parsl 2024.4.15__py3-none-any.whl → 2024.4.29__py3-none-any.whl

parsl/addresses.py

@@ -113,7 +113,7 @@ def get_all_addresses() -> Set[str]:
         try:
             s_addresses.add(address_by_interface(interface))
         except Exception:
-            logger.info("Ignoring failure to fetch address from interface {}".format(interface))
+            logger.debug("Ignoring failure to fetch address from interface {}".format(interface))
 
     resolution_functions: List[Callable[[], str]]
     resolution_functions = [address_by_hostname, address_by_route, address_by_query]
@@ -121,7 +121,7 @@ def get_all_addresses() -> Set[str]:
         try:
             s_addresses.add(f())
         except Exception:
-            logger.info("Ignoring an address finder exception")
+            logger.debug("Ignoring an address finder exception")
 
     return s_addresses
 

parsl/app/bash.py

@@ -5,6 +5,7 @@ import logging
 
 from parsl.app.errors import wrap_error
 from parsl.app.app import AppBase
+from parsl.data_provider.files import File
 from parsl.dataflow.dflow import DataFlowKernelLoader
 
 logger = logging.getLogger(__name__)
@@ -54,13 +55,20 @@ def remote_side_bash_executor(func, *args, **kwargs):
         if stdfspec is None:
             return None
 
-        fname, mode = get_std_fname_mode(fdname, stdfspec)
+        if isinstance(stdfspec, File):
+            # a File is an os.PathLike and so we can use it directly for
+            # the subsequent file operations
+            fname = stdfspec
+            mode = "w"
+        else:
+            fname, mode = get_std_fname_mode(fdname, stdfspec)
+
         try:
             if os.path.dirname(fname):
                 os.makedirs(os.path.dirname(fname), exist_ok=True)
             fd = open(fname, mode)
         except Exception as e:
-            raise pe.BadStdStreamFile(fname, e)
+            raise pe.BadStdStreamFile(str(fname)) from e
         return fd
 
     std_out = open_std_fd('stdout')

parsl/app/errors.py

@@ -78,16 +78,14 @@ class BadStdStreamFile(ParslError):
 
     Contains:
        reason(string)
-       exception object
     """
 
-    def __init__(self, reason: str, exception: Exception) -> None:
-        super().__init__(reason, exception)
+    def __init__(self, reason: str) -> None:
+        super().__init__(reason)
         self._reason = reason
-        self._exception = exception
 
     def __repr__(self) -> str:
-        return "Bad Stream File: {} Exception: {}".format(self._reason, self._exception)
+        return "Bad Stream File: {}".format(self._reason)
 
     def __str__(self) -> str:
         return self.__repr__()

parsl/config.py

@@ -10,11 +10,12 @@ from parsl.executors.threads import ThreadPoolExecutor
 from parsl.errors import ConfigurationError
 from parsl.dataflow.taskrecord import TaskRecord
 from parsl.monitoring import MonitoringHub
+from parsl.usage_tracking.api import UsageInformation
 
 logger = logging.getLogger(__name__)
 
 
-class Config(RepresentationMixin):
+class Config(RepresentationMixin, UsageInformation):
     """
     Specification of Parsl configuration options.
 
@@ -50,6 +51,9 @@ class Config(RepresentationMixin):
         of 1.
     run_dir : str, optional
         Path to run directory. Default is 'runinfo'.
+    std_autopath : function, optional
+        Sets the function used to generate stdout/stderr specifications when parsl.AUTO_LOGPATH is used. If no function
+        is specified, generates paths that look like: ``rundir/NNN/task_logs/X/task_{id}_{name}{label}.{out/err}``
     strategy : str, optional
         Strategy to use for scaling blocks according to workflow needs. Can be 'simple', 'htex_auto_scale', 'none'
         or `None`.
@@ -89,6 +93,7 @@ class Config(RepresentationMixin):
                  retries: int = 0,
                  retry_handler: Optional[Callable[[Exception, TaskRecord], float]] = None,
                  run_dir: str = 'runinfo',
+                 std_autopath: Optional[Callable] = None,
                  strategy: Optional[str] = 'simple',
                  strategy_period: Union[float, int] = 5,
                  max_idletime: float = 120.0,
@@ -129,6 +134,7 @@ class Config(RepresentationMixin):
         self.usage_tracking = usage_tracking
         self.initialize_logging = initialize_logging
         self.monitoring = monitoring
+        self.std_autopath: Optional[Callable] = std_autopath
 
     @property
     def executors(self) -> Sequence[ParslExecutor]:
@@ -144,3 +150,6 @@ class Config(RepresentationMixin):
         if len(duplicates) > 0:
             raise ConfigurationError('Executors must have unique labels ({})'.format(
                 ', '.join(['label={}'.format(repr(d)) for d in duplicates])))
+
+    def get_usage_information(self):
+        return {"executors_len": len(self.executors)}

parsl/data_provider/zip.py

@@ -42,6 +42,12 @@ class ZipFileStaging(Staging):
     """
 
     def can_stage_out(self, file: File) -> bool:
+        return self.is_zip_url(file)
+
+    def can_stage_in(self, file: File) -> bool:
+        return self.is_zip_url(file)
+
+    def is_zip_url(self, file: File) -> bool:
         logger.debug("archive provider checking File {}".format(repr(file)))
 
         # First check if this is the scheme we care about
@@ -76,6 +82,20 @@ class ZipFileStaging(Staging):
         app_fut = stage_out_app(zip_path, inside_path, working_dir, inputs=[file], _parsl_staging_inhibit=True, parent_fut=parent_fut)
         return app_fut
 
+    def stage_in(self, dm, executor, file, parent_fut):
+        assert file.scheme == 'zip'
+
+        zip_path, inside_path = zip_path_split(file.path)
+
+        working_dir = dm.dfk.executors[executor].working_dir
+
+        if working_dir:
+            file.local_path = os.path.join(working_dir, inside_path)
+
+        stage_in_app = _zip_stage_in_app(dm)
+        app_fut = stage_in_app(zip_path, inside_path, working_dir, outputs=[file], _parsl_staging_inhibit=True, parent_fut=parent_fut)
+        return app_fut._outputs[0]
+
 
 def _zip_stage_out(zip_file, inside_path, working_dir, parent_fut=None, inputs=[], _parsl_staging_inhibit=True):
     file = inputs[0]
@@ -93,6 +113,18 @@ def _zip_stage_out_app(dm):
     return parsl.python_app(executors=['_parsl_internal'], data_flow_kernel=dm.dfk)(_zip_stage_out)
 
 
+def _zip_stage_in(zip_file, inside_path, working_dir, *, parent_fut, outputs, _parsl_staging_inhibit=True):
+    with filelock.FileLock(zip_file + ".lock"):
+        with zipfile.ZipFile(zip_file, mode='r') as z:
+            content = z.read(inside_path)
+        with open(outputs[0], "wb") as of:
+            of.write(content)
+
+
+def _zip_stage_in_app(dm):
+    return parsl.python_app(executors=['_parsl_internal'], data_flow_kernel=dm.dfk)(_zip_stage_in)
+
+
 def zip_path_split(path: str) -> Tuple[str, str]:
     """Split zip: path into a zipfile name and a contained-file name.
     """

parsl/dataflow/dflow.py

@@ -219,14 +219,18 @@ class DataFlowKernel:
             task_log_info = self._create_task_log_info(task_record)
             self.monitoring.send(MessageType.TASK_INFO, task_log_info)
 
-    def _create_task_log_info(self, task_record):
+    def _create_task_log_info(self, task_record: TaskRecord) -> Dict[str, Any]:
         """
         Create the dictionary that will be included in the log.
         """
         info_to_monitor = ['func_name', 'memoize', 'hashsum', 'fail_count', 'fail_cost', 'status',
                            'id', 'time_invoked', 'try_time_launched', 'time_returned', 'try_time_returned', 'executor']
 
-        task_log_info = {"task_" + k: task_record[k] for k in info_to_monitor}
+        # mypy cannot verify that these task_record[k] references are valid:
+        # They are valid if all entries in info_to_monitor are declared in the definition of TaskRecord
+        # This type: ignore[literal-required] asserts that fact.
+        task_log_info = {"task_" + k: task_record[k] for k in info_to_monitor}  # type: ignore[literal-required]
+
         task_log_info['run_id'] = self.run_id
         task_log_info['try_id'] = task_record['try_id']
         task_log_info['timestamp'] = datetime.datetime.now()
@@ -238,33 +242,28 @@ class DataFlowKernel:
         task_log_info['task_inputs'] = str(task_record['kwargs'].get('inputs', None))
         task_log_info['task_outputs'] = str(task_record['kwargs'].get('outputs', None))
         task_log_info['task_stdin'] = task_record['kwargs'].get('stdin', None)
-        stdout_spec = task_record['kwargs'].get('stdout', None)
-        stderr_spec = task_record['kwargs'].get('stderr', None)
 
-        # stdout and stderr strings are set to the filename if we can
-        # interpret the specification; otherwise, set to the empty string
-        # (on exception, or when not specified)
+        def std_spec_to_name(name, spec):
+            if spec is None:
+                name = ""
+            elif isinstance(spec, File):
+                name = spec.url
+            else:
+                # fallthrough case is various str, os.PathLike, tuple modes that
+                # can be interpreted by get_std_fname_mode.
+                try:
+                    name, _ = get_std_fname_mode(name, spec)
+                except Exception:
+                    logger.exception(f"Could not parse {name} specification {spec} for task {task_record['id']}")
+                    name = ""
+            return name
 
-        if stdout_spec is not None:
-            try:
-                stdout_name, _ = get_std_fname_mode('stdout', stdout_spec)
-            except Exception:
-                logger.exception("Could not parse stdout specification {} for task {}".format(stdout_spec, task_record['id']))
-                stdout_name = ""
-        else:
-            stdout_name = ""
+        stdout_spec = task_record['kwargs'].get('stdout')
+        task_log_info['task_stdout'] = std_spec_to_name('stdout', stdout_spec)
 
-        if stderr_spec is not None:
-            try:
-                stderr_name, _ = get_std_fname_mode('stderr', stderr_spec)
-            except Exception:
-                logger.exception("Could not parse stderr specification {} for task {}".format(stderr_spec, task_record['id']))
-                stderr_name = ""
-        else:
-            stderr_name = ""
+        stderr_spec = task_record['kwargs'].get('stderr')
+        task_log_info['task_stderr'] = std_spec_to_name('stderr', stderr_spec)
 
-        task_log_info['task_stdout'] = stdout_name
-        task_log_info['task_stderr'] = stderr_name
         task_log_info['task_fail_history'] = ",".join(task_record['fail_history'])
         task_log_info['task_depends'] = None
         if task_record['depends'] is not None:
@@ -774,6 +773,10 @@ class DataFlowKernel:
             (inputs[idx], func) = self.data_manager.optionally_stage_in(f, func, executor)
 
         for kwarg, f in kwargs.items():
+            # stdout and stderr files should not be staging in (they will be staged *out*
+            # in _add_output_deps)
+            if kwarg in ['stdout', 'stderr']:
+                continue
             (kwargs[kwarg], func) = self.data_manager.optionally_stage_in(f, func, executor)
 
         newargs = list(args)
@@ -786,33 +789,55 @@ class DataFlowKernel:
         logger.debug("Adding output dependencies")
         outputs = kwargs.get('outputs', [])
         app_fut._outputs = []
-        for idx, f in enumerate(outputs):
-            if isinstance(f, File) and not self.check_staging_inhibited(kwargs):
+
+        # Pass over all possible outputs: the outputs kwarg, stdout and stderr
+        # and for each of those, perform possible stage-out. This can result in:
+        # a DataFuture to be exposed in app_fut to represent the completion of
+        # that stageout (sometimes backed by a new sub-workflow for separate-task
+        # stageout), a replacement for the function to be executed (intended to
+        # be the original function wrapped with an in-task stageout wrapper), a
+        # rewritten File object to be passed to task to be executed
+
+        def stageout_one_file(file: File, rewritable_func: Callable):
+            if not self.check_staging_inhibited(kwargs):
                 # replace a File with a DataFuture - either completing when the stageout
                 # future completes, or if no stage out future is returned, then when the
                 # app itself completes.
 
                 # The staging code will get a clean copy which it is allowed to mutate,
                 # while the DataFuture-contained original will not be modified by any staging.
-                f_copy = f.cleancopy()
-                outputs[idx] = f_copy
+                f_copy = file.cleancopy()
 
-                logger.debug("Submitting stage out for output file {}".format(repr(f)))
+                logger.debug("Submitting stage out for output file {}".format(repr(file)))
                 stageout_fut = self.data_manager.stage_out(f_copy, executor, app_fut)
                 if stageout_fut:
-                    logger.debug("Adding a dependency on stageout future for {}".format(repr(f)))
-                    app_fut._outputs.append(DataFuture(stageout_fut, f, tid=app_fut.tid))
+                    logger.debug("Adding a dependency on stageout future for {}".format(repr(file)))
+                    df = DataFuture(stageout_fut, file, tid=app_fut.tid)
                 else:
-                    logger.debug("No stageout dependency for {}".format(repr(f)))
-                    app_fut._outputs.append(DataFuture(app_fut, f, tid=app_fut.tid))
+                    logger.debug("No stageout dependency for {}".format(repr(file)))
+                    df = DataFuture(app_fut, file, tid=app_fut.tid)
 
                 # this is a hook for post-task stageout
                 # note that nothing depends on the output - which is maybe a bug
                 # in the not-very-tested stageout system?
-                func = self.data_manager.replace_task_stage_out(f_copy, func, executor)
+                rewritable_func = self.data_manager.replace_task_stage_out(f_copy, rewritable_func, executor)
+                return rewritable_func, f_copy, df
             else:
-                logger.debug("Not performing output staging for: {}".format(repr(f)))
-                app_fut._outputs.append(DataFuture(app_fut, f, tid=app_fut.tid))
+                logger.debug("Not performing output staging for: {}".format(repr(file)))
+                return rewritable_func, file, DataFuture(app_fut, file, tid=app_fut.tid)
+
+        for idx, file in enumerate(outputs):
+            func, outputs[idx], o = stageout_one_file(file, func)
+            app_fut._outputs.append(o)
+
+        file = kwargs.get('stdout')
+        if isinstance(file, File):
+            func, kwargs['stdout'], app_fut._stdout_future = stageout_one_file(file, func)
+
+        file = kwargs.get('stderr')
+        if isinstance(file, File):
+            func, kwargs['stderr'], app_fut._stderr_future = stageout_one_file(file, func)
+
         return func
 
     def _gather_all_deps(self, args: Sequence[Any], kwargs: Dict[str, Any]) -> List[Future]:
@@ -970,32 +995,16 @@ class DataFlowKernel:
         executor = random.choice(choices)
         logger.debug("Task {} will be sent to executor {}".format(task_id, executor))
 
-        # The below uses func.__name__ before it has been wrapped by any staging code.
-
-        label = app_kwargs.get('label')
-        for kw in ['stdout', 'stderr']:
-            if kw in app_kwargs:
-                if app_kwargs[kw] == parsl.AUTO_LOGNAME:
-                    if kw not in ignore_for_cache:
-                        ignore_for_cache += [kw]
-                    app_kwargs[kw] = os.path.join(
-                                self.run_dir,
-                                'task_logs',
-                                str(int(task_id / 10000)).zfill(4),  # limit logs to 10k entries per directory
-                                'task_{}_{}{}.{}'.format(
-                                    str(task_id).zfill(4),
-                                    func.__name__,
-                                    '' if label is None else '_{}'.format(label),
-                                    kw)
-                    )
-
         resource_specification = app_kwargs.get('parsl_resource_specification', {})
 
         task_record: TaskRecord
-        task_record = {'depends': [],
+        task_record = {'args': app_args,
+                       'depends': [],
                        'dfk': self,
                        'executor': executor,
+                       'func': func,
                        'func_name': func.__name__,
+                       'kwargs': app_kwargs,
                        'memoize': cache,
                        'hashsum': None,
                        'exec_fu': None,
@@ -1017,18 +1026,30 @@ class DataFlowKernel:
 
         self.update_task_state(task_record, States.unsched)
 
+        for kw in ['stdout', 'stderr']:
+            if kw in app_kwargs:
+                if app_kwargs[kw] == parsl.AUTO_LOGNAME:
+                    if kw not in ignore_for_cache:
+                        ignore_for_cache += [kw]
+                    if self.config.std_autopath is None:
+                        app_kwargs[kw] = self.default_std_autopath(task_record, kw)
+                    else:
+                        app_kwargs[kw] = self.config.std_autopath(task_record, kw)
+
         app_fu = AppFuture(task_record)
+        task_record['app_fu'] = app_fu
 
         # Transform remote input files to data futures
         app_args, app_kwargs, func = self._add_input_deps(executor, app_args, app_kwargs, func)
 
         func = self._add_output_deps(executor, app_args, app_kwargs, app_fu, func)
 
+        # Replace the function invocation in the TaskRecord with whatever file-staging
+        # substitutions have been made.
         task_record.update({
                     'args': app_args,
                     'func': func,
-                    'kwargs': app_kwargs,
-                    'app_fu': app_fu})
+                    'kwargs': app_kwargs})
 
         assert task_id not in self.tasks
 
@@ -1140,8 +1161,6 @@ class DataFlowKernel:
             executor.hub_port = self.hub_zmq_port
             if self.monitoring:
                 executor.monitoring_radio = self.monitoring.radio
-            else:
-                executor.monitoring_radio = None
             if hasattr(executor, 'provider'):
                 if hasattr(executor.provider, 'script_dir'):
                     executor.provider.script_dir = os.path.join(self.run_dir, 'submit_scripts')
@@ -1221,8 +1240,10 @@ class DataFlowKernel:
                 self._checkpoint_timer.close()
 
         # Send final stats
+        logger.info("Sending end message for usage tracking")
         self.usage_tracker.send_end_message()
         self.usage_tracker.close()
+        logger.info("Closed usage tracking")
 
         logger.info("Closing job status poller")
         self.job_status_poller.close()
@@ -1402,14 +1423,33 @@ class DataFlowKernel:
                 logger.info(f"{name} for task {tid} will not be redirected.")
             elif isinstance(target, str):
                 logger.info(f"{name} for task {tid} will be redirected to {target}")
-            elif isinstance(target, tuple) and len(target) == 2:
+            elif isinstance(target, os.PathLike):
+                logger.info(f"{name} for task {tid} will be redirected to {os.fspath(target)}")
+            elif isinstance(target, tuple) and len(target) == 2 and isinstance(target[0], str):
                 logger.info(f"{name} for task {tid} will be redirected to {target[0]} with mode {target[1]}")
+            elif isinstance(target, tuple) and len(target) == 2 and isinstance(target[0], os.PathLike):
+                logger.info(f"{name} for task {tid} will be redirected to {os.fspath(target[0])} with mode {target[1]}")
+            elif isinstance(target, DataFuture):
+                logger.info(f"{name} for task {tid} will staged to {target.file_obj.url}")
             else:
                 logger.error(f"{name} for task {tid} has unknown specification: {target!r}")
 
         log_std_stream("Standard out", task_record['app_fu'].stdout)
         log_std_stream("Standard error", task_record['app_fu'].stderr)
 
+    def default_std_autopath(self, taskrecord, kw):
+        label = taskrecord['kwargs'].get('label')
+        task_id = taskrecord['id']
+        return os.path.join(
+            self.run_dir,
+            'task_logs',
+            str(int(task_id / 10000)).zfill(4),  # limit logs to 10k entries per directory
+            'task_{}_{}{}.{}'.format(
+                str(task_id).zfill(4),
+                taskrecord['func_name'],
+                '' if label is None else '_{}'.format(label),
+                kw))
+
 
 class DataFlowKernelLoader:
     """Manage which DataFlowKernel is active.

parsl/dataflow/futures.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 from concurrent.futures import Future
 import logging
 import threading
-from typing import Any, Optional, Sequence
+from typing import Any, Optional, Sequence, Union
 
 import parsl.app.app as app
 
@@ -70,13 +70,34 @@ class AppFuture(Future):
         self._outputs = []
         self.task_record = task_record
 
+        self._stdout_future: Optional[DataFuture] = None
+        self._stderr_future: Optional[DataFuture] = None
+
     @property
-    def stdout(self) -> Optional[str]:
-        return self.task_record['kwargs'].get('stdout')
+    def stdout(self) -> Union[None, str, DataFuture]:
+        """Return app stdout. If stdout was specified as a string, then this
+        property will return that string. If stdout was specified as a File,
+        then this property will return a DataFuture representing that file
+        stageout.
+        TODO: this can be a tuple too I think?"""
+        if self._stdout_future:
+            return self._stdout_future
+        else:
+            # this covers the str and None cases
+            return self.task_record['kwargs'].get('stdout')
 
     @property
-    def stderr(self) -> Optional[str]:
-        return self.task_record['kwargs'].get('stderr')
+    def stderr(self) -> Union[None, str, DataFuture]:
+        """Return app stderr. If stdout was specified as a string, then this
+        property will return that string. If stdout was specified as a File,
+        then this property will return a DataFuture representing that file
+        stageout.
+        TODO: this can be a tuple too I think?"""
+        if self._stderr_future:
+            return self._stderr_future
+        else:
+            # this covers the str and None cases
+            return self.task_record['kwargs'].get('stderr')
 
     @property
     def tid(self) -> int:

parsl/executors/base.py

@@ -1,3 +1,4 @@
+import os
 from abc import ABCMeta, abstractmethod
 from concurrent.futures import Future
 from typing import Any, Callable, Dict, Optional
@@ -45,6 +46,21 @@ class ParslExecutor(metaclass=ABCMeta):
     label: str = "undefined"
     radio_mode: str = "udp"
 
+    def __init__(
+        self,
+        *,
+        hub_address: Optional[str] = None,
+        hub_port: Optional[int] = None,
+        monitoring_radio: Optional[MonitoringRadio] = None,
+        run_dir: str = ".",
+        run_id: Optional[str] = None,
+    ):
+        self.hub_address = hub_address
+        self.hub_port = hub_port
+        self.monitoring_radio = monitoring_radio
+        self.run_dir = os.path.abspath(run_dir)
+        self.run_id = run_id
+
     def __enter__(self) -> Self:
         return self
 

parsl/executors/high_throughput/executor.py

@@ -14,6 +14,7 @@ import math
 import warnings
 
 import parsl.launchers
+from parsl.usage_tracking.api import UsageInformation
 from parsl.serialize import pack_res_spec_apply_message, deserialize
 from parsl.serialize.errors import SerializationError, DeserializationError
 from parsl.app.errors import RemoteExceptionWrapper
@@ -62,7 +63,7 @@ DEFAULT_LAUNCH_CMD = ("process_worker_pool.py {debug} {max_workers_per_node} "
                       "--available-accelerators {accelerators}")
 
 
-class HighThroughputExecutor(BlockProviderExecutor, RepresentationMixin):
+class HighThroughputExecutor(BlockProviderExecutor, RepresentationMixin, UsageInformation):
     """Executor designed for cluster-scale
 
     The HighThroughputExecutor system has the following components:
@@ -818,4 +819,9 @@ class HighThroughputExecutor(BlockProviderExecutor, RepresentationMixin):
             logger.info("Unable to terminate Interchange process; sending SIGKILL")
             self.interchange_proc.kill()
 
+        self.interchange_proc.close()
+
         logger.info("Finished HighThroughputExecutor shutdown attempt")
+
+    def get_usage_information(self):
+        return {"mpi": self.enable_mpi_mode}

parsl/executors/taskvine/executor.py

@@ -644,6 +644,12 @@ class TaskVineExecutor(BlockProviderExecutor, putils.RepresentationMixin):
             logger.debug("Joining on factory process")
             self._factory_process.join()
 
+        # Shutdown multiprocessing queues
+        self._ready_task_queue.close()
+        self._ready_task_queue.join_thread()
+        self._finished_task_queue.close()
+        self._finished_task_queue.join_thread()
+
         self._is_shutdown = True
         logger.debug("TaskVine shutdown completed")
 

parsl/executors/workqueue/executor.py

@@ -735,6 +735,12 @@ class WorkQueueExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         logger.debug("Joining on collector thread")
         self.collector_thread.join()
 
+        logger.debug("Closing multiprocessing queues")
+        self.task_queue.close()
+        self.task_queue.join_thread()
+        self.collector_queue.close()
+        self.collector_queue.join_thread()
+
         self.is_shutdown = True
         logger.debug("Work Queue shutdown completed")
 

parsl/monitoring/monitoring.py

@@ -195,6 +195,8 @@ class MonitoringHub(RepresentationMixin):
 
         try:
             comm_q_result = comm_q.get(block=True, timeout=120)
+            comm_q.close()
+            comm_q.join_thread()
         except queue.Empty:
             logger.error("Hub has not completed initialization in 120s. Aborting")
             raise Exception("Hub failed to start")
@@ -258,6 +260,19 @@ class MonitoringHub(RepresentationMixin):
             self.filesystem_proc.terminate()
             self.filesystem_proc.join()
 
+            logger.info("Closing monitoring multiprocessing queues")
+            self.exception_q.close()
+            self.exception_q.join_thread()
+            self.priority_msgs.close()
+            self.priority_msgs.join_thread()
+            self.resource_msgs.close()
+            self.resource_msgs.join_thread()
+            self.node_msgs.close()
+            self.node_msgs.join_thread()
+            self.block_msgs.close()
+            self.block_msgs.join_thread()
+            logger.info("Closed monitoring multiprocessing queues")
+
 
 @wrap_with_logs
 def filesystem_receiver(logdir: str, q: "queue.Queue[AddressedMonitoringMessage]", run_dir: str) -> None:

parsl/providers/kubernetes/kube.py

@@ -105,7 +105,26 @@ class KubernetesProvider(ExecutionProvider, RepresentationMixin):
         if not _kubernetes_enabled:
             raise OptionalModuleMissing(['kubernetes'],
                                         "Kubernetes provider requires kubernetes module and config.")
-        config.load_kube_config()
+        try:
+            config.load_kube_config()
+        except config.config_exception.ConfigException:
+            # `load_kube_config` assumes a local kube-config file, and fails if not
+            # present, raising:
+            #
+            #     kubernetes.config.config_exception.ConfigException: Invalid
+            #     kube-config file. No configuration found.
+            #
+            # Since running a parsl driver script on a kubernetes cluster is a common
+            # pattern to enable worker-interchange communication, this enables an
+            # in-cluster config to be loaded if a kube-config file isn't found.
+            #
+            # Based on: https://github.com/kubernetes-client/python/issues/1005
+            try:
+                config.load_incluster_config()
+            except config.config_exception.ConfigException:
+                raise config.config_exception.ConfigException(
+                    "Failed to load both kube-config file and in-cluster configuration."
+                )
 
         self.namespace = namespace
         self.image = image

parsl/tests/configs/local_threads_checkpoint_periodic.py

@@ -1,13 +1,11 @@
 from parsl.config import Config
 from parsl.executors.threads import ThreadPoolExecutor
 
-config = Config(
-    executors=[
-        ThreadPoolExecutor(
-            label='local_threads_checkpoint_periodic',
-            max_threads=1
-        )
-    ],
-    checkpoint_mode='periodic',
-    checkpoint_period='00:00:05'
-)
+
+def fresh_config():
+    tpe = ThreadPoolExecutor(label='local_threads_checkpoint_periodic', max_threads=1)
+    return Config(
+        executors=[tpe],
+        checkpoint_mode='periodic',
+        checkpoint_period='00:00:02'
+    )