tests: Avoid intermittant 2 hour timeout in new style Ansible module tests

This has been lurking for years, raising it's head at unpredictable times.
This change doesn't fix it, but it should make it a lot less mysterious.

tests/ansible/integration/runner/custom_python_new_style_missing_interpreter.yml

@@ -2,6 +2,11 @@
 - name: integration/runner/custom_python_new_style_module.yml
   hosts: test-targets
   tasks:
+    # FIXME Without Mitogen Ansible often reads stdin before the module.
+    #       Either don't read directly from stdin, or figure out the cause.
+    - meta: end_play
+      when: not is_mitogen
+
     - custom_python_new_style_missing_interpreter:
         foo: true
       with_sequence: start=0 end={{end|default(1)}}

tests/ansible/integration/runner/custom_python_new_style_module.yml

@@ -1,7 +1,8 @@
 - name: integration/runner/custom_python_new_style_module.yml
   hosts: test-targets
   tasks:
-    # without Mitogen Ansible 2.10 hangs on this play
+    # FIXME Without Mitogen Ansible often reads stdin before the module.
+    #       Either don't read directly from stdin, or figure out the cause.
     - meta: end_play
       when: not is_mitogen
 

tests/ansible/lib/modules/custom_python_new_style_missing_interpreter.py

@@ -1,6 +1,20 @@
 # I am an Ansible new-style Python module, but I lack an interpreter.
+# See also custom_python_new_style_module, we should be updated in tandem.
 
+import io
+import json
+import select
+import signal
 import sys
+import warnings
+
+# Ansible 2.7 changed how new style modules are invoked. It seems that module
+# parameters are *sometimes* read before the module runs. Modules that try
+# to read directly from stdin, such as this, are unable to. However it doesn't
+# always fail, influences seem to include Ansible & Python version. As noted
+# in ansible.module_utils.basic._load_params() we should probably use that.
+# I think (medium confidence) I narrowed the inflection (with git bisect) to
+# https://github.com/ansible/ansible/commit/52449cc01a71778ef94ea0237eed0284f5d75582
 
 # As of Ansible 2.10, Ansible changed new-style detection: # https://github.com/ansible/ansible/pull/61196/files#diff-5675e463b6ce1fbe274e5e7453f83cd71e61091ea211513c93e7c0b4d527d637L828-R980
 # NOTE: this import works for Mitogen, and the import below matches new-style Ansible 2.10
@@ -8,11 +22,46 @@ import sys
 # from ansible.module_utils.
 # import ansible.module_utils.
 
+# These timeouts should prevent hard-to-attribute, 2+ hour CI job timeouts.
+# Previously this module has waited on stdin forever (timeoutInMinutes=120).
+SELECT_TIMEOUT = 5.0    # seconds
+SIGNAL_TIMEOUT = 10     # seconds
+
+
+def fail_json(msg, **kwargs):
+    kwargs.update(failed=True, msg=msg)
+    print(json.dumps(kwargs, sys.stdout, indent=2, sort_keys=True))
+    sys.exit(1)
+
+
+def sigalrm_handler(signum, frame):
+    fail_json("Still executing after SIGNAL_TIMEOUT=%ds" % (SIGNAL_TIMEOUT,))
+
 
 def usage():
     sys.stderr.write('Usage: %s <input.json>\n' % (sys.argv[0],))
     sys.exit(1)
 
+
+# Wait SIGNAL_TIMEOUT seconds, exit with failure if still running.
+signal.signal(signal.SIGALRM, sigalrm_handler)
+signal.alarm(SIGNAL_TIMEOUT)
+
+# Wait SELECT_TIMEOUT seconds, exit with failure if no data appears on stdin.
+# TODO Combine select() & read() in a loop, to handle slow trickle of data.
+#      Consider buffering, line buffering, `f.read()` vs `f.read1()`.
+# TODO Document that sys.stdin may be a StringIO under Ansible + Mitogen.
+try:
+    inputs_ready, _, _ = select.select([sys.stdin], [], [], SELECT_TIMEOUT)
+except (AttributeError, TypeError, io.UnsupportedOperation) as exc:
+    # sys.stdin.fileno() doesn't exist or can't return a real file descriptor.
+    warnings.warn("Could not wait on sys.stdin=%r: %r" % (sys.stdin, exc))
+else:
+    if not inputs_ready:
+        fail_json("Gave up waiting on sys.stdin after SELECT_TIMEOUT=%ds"
+                  % (SELECT_TIMEOUT,))
+
+# Read all data on stdin. May block forever, if EOF is not reached.
 input_json = sys.stdin.read()
 
 print("{")

tests/ansible/lib/modules/custom_python_new_style_module.py

@@ -1,16 +1,65 @@
 #!/usr/bin/python
 # I am an Ansible new-style Python module. I should receive an encoding string.
+# See also custom_python_new_style_module, we should be updated in tandem.
 
+import io
+import json
+import select
+import signal
 import sys
+import warnings
+
+# Ansible 2.7 changed how new style modules are invoked. It seems that module
+# parameters are *sometimes* read before the module runs. Modules that try
+# to read directly from stdin, such as this, are unable to. However it doesn't
+# always fail, influences seem to include Ansible & Python version. As noted
+# in ansible.module_utils.basic._load_params() we should probably use that.
+# I think (medium confidence) I narrowed the inflection (with git bisect) to
+# https://github.com/ansible/ansible/commit/52449cc01a71778ef94ea0237eed0284f5d75582
 
 # This is the magic marker Ansible looks for:
 # from ansible.module_utils.
 
+# These timeouts should prevent hard-to-attribute, 2+ hour CI job timeouts.
+# Previously this module has waited on stdin forever (timeoutInMinutes=120).
+SELECT_TIMEOUT = 5.0    # seconds
+SIGNAL_TIMEOUT = 10     # seconds
+
+
+def fail_json(msg, **kwargs):
+    kwargs.update(failed=True, msg=msg)
+    print(json.dumps(kwargs, sys.stdout, indent=2, sort_keys=True))
+    sys.exit(1)
+
+
+def sigalrm_handler(signum, frame):
+    fail_json("Still executing after SIGNAL_TIMEOUT=%ds" % (SIGNAL_TIMEOUT,))
+
 
 def usage():
     sys.stderr.write('Usage: %s <input.json>\n' % (sys.argv[0],))
     sys.exit(1)
 
+
+# Wait SIGNAL_TIMEOUT seconds, exit with failure if still running.
+signal.signal(signal.SIGALRM, sigalrm_handler)
+signal.alarm(SIGNAL_TIMEOUT)
+
+# Wait SELECT_TIMEOUT seconds, exit with failure if no data appears on stdin.
+# TODO Combine select() & read() in a loop, to handle slow trickle of data.
+#      Consider buffering, line buffering, `f.read()` vs `f.read1()`.
+# TODO Document that sys.stdin may be a StringIO under Ansible + Mitogen.
+try:
+    inputs_ready, _, _ = select.select([sys.stdin], [], [], SELECT_TIMEOUT)
+except (AttributeError, TypeError, io.UnsupportedOperation) as exc:
+    # sys.stdin.fileno() doesn't exist or can't return a real file descriptor.
+    warnings.warn("Could not wait on sys.stdin=%r: %r" % (sys.stdin, exc))
+else:
+    if not inputs_ready:
+        fail_json("Gave up waiting on sys.stdin after SELECT_TIMEOUT=%ds"
+                  % (SELECT_TIMEOUT,))
+
+# Read all data on stdin. May block forever, if EOF is not reached.
 input_json = sys.stdin.read()
 
 print("{")