fix(profiling): let the task recap reach past the twenty slowest

ansible.cfg enables profile_tasks, whose output_limit defaults to 20. On a deploy round of this size the recap therefore cut off at 66.07s and printed nothing below, so a role whose cost is spread over many cheap tasks could not be attributed at all: in run 33828309745 that left 22.7 of sys-svc-compose 51.9 minutes unexplained, and the same blind spot covers web-app-nextcloud and web-app-keycloak almost entirely.

The callback accepts all, which disables the cap, but the recap prints one line per task execution and pass 1 alone ran over 7500 of them. Since the Actions worker dies silently on ENOSPC while writing its own logs, which is why utils/tests/swarm/matrix.py aborts a step at DISK_FLOOR_MB, the limit stays bounded at 300: about 600 extra lines across both passes.

Measured with a 30-task play: 30 recap entries with this config, 20 with the upstream default forced through PROFILE_TASKS_TASK_OUTPUT_LIMIT. The lint pins the setting so removing the section cannot silently restore the truncation.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

ansible.cfg

@@ -25,6 +25,9 @@ lookup_plugins  = ./plugins/lookup
 library         = ./library
 module_utils    = ./utils
 
+[callback_profile_tasks]
+task_output_limit = 300
+
 [ssh_connection]
 ssh_args = -o ControlMaster=auto -o ControlPersist=20s -o ControlPath=~/.ssh/ansible-%h-%p-%r -o ServerAliveInterval=15 -o ServerAliveCountMax=3 -o StrictHostKeyChecking=accept-new -o PreferredAuthentications=publickey,password,keyboard-interactive
 pipelining = True

tests/lint/filesystem/test_profile_tasks_output_limit.py

@@ -0,0 +1,51 @@
+"""Lint check: the task profiler must not truncate its summary to the default.
+
+``ansible.cfg`` enables ``profile_tasks``, whose ``output_limit`` defaults to
+20. A deploy round of this size then reports only the twenty slowest tasks and
+prints nothing below them, so any role whose cost is spread over many cheap
+tasks cannot be attributed at all: in run 33828309745 the recap cut off at
+66.07s, leaving 22.7 of ``sys-svc-compose``'s 51.9 minutes unaccounted.
+
+The limit stays bounded rather than ``all`` because the recap prints one line
+per task execution and the Actions worker dies silently on ENOSPC while writing
+its own logs, which is the same reason ``utils/tests/swarm/matrix.py`` aborts a
+step at ``DISK_FLOOR_MB``.
+"""
+
+from __future__ import annotations
+
+import configparser
+import unittest
+
+from utils import PROJECT_ROOT
+
+CONFIG = PROJECT_ROOT / "ansible.cfg"
+SECTION = "callback_profile_tasks"
+KEY = "task_output_limit"
+DEFAULT_LIMIT = 20
+
+
+class TestProfileTasksOutputLimit(unittest.TestCase):
+    def setUp(self) -> None:
+        self.parser = configparser.ConfigParser()
+        self.parser.read(CONFIG)
+
+    def test_the_profiler_is_enabled(self) -> None:
+        enabled = self.parser.get("defaults", "callbacks_enabled", fallback="")
+        self.assertIn("profile_tasks", enabled)
+
+    def test_the_summary_is_not_left_at_the_default_limit(self) -> None:
+        self.assertTrue(
+            self.parser.has_option(SECTION, KEY),
+            f"ansible.cfg must set [{SECTION}] {KEY}; without it the recap "
+            f"stops after {DEFAULT_LIMIT} tasks and the cost of a role spread "
+            "over many cheap tasks cannot be attributed",
+        )
+        value = self.parser.get(SECTION, KEY).strip()
+        if value == "all":
+            return
+        self.assertGreater(int(value), DEFAULT_LIMIT)
+
+
+if __name__ == "__main__":
+    unittest.main()