parsl 2024.4.22__tar.gz → 2024.5.6__tar.gz

{parsl-2024.4.22/parsl.egg-info → parsl-2024.5.6}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: parsl
-Version: 2024.4.22
+Version: 2024.5.6
 Summary: Simple data dependent workflows in Python
 Home-page: https://github.com/Parsl/parsl
-Download-URL: https://github.com/Parsl/parsl/archive/2024.04.22.tar.gz
+Download-URL: https://github.com/Parsl/parsl/archive/2024.05.06.tar.gz
 Author: The Parsl Team
 Author-email: parsl@googlegroups.com
 License: Apache 2.0

{parsl-2024.4.22 → parsl-2024.5.6}/README.rst

@@ -15,8 +15,6 @@ Parsl lets you chain functions together and will launch each function as inputs
     import parsl
     from parsl import python_app
 
-    # Start Parsl on a single computer
-    parsl.load()
 
     # Make functions parallel by decorating them
     @python_app
@@ -27,12 +25,14 @@ Parsl lets you chain functions together and will launch each function as inputs
     def g(x):
         return x * 2
 
-    # These functions now return Futures, and can be chained
-    future = f(1)
-    assert future.result() == 2
+    # Start Parsl on a single computer
+    with parsl.load():
+        # These functions now return Futures, and can be chained
+        future = f(1)
+        assert future.result() == 2
 
-    future = g(f(1))
-    assert future.result() == 4
+        future = g(f(1))
+        assert future.result() == 4
 
 
 Start with the `configuration quickstart <https://parsl.readthedocs.io/en/stable/quickstart.html#getting-started>`_ to learn how to tell Parsl how to use your computing resource,

{parsl-2024.4.22 → parsl-2024.5.6}/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-2024.4.22 → parsl-2024.5.6}/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-2024.4.22 → parsl-2024.5.6}/parsl/dataflow/dflow.py

@@ -177,11 +177,9 @@ class DataFlowKernel:
 
         # this must be set before executors are added since add_executors calls
         # job_status_poller.add_executors.
-        radio = self.monitoring.radio if self.monitoring else None
         self.job_status_poller = JobStatusPoller(strategy=self.config.strategy,
                                                  strategy_period=self.config.strategy_period,
-                                                 max_idletime=self.config.max_idletime,
-                                                 monitoring=radio)
+                                                 max_idletime=self.config.max_idletime)
 
         self.executors: Dict[str, ParslExecutor] = {}
 
@@ -798,7 +796,6 @@ class DataFlowKernel:
         # be the original function wrapped with an in-task stageout wrapper), a
         # rewritten File object to be passed to task to be executed
 
-        @typechecked
         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
@@ -996,32 +993,16 @@ 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,
@@ -1043,18 +1024,30 @@ 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
 
@@ -1245,8 +1238,10 @@ 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()
@@ -1440,6 +1435,19 @@ class DataFlowKernel:
         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-2024.4.22 → parsl-2024.5.6}/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-2024.4.22 → parsl-2024.5.6}/parsl/executors/status_handling.py

@@ -43,9 +43,6 @@ class BlockProviderExecutor(ParslExecutor):
     invoking scale_out, but it will not initialize the blocks requested by
     any init_blocks parameter. Subclasses must implement that behaviour
     themselves.
-
-    BENC: TODO: block error handling: maybe I want this more user pluggable?
-    I'm not sure of use cases for switchability at the moment beyond "yes or no"
     """
     def __init__(self, *,
                  provider: Optional[ExecutionProvider],

{parsl-2024.4.22 → parsl-2024.5.6}/parsl/executors/taskvine/executor.py

@@ -4,7 +4,6 @@ high-throughput system for delegating Parsl tasks to thousands of remote machine
 """
 
 # Import Python built-in libraries
-import atexit
 import threading
 import multiprocessing
 import logging
@@ -180,24 +179,6 @@ class TaskVineExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         else:
             self._poncho_available = True
 
-        # Register atexit handler to cleanup when Python shuts down
-        atexit.register(self.atexit_cleanup)
-
-        # Attribute indicating whether this executor was started to shut it down properly.
-        # This safeguards cases where an object of this executor is created but
-        # the executor never starts, so it shouldn't be shutdowned.
-        self._is_started = False
-
-        # Attribute indicating whether this executor was shutdown before.
-        # This safeguards cases where this object is automatically shut down (e.g.,
-        # via atexit) and the user also explicitly calls shut down. While this is
-        # permitted, the effect of an executor shutdown should happen only once.
-        self._is_shutdown = False
-
-    def atexit_cleanup(self):
-        # Calls this executor's shutdown method upon Python exiting the process.
-        self.shutdown()
-
     def _get_launch_command(self, block_id):
         # Implements BlockProviderExecutor's abstract method.
         # This executor uses different terminology for worker/launch
@@ -257,9 +238,6 @@ class TaskVineExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         retrieve Parsl tasks within the TaskVine system.
         """
 
-        # Mark this executor object as started
-        self._is_started = True
-
         # Synchronize connection and communication settings between the manager and factory
         self.__synchronize_manager_factory_comm_settings()
 
@@ -618,14 +596,6 @@ class TaskVineExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         """Shutdown the executor. Sets flag to cancel the submit process and
         collector thread, which shuts down the TaskVine system submission.
         """
-        if not self._is_started:
-            # Don't shutdown if the executor never starts.
-            return
-
-        if self._is_shutdown:
-            # Don't shutdown this executor again.
-            return
-
         logger.debug("TaskVine shutdown started")
         self._should_stop.set()
 
@@ -650,7 +620,6 @@ class TaskVineExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         self._finished_task_queue.close()
         self._finished_task_queue.join_thread()
 
-        self._is_shutdown = True
         logger.debug("TaskVine shutdown completed")
 
     @wrap_with_logs

{parsl-2024.4.22 → parsl-2024.5.6}/parsl/executors/workqueue/executor.py

@@ -3,7 +3,6 @@ Cooperative Computing Lab (CCL) at Notre Dame to provide a fault-tolerant,
 high-throughput system for delegating Parsl tasks to thousands of remote machines
 """
 
-import atexit
 import threading
 import multiprocessing
 import logging
@@ -298,24 +297,6 @@ class WorkQueueExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         if self.init_command != "":
             self.launch_cmd = self.init_command + "; " + self.launch_cmd
 
-        # register atexit handler to cleanup when Python shuts down
-        atexit.register(self.atexit_cleanup)
-
-        # Attribute indicating whether this executor was started to shut it down properly.
-        # This safeguards cases where an object of this executor is created but
-        # the executor never starts, so it shouldn't be shutdowned.
-        self.is_started = False
-
-        # Attribute indicating whether this executor was shutdown before.
-        # This safeguards cases where this object is automatically shut down (e.g.,
-        # via atexit) and the user also explicitly calls shut down. While this is
-        # permitted, the effect of an executor shutdown should happen only once.
-        self.is_shutdown = False
-
-    def atexit_cleanup(self):
-        # Calls this executor's shutdown method upon Python exiting the process.
-        self.shutdown()
-
     def _get_launch_command(self, block_id):
         # this executor uses different terminology for worker/launch
         # commands than in htex
@@ -325,8 +306,6 @@ class WorkQueueExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         """Create submit process and collector thread to create, send, and
         retrieve Parsl tasks within the Work Queue system.
         """
-        # Mark this executor object as started
-        self.is_started = True
         self.tasks_lock = threading.Lock()
 
         # Create directories for data and results
@@ -713,14 +692,6 @@ class WorkQueueExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         """Shutdown the executor. Sets flag to cancel the submit process and
         collector thread, which shuts down the Work Queue system submission.
         """
-        if not self.is_started:
-            # Don't shutdown if the executor never starts.
-            return
-
-        if self.is_shutdown:
-            # Don't shutdown this executor again.
-            return
-
         logger.debug("Work Queue shutdown started")
         self.should_stop.value = True
 
@@ -741,7 +712,6 @@ class WorkQueueExecutor(BlockProviderExecutor, putils.RepresentationMixin):
         self.collector_queue.close()
         self.collector_queue.join_thread()
 
-        self.is_shutdown = True
         logger.debug("Work Queue shutdown completed")
 
     @wrap_with_logs

{parsl-2024.4.22 → parsl-2024.5.6}/parsl/jobs/job_status_poller.py

@@ -1,5 +1,4 @@
 import logging
-import parsl
 from typing import List, Sequence, Optional, Union
 
 from parsl.jobs.strategy import Strategy
@@ -14,8 +13,7 @@ logger = logging.getLogger(__name__)
 
 class JobStatusPoller(Timer):
     def __init__(self, *, strategy: Optional[str], max_idletime: float,
-                 strategy_period: Union[float, int],
-                 monitoring: Optional["parsl.monitoring.radios.MonitoringRadio"] = None) -> None:
+                 strategy_period: Union[float, int]) -> None:
         self._executors = []  # type: List[BlockProviderExecutor]
         self._strategy = Strategy(strategy=strategy,
                                   max_idletime=max_idletime)

{parsl-2024.4.22 → parsl-2024.5.6}/parsl/monitoring/monitoring.py

@@ -244,6 +244,7 @@ class MonitoringHub(RepresentationMixin):
             self.router_exit_event.set()
             logger.info("Waiting for router to terminate")
             self.router_proc.join()
+            self.router_proc.close()
             logger.debug("Finished waiting for router termination")
             if len(exception_msgs) == 0:
                 logger.debug("Sending STOP to DBM")
@@ -252,6 +253,7 @@ class MonitoringHub(RepresentationMixin):
                 logger.debug("Not sending STOP to DBM, because there were DBM exceptions")
             logger.debug("Waiting for DB termination")
             self.dbm_proc.join()
+            self.dbm_proc.close()
             logger.debug("Finished waiting for DBM termination")
 
             # should this be message based? it probably doesn't need to be if
@@ -259,6 +261,7 @@ class MonitoringHub(RepresentationMixin):
             logger.info("Terminating filesystem radio receiver process")
             self.filesystem_proc.terminate()
             self.filesystem_proc.join()
+            self.filesystem_proc.close()
 
             logger.info("Closing monitoring multiprocessing queues")
             self.exception_q.close()

{parsl-2024.4.22 → parsl-2024.5.6}/parsl/monitoring/radios.py

@@ -177,7 +177,7 @@ class UDPRadio(MonitoringRadio):
 
 
 class MultiprocessingQueueRadio(MonitoringRadio):
-    """A monitoring radio intended which connects over a multiprocessing Queue.
+    """A monitoring radio which connects over a multiprocessing Queue.
     This radio is intended to be used on the submit side, where components
     in the submit process, or processes launched by multiprocessing, will have
     access to a Queue shared with the monitoring database code (bypassing the

{parsl-2024.4.22 → parsl-2024.5.6}/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-2024.5.6/parsl/tests/configs/local_threads_checkpoint_periodic.py

@@ -0,0 +1,11 @@
+from parsl.config import Config
+from parsl.executors.threads import ThreadPoolExecutor
+
+
+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'
+    )

{parsl-2024.4.22 → parsl-2024.5.6}/parsl/tests/conftest.py

@@ -3,8 +3,10 @@ import itertools
 import logging
 import os
 import pathlib
+import random
 import re
 import shutil
+import string
 import time
 import types
 import signal
@@ -139,7 +141,7 @@ def pytest_configure(config):
     )
     config.addinivalue_line(
         'markers',
-        'staging_required: Marks tests that require a staging provider, when there is no sharedFS)'
+        'staging_required: Marks tests that require a staging provider, when there is no sharedFS'
     )
     config.addinivalue_line(
         'markers',
@@ -245,6 +247,7 @@ def load_dfk_local_module(request, pytestconfig, tmpd_cwd_session):
 
         if callable(local_teardown):
             local_teardown()
+            assert DataFlowKernelLoader._dfk is None, "Expected teardown to clear DFK"
 
         if local_config:
             if parsl.dfk() != dfk:
@@ -421,3 +424,11 @@ def try_assert():
             raise AssertionError("Bad assert call: no attempts or timeout period")
 
     yield _impl
+
+
+@pytest.fixture
+def randomstring():
+    def func(length=5, alphabet=string.ascii_letters):
+        return "".join(random.choice(alphabet) for _ in range(length))
+
+    return func