Emit periodic keepalive events from Worker (#1191) (#1201)

* new CLI arg and envvar for `Worker` mode to optionally emit regular keepalive events; fixes issues with container runtimes that assume long-silent stdout == hung process

---------

Co-authored-by: Alan Rominger <arominge@redhat.com>
(cherry picked from commit fd9d67a)

ansible_runner/__main__.py

@@ -611,6 +611,15 @@ def main(sys_args=None):
             "Using this will also assure that the directory is deleted when the job finishes."
         )
     )
+    worker_subparser.add_argument(
+        "--keepalive-seconds",
+        dest="keepalive_seconds",
+        default=None,
+        type=int,
+        help=(
+            "Emit a synthetic keepalive event every N seconds of idle. (default=0, disabled)"
+        )
+    )
     process_subparser = subparser.add_parser(
         'process',
         help="Receive the output of remote ansible-runner work and distribute the results"
@@ -859,6 +868,7 @@ def main(sys_args=None):
                                    limit=vargs.get('limit'),
                                    streamer=streamer,
                                    suppress_env_files=vargs.get("suppress_env_files"),
+                                   keepalive_seconds=vargs.get("keepalive_seconds"),
                                    )
                 try:
                     res = run(**run_options)
@@ -887,3 +897,7 @@ def main(sys_args=None):
             return 0
         except OSError:
             return 1
+
+
+if __name__ == '__main__':
+    sys.exit(main())

ansible_runner/config/_base.py

@@ -67,7 +67,7 @@ def __init__(self,
                  process_isolation=False, process_isolation_executable=None,
                  container_image=None, container_volume_mounts=None, container_options=None, container_workdir=None, container_auth_data=None,
                  ident=None, rotate_artifacts=0, timeout=None, ssh_key=None, quiet=False, json_mode=False,
-                 check_job_event_data=False, suppress_env_files=False):
+                 check_job_event_data=False, suppress_env_files=False, keepalive_seconds=None):
         # common params
         self.host_cwd = host_cwd
         self.envvars = envvars
@@ -95,6 +95,8 @@ def __init__(self,
         self.timeout = timeout
         self.check_job_event_data = check_job_event_data
         self.suppress_env_files = suppress_env_files
+        # ignore this for now since it's worker-specific and would just trip up old runners
+        # self.keepalive_seconds = keepalive_seconds
 
         # setup initial environment
         if private_data_dir:

ansible_runner/streaming.py

@@ -1,3 +1,5 @@
+from __future__ import annotations  # allow newer type syntax until 3.10 is our minimum
+
 import codecs
 import json
 import os
@@ -6,17 +8,16 @@
 import tempfile
 import uuid
 import traceback
-try:
-    from collections.abc import Mapping
-except ImportError:
-    from collections import Mapping
 
 import ansible_runner
 from ansible_runner.exceptions import ConfigurationError
 from ansible_runner.loader import ArtifactLoader
 import ansible_runner.plugins
 from ansible_runner.utils import register_for_cleanup
 from ansible_runner.utils.streaming import stream_dir, unstream_dir
+from collections.abc import Mapping
+from functools import wraps
+from threading import Event, RLock, Thread
 
 
 class UUIDEncoder(json.JSONEncoder):
@@ -38,6 +39,9 @@ def __init__(self, _output=None, **kwargs):
         self._output = _output
         self.private_data_dir = os.path.abspath(kwargs.pop('private_data_dir'))
         self.only_transmit_kwargs = kwargs.pop('only_transmit_kwargs', False)
+        if 'keepalive_seconds' in kwargs:
+            kwargs.pop('keepalive_seconds')  # don't confuse older runners with this Worker-only arg
+
         self.kwargs = kwargs
 
         self.status = "unstarted"
@@ -60,12 +64,22 @@ def run(self):
         return self.status, self.rc
 
 
-class Worker(object):
-    def __init__(self, _input=None, _output=None, **kwargs):
+class Worker:
+    def __init__(self, _input=None, _output=None, keepalive_seconds: float | None = None, **kwargs):
         if _input is None:
             _input = sys.stdin.buffer
         if _output is None:
             _output = sys.stdout.buffer
+
+        if keepalive_seconds is None:  # if we didn't get an explicit int value, fall back to envvar
+            # FIXME: emit/log a warning and silently continue if this value won't parse
+            keepalive_seconds = float(os.environ.get('ANSIBLE_RUNNER_KEEPALIVE_SECONDS', 0))
+
+        self._keepalive_interval_sec = keepalive_seconds
+        self._keepalive_thread: Thread | None = None
+        self._output_event = Event()
+        self._output_lock = RLock()
+
         self._input = _input
         self._output = _output
 
@@ -81,6 +95,64 @@ def __init__(self, _input=None, _output=None, **kwargs):
         self.status = "unstarted"
         self.rc = None
 
+    def _begin_keepalive(self):
+        """Starts a keepalive thread at most once"""
+        if not self._keepalive_thread:
+            self._keepalive_thread = Thread(target=self._keepalive_loop, daemon=True)
+            self._keepalive_thread.start()
+
+    def _end_keepalive(self):
+        """Disable the keepalive interval and notify the keepalive thread to shut down"""
+        self._keepalive_interval_sec = 0
+        self._output_event.set()
+
+    def _keepalive_loop(self):
+        """Main loop for keepalive injection thread; exits when keepalive interval is <= 0"""
+        while self._keepalive_interval_sec > 0:
+            # block until output has occurred or keepalive interval elapses
+            if self._output_event.wait(timeout=self._keepalive_interval_sec):
+                # output was sent before keepalive timeout; reset the event and start waiting again
+                self._output_event.clear()
+                continue
+
+            # keepalive interval elapsed; try to send a keepalive...
+            # pre-acquire the output lock without blocking
+            if not self._output_lock.acquire(blocking=False):
+                # something else has the lock; output is imminent, so just skip this keepalive
+                # NB: a long-running operation under an event handler that's holding this lock but not actually moving
+                # output could theoretically block keepalives long enough to cause problems, but it's probably not
+                # worth the added locking hassle to be pedantic about it
+                continue
+
+            try:
+                # were keepalives recently disabled?
+                if self._keepalive_interval_sec <= 0:
+                    # we're probably shutting down; don't risk corrupting output by writing now, just bail out
+                    return
+                # output a keepalive event
+                # FIXME: this could be a lot smaller (even just `{}`) if a short-circuit discard was guaranteed in
+                #  Processor or if other layers were more defensive about missing event keys and/or unknown dictionary
+                #  values...
+                self.event_handler(dict(event='keepalive', counter=0, uuid=0))
+            finally:
+                # always release the output lock (
+                self._output_lock.release()
+
+    def _synchronize_output_reset_keepalive(wrapped_method):
+        """
+        Utility decorator to synchronize event writes and flushes to avoid keepalives splatting in the middle of
+        mid-write events, and reset keepalive interval on write completion.
+        """
+        @wraps(wrapped_method)
+        def wrapper(self, *args, **kwargs):
+            with self._output_lock:
+                ret = wrapped_method(self, *args, **kwargs)
+                # signal the keepalive thread last, so the timeout restarts after the last write, not before the first
+                self._output_event.set()
+                return ret
+
+        return wrapper
+
     def update_paths(self, kwargs):
         if kwargs.get('envvars'):
             if 'ANSIBLE_ROLES_PATH' in kwargs['envvars']:
@@ -93,63 +165,72 @@ def update_paths(self, kwargs):
         return kwargs
 
     def run(self):
-        while True:
-            try:
-                line = self._input.readline()
-                data = json.loads(line)
-            except (json.decoder.JSONDecodeError, IOError):
-                self.status_handler({'status': 'error', 'job_explanation': 'Failed to JSON parse a line from transmit stream.'}, None)
-                self.finished_callback(None)  # send eof line
-                return self.status, self.rc
-
-            if 'kwargs' in data:
-                self.job_kwargs = self.update_paths(data['kwargs'])
-            elif 'zipfile' in data:
+        self._begin_keepalive()
+        try:
+            while True:
                 try:
-                    unstream_dir(self._input, data['zipfile'], self.private_data_dir)
-                except Exception:
-                    self.status_handler({
-                        'status': 'error',
-                        'job_explanation': 'Failed to extract private data directory on worker.',
-                        'result_traceback': traceback.format_exc()
-                    }, None)
+                    line = self._input.readline()
+                    data = json.loads(line)
+                except (json.decoder.JSONDecodeError, IOError):
+                    self.status_handler({'status': 'error', 'job_explanation': 'Failed to JSON parse a line from transmit stream.'}, None)
                     self.finished_callback(None)  # send eof line
                     return self.status, self.rc
-            elif 'eof' in data:
-                break
 
-        self.kwargs.update(self.job_kwargs)
-        self.kwargs['quiet'] = True
-        self.kwargs['suppress_ansible_output'] = True
-        self.kwargs['private_data_dir'] = self.private_data_dir
-        self.kwargs['status_handler'] = self.status_handler
-        self.kwargs['event_handler'] = self.event_handler
-        self.kwargs['artifacts_handler'] = self.artifacts_handler
-        self.kwargs['finished_callback'] = self.finished_callback
-
-        r = ansible_runner.interface.run(**self.kwargs)
-        self.status, self.rc = r.status, r.rc
-
-        # FIXME: do cleanup on the tempdir
+                if 'kwargs' in data:
+                    self.job_kwargs = self.update_paths(data['kwargs'])
+                elif 'zipfile' in data:
+                    try:
+                        unstream_dir(self._input, data['zipfile'], self.private_data_dir)
+                    except Exception:
+                        self.status_handler({
+                            'status': 'error',
+                            'job_explanation': 'Failed to extract private data directory on worker.',
+                            'result_traceback': traceback.format_exc()
+                        }, None)
+                        self.finished_callback(None)  # send eof line
+                        return self.status, self.rc
+                elif 'eof' in data:
+                    break
+
+            self.kwargs.update(self.job_kwargs)
+            self.kwargs['quiet'] = True
+            self.kwargs['suppress_ansible_output'] = True
+            self.kwargs['private_data_dir'] = self.private_data_dir
+            self.kwargs['status_handler'] = self.status_handler
+            self.kwargs['event_handler'] = self.event_handler
+            self.kwargs['artifacts_handler'] = self.artifacts_handler
+            self.kwargs['finished_callback'] = self.finished_callback
+
+            r = ansible_runner.interface.run(**self.kwargs)
+            self.status, self.rc = r.status, r.rc
+
+            # FIXME: do cleanup on the tempdir
+        finally:
+            self._end_keepalive()
 
         return self.status, self.rc
 
+    @_synchronize_output_reset_keepalive
     def status_handler(self, status_data, runner_config):
         self.status = status_data['status']
         self._output.write(json.dumps(status_data).encode('utf-8'))
         self._output.write(b'\n')
         self._output.flush()
 
+    @_synchronize_output_reset_keepalive
     def event_handler(self, event_data):
         self._output.write(json.dumps(event_data).encode('utf-8'))
         self._output.write(b'\n')
         self._output.flush()
 
+    @_synchronize_output_reset_keepalive
     def artifacts_handler(self, artifact_dir):
         stream_dir(artifact_dir, self._output)
         self._output.flush()
 
+    @_synchronize_output_reset_keepalive
     def finished_callback(self, runner_obj):
+        self._end_keepalive()  # ensure that we can't splat a keepalive event after the eof event
         self._output.write(json.dumps({'eof': True}).encode('utf-8'))
         self._output.write(b'\n')
         self._output.flush()
@@ -210,10 +291,18 @@ def status_callback(self, status_data):
             self.status_handler(status_data, runner_config=self.config)
 
     def event_callback(self, event_data):
+        # FIXME: this needs to be more defensive to not blow up on "malformed" events or new values it doesn't recognize
+        counter = event_data.get('counter')
+        uuid = event_data.get('uuid')
+
+        if not counter or not uuid:
+            # FIXME: log a warning about a malformed event?
+            return
+
         full_filename = os.path.join(self.artifact_dir,
                                      'job_events',
-                                     '{}-{}.json'.format(event_data['counter'],
-                                                         event_data['uuid']))
+                                     f'{counter}-{uuid}.json')
+
         if not self.quiet and 'stdout' in event_data:
             print(event_data['stdout'])
 
@@ -254,6 +343,9 @@ def run(self):
                 self.artifacts_callback(data)
             elif 'eof' in data:
                 break
+            elif data.get('event') == 'keepalive':
+                # just ignore keepalives
+                continue
             else:
                 self.event_callback(data)