parsl 2024.12.23__tar.gz → 2025.1.6__tar.gz

{parsl-2024.12.23/parsl.egg-info → parsl-2025.1.6}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: parsl
-Version: 2024.12.23
+Version: 2025.1.6
 Summary: Simple data dependent workflows in Python
 Home-page: https://github.com/Parsl/parsl
-Download-URL: https://github.com/Parsl/parsl/archive/2024.12.23.tar.gz
+Download-URL: https://github.com/Parsl/parsl/archive/2025.01.06.tar.gz
 Author: The Parsl Team
 Author-email: parsl@googlegroups.com
 License: Apache 2.0

{parsl-2024.12.23 → parsl-2025.1.6}/parsl/jobs/strategy.py

@@ -298,8 +298,8 @@ class Strategy:
                     # Scale in for htex
                     if isinstance(executor, HighThroughputExecutor):
                         if active_blocks > min_blocks:
-                            excess_slots = math.ceil(active_slots - (active_tasks * parallelism))
-                            excess_blocks = math.ceil(float(excess_slots) / (tasks_per_node * nodes_per_block))
+                            excess_slots = math.floor(active_slots - (active_tasks * parallelism))
+                            excess_blocks = math.floor(float(excess_slots) / (tasks_per_node * nodes_per_block))
                             excess_blocks = min(excess_blocks, active_blocks - min_blocks)
                             logger.debug(f"Requesting scaling in by {excess_blocks} blocks with idle time {self.max_idletime}s")
                             executor.scale_in_facade(excess_blocks, max_idletime=self.max_idletime)

{parsl-2024.12.23 → parsl-2025.1.6}/parsl/monitoring/db_manager.py

@@ -78,7 +78,7 @@ class Database:
 
     def _get_mapper(self, table_obj: Table) -> Mapper:
         all_mappers: Set[Mapper] = set()
-        for mapper_registry in mapperlib._all_registries():  # type: ignore[attr-defined]
+        for mapper_registry in mapperlib._all_registries():
             all_mappers.update(mapper_registry.mappers)
         mapper_gen = (
             mapper for mapper in all_mappers

{parsl-2024.12.23 → parsl-2025.1.6}/parsl/tests/test_monitoring/test_app_names.py

@@ -67,7 +67,7 @@ def test_app_name(get_app, expected_name, expected_result, tmpd_cwd):
     with engine.begin() as connection:
 
         def count_rows(table: str):
-            result = connection.execute(f"SELECT COUNT(*) FROM {table}")
+            result = connection.execute(sqlalchemy.text(f"SELECT COUNT(*) FROM {table}"))
             (c, ) = result.first()
             return c
 
@@ -81,6 +81,6 @@ def test_app_name(get_app, expected_name, expected_result, tmpd_cwd):
         assert count_rows("try") == 1
 
         # ... and has the expected name.
-        result = connection.execute("SELECT task_func_name FROM task")
+        result = connection.execute(sqlalchemy.text("SELECT task_func_name FROM task"))
         (c, ) = result.first()
         assert c == expected_name

{parsl-2024.12.23 → parsl-2025.1.6}/parsl/tests/test_monitoring/test_stdouterr.py

@@ -106,7 +106,7 @@ def test_stdstream_to_monitoring(stdx, expected_stdx, stream, tmpd_cwd, caplog):
     with engine.begin() as connection:
 
         def count_rows(table: str):
-            result = connection.execute(f"SELECT COUNT(*) FROM {table}")
+            result = connection.execute(sqlalchemy.text(f"SELECT COUNT(*) FROM {table}"))
             (c, ) = result.first()
             return c
 
@@ -120,7 +120,7 @@ def test_stdstream_to_monitoring(stdx, expected_stdx, stream, tmpd_cwd, caplog):
         assert count_rows("try") == 1
 
         # ... and has the expected name.
-        result = connection.execute(f"SELECT task_{stream} FROM task")
+        result = connection.execute(sqlalchemy.text(f"SELECT task_{stream} FROM task"))
         (c, ) = result.first()
 
         if isinstance(expected_stdx, str):

parsl-2025.1.6/parsl/tests/test_scaling/test_regression_3696_oscillation.py

@@ -0,0 +1,103 @@
+import math
+from unittest.mock import MagicMock
+
+import pytest
+
+from parsl.executors.high_throughput.executor import HighThroughputExecutor
+from parsl.jobs.states import JobState, JobStatus
+from parsl.jobs.strategy import Strategy
+
+
+# the parameterize tuple consists of:
+# Input:
+#  * number of tasks to mock the load as
+#  * number of workers per node
+# Expected output:
+#  * the number of blocks we should expect to be launched
+#    in this situation
+#
+# This test will configure an executor, then run strategize
+# a few times, asserting that it converges to the correct
+# number of blocks without oscillating.
+@pytest.mark.local
+@pytest.mark.parametrize("ns", [(14, 48, 1),    # values from issue #3696
+
+                                (1, 1, 1),      # one task needs one block
+
+                                (100, 1, 20),   # many one-task blocks, hitting hard-coded max blocks
+
+                                (47, 48, 1),    # some edge cases around #3696 values
+                                (48, 48, 1),    # "
+                                (49, 48, 2),    # "
+                                (149, 50, 3)])  # "
+def test_htex_strategy_does_not_oscillate(ns):
+    """Check for oscillations in htex scaling.
+    In issue 3696, with a large number of workers per block
+    and a smaller number of active tasks, the htex scaling
+    strategy oscillates between 0 and 1 active block, rather
+    than converging to 1 active block.
+    """
+
+    n_tasks, n_workers, n_blocks = ns
+
+    s = Strategy(strategy='htex_auto_scale', max_idletime=0)
+
+    provider = MagicMock()
+    executor = MagicMock(spec=HighThroughputExecutor)
+
+    statuses = {}
+
+    executor.provider = provider
+    executor.outstanding = n_tasks
+    executor.status_facade = statuses
+    executor.workers_per_node = n_workers
+
+    provider.parallelism = 1
+    provider.init_blocks = 0
+    provider.min_blocks = 0
+    provider.max_blocks = 20
+    provider.nodes_per_block = 1
+
+    def scale_out(n):
+        for _ in range(n):
+            statuses[len(statuses)] = JobStatus(state=JobState.PENDING)
+
+    executor.scale_out_facade.side_effect = scale_out
+
+    def scale_in(n, max_idletime=None):
+        # find n PENDING jobs and set them to CANCELLED
+        for k in statuses:
+            if n == 0:
+                return
+            if statuses[k].state == JobState.PENDING:
+                statuses[k].state = JobState.CANCELLED
+                n -= 1
+
+    executor.scale_in_facade.side_effect = scale_in
+
+    s.add_executors([executor])
+
+    # In issue #3696, this first strategise does initial and load based
+    # scale outs, because n_tasks > n_workers*0
+    s.strategize([executor])
+
+    executor.scale_out_facade.assert_called()
+    assert len(statuses) == n_blocks, "Should have launched n_blocks"
+    assert len([k for k in statuses if statuses[k].state == JobState.PENDING]) == n_blocks
+    # there might be several calls to scale_out_facade inside strategy,
+    # but the end effect should be that exactly one block is scaled out.
+
+    executor.scale_in_facade.assert_not_called()
+
+    # In issue #3696, this second strategize does a scale in, because n_tasks < n_workers*1
+    s.strategize([executor])
+
+    # assert that there should still be n_blocks pending blocks
+    assert len([k for k in statuses if statuses[k].state == JobState.PENDING]) == n_blocks
+    # this assert fails due to issue #3696
+
+    # Now check scale in happens with 0 load
+    executor.outstanding = 0
+    s.strategize([executor])
+    executor.scale_in_facade.assert_called()
+    assert len([k for k in statuses if statuses[k].state == JobState.PENDING]) == 0

{parsl-2024.12.23 → parsl-2025.1.6}/parsl/version.py

@@ -3,4 +3,4 @@
 Year.Month.Day[alpha/beta/..]
 Alphas will be numbered like this -> 2024.12.10a0
 """
-VERSION = '2024.12.23'
+VERSION = '2025.01.06'

{parsl-2024.12.23 → parsl-2025.1.6/parsl.egg-info}/PKG-INFO

@@ -1,9 +1,9 @@
 Metadata-Version: 2.1
 Name: parsl
-Version: 2024.12.23
+Version: 2025.1.6
 Summary: Simple data dependent workflows in Python
 Home-page: https://github.com/Parsl/parsl
-Download-URL: https://github.com/Parsl/parsl/archive/2024.12.23.tar.gz
+Download-URL: https://github.com/Parsl/parsl/archive/2025.01.06.tar.gz
 Author: The Parsl Team
 Author-email: parsl@googlegroups.com
 License: Apache 2.0

{parsl-2024.12.23 → parsl-2025.1.6}/parsl.egg-info/SOURCES.txt

@@ -412,6 +412,7 @@ parsl/tests/test_scaling/__init__.py
 parsl/tests/test_scaling/test_block_error_handler.py
 parsl/tests/test_scaling/test_regression_1621.py
 parsl/tests/test_scaling/test_regression_3568_scaledown_vs_MISSING.py
+parsl/tests/test_scaling/test_regression_3696_oscillation.py
 parsl/tests/test_scaling/test_scale_down.py
 parsl/tests/test_scaling/test_scale_down_htex_auto_scale.py
 parsl/tests/test_scaling/test_scale_down_htex_unregistered.py

{parsl-2024.12.23 → parsl-2025.1.6}/parsl.egg-info/requires.txt

@@ -10,9 +10,9 @@ setproctitle
 filelock<4,>=3.13
 
 [all]
-sqlalchemy<2,>=1.4
-pydot
-networkx<2.6,>=2.5
+sqlalchemy<2.1,>=2
+pydot>=1.4.2
+networkx<3.3,>=3.2
 Flask>=1.0.2
 flask_sqlalchemy
 pandas<2.2
@@ -66,7 +66,7 @@ python-gssapi
 kubernetes
 
 [monitoring]
-sqlalchemy<2,>=1.4
+sqlalchemy<2.1,>=2
 
 [proxystore]
 proxystore
@@ -76,8 +76,8 @@ radical.pilot==1.90
 radical.utils==1.90
 
 [visualization]
-pydot
-networkx<2.6,>=2.5
+pydot>=1.4.2
+networkx<3.3,>=3.2
 Flask>=1.0.2
 flask_sqlalchemy
 pandas<2.2

{parsl-2024.12.23 → parsl-2025.1.6}/setup.py

@@ -8,11 +8,18 @@ with open('requirements.txt') as f:
 
 extras_require = {
     'monitoring' : [
-        'sqlalchemy>=1.4,<2'
+        # sqlalchemy does not use semantic versioning.
+        # see https://github.com/sqlalchemy/sqlalchemy/discussions/11391#discussioncomment-9472033
+        'sqlalchemy>=2,<2.1'
     ],
     'visualization' : [
-        'pydot',
-        'networkx>=2.5,<2.6',
+        # this pydot bound is copied from networkx's pyproject.toml,
+        # version 3.2 (aa2de1adecea09f7b86ff6093b212ca86f22b3ef),
+        # because networkx[extra] installs quite a lot of extra stuff
+        # that needs more OS dependencies in addition to pydot.
+        'pydot>=1.4.2',
+
+        'networkx>=3.2,<3.3',
         'Flask>=1.0.2',
         'flask_sqlalchemy',
         'pandas<2.2',