69,6 @@ def __init__(self, *args, **kwargs):
         }
         kwargs.setdefault("client", mock_client)
         super().__init__(*args, **kwargs)
-        Executor._default_task_group_id = None  # Reset for each test
         self._test_task_submitter_exception: t.Type[Exception] | None = None
         self._test_task_submitter_done = False
 
@@ -112,6 +111,11 @@ def join(self, timeout: float | None = None) -> None:
         super().join(timeout=timeout)
 
 
+@pytest.fixture(autouse=True)
+def reset_default_executor_id():
+    Executor._default_task_group_id = None  # Reset for each test
+
+
 @pytest.fixture
 def mock_result_watcher(mocker: MockerFixture):
     rw = mocker.patch(f"{_MOCK_BASE}_ResultWatcher", autospec=True)
@@ -1581,6 +1585,7 @@ def test_resultwatcher_onmessage_verifies_result_type(mocker, unpacked):
     mrw._on_message(mock_channel, mock_deliver, mock_props, b"some_bytes")
     mock_channel.basic_nack.assert_called()
     assert not mrw._received_results
+    mrw.shutdown()
 
 
 def test_resultwatcher_onmessage_sets_check_results_flag():
@@ -1594,6 +1599,7 @@ def test_resultwatcher_onmessage_sets_check_results_flag():
     mock_channel.basic_nack.assert_not_called()
     assert mrw._received_results
     assert mrw._time_to_check_results.is_set()
+    mrw.shutdown()
 
 
 @pytest.mark.parametrize("exc", (MemoryError("some description"), "some description"))

compute_sdk/tox.ini

@@ -9,7 +9,7 @@ usedevelop = true
 extras = test
 commands =
     coverage erase
-    coverage run -m pytest --durations 5 {posargs}
+    coverage run -m pytest --durations 5 --random-order {posargs}
     coverage report --skip-covered
 
 [testenv:mypy]