lsst-ctrl-execute 29.2025.1000__tar.gz → 29.2025.2100__tar.gz

{lsst_ctrl_execute-29.2025.1000 → lsst_ctrl_execute-29.2025.2100}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lsst-ctrl-execute
-Version: 29.2025.1000
+Version: 29.2025.2100
 Summary: Utilities for executing and managing workloads.
 Project-URL: Homepage, https://github.com/lsst/ctrl_execute
 Author-email: Rubin Observatory Data Management <dm-admin@lists.lsst.org>

{lsst_ctrl_execute-29.2025.1000 → lsst_ctrl_execute-29.2025.2100}/python/lsst/ctrl/execute/allocator.py

@@ -30,9 +30,7 @@ from datetime import datetime
 from string import Template
 
 from lsst.ctrl.execute.allocationConfig import AllocationConfig
-from lsst.ctrl.execute.condorConfig import CondorConfig
 from lsst.ctrl.execute.condorInfoConfig import CondorInfoConfig
-from lsst.ctrl.execute.findPackageFile import find_package_file
 from lsst.ctrl.execute.templateWriter import TemplateWriter
 from lsst.resources import ResourcePath, ResourcePathExpression
 
@@ -79,10 +77,6 @@ class Allocator:
 
         self.platform = platform
 
-        execConfigName = find_package_file("execConfig.py", platform=platform)
-        execConfig = CondorConfig()
-        execConfig.loadFromStream(execConfigName.read())
-
         # Look up the user's name and home and scratch directory in the
         # $HOME/.lsst/condor-info.py file
         user_name = None
@@ -115,7 +109,7 @@ class Allocator:
         self.commandLineDefaults["COLLECTOR"] = self.opts.collector
         self.commandLineDefaults["CPORT"] = self.opts.collectorport
         if self.opts.exclusive:
-            self.commandLineDefaults["CPUS"] = execConfig.platform.peakcpus
+            self.commandLineDefaults["CPUS"] = self.configuration.platform.peakcpus
         else:
             self.commandLineDefaults["CPUS"] = self.opts.cpus
         self.commandLineDefaults["WALL_CLOCK"] = self.opts.maximumWallClock
@@ -123,6 +117,7 @@ class Allocator:
         self.commandLineDefaults["MEMPERCORE"] = 4096
         self.commandLineDefaults["ALLOWEDAUTO"] = 500
         self.commandLineDefaults["AUTOCPUS"] = 16
+        self.commandLineDefaults["MINAUTOCPUS"] = 15
         self.commandLineDefaults["QUEUE"] = self.opts.queue
         self.load()
 
@@ -361,6 +356,12 @@ class Allocator:
         else:
             return self.getParameter("AUTOCPUS")
 
+    def getMinAutoCPUs(self):
+        """Minimum Size of standard glideins for allocateNodes auto
+        @return the value of minAutoCPUs
+        """
+        return self.getParameter("MINAUTOCPUS")
+
     def getWallClock(self):
         """Accessor for WALL_CLOCK
         @return the value of WALL_CLOCK

{lsst_ctrl_execute-29.2025.1000 → lsst_ctrl_execute-29.2025.2100}/python/lsst/ctrl/execute/allocatorParser.py

@@ -178,6 +178,24 @@ class AllocatorParser:
             dest="packnodes",
             help="encourage nodes to pack jobs rather than spread",
         )
+        parser.add_argument(
+            "--exclude",
+            action="store",
+            default=None,
+            dest="exclude",
+            help="machine name(s) to exclude Slurm jobs from",
+            type=str,
+            required=False,
+        )
+        parser.add_argument(
+            "--nodelist",
+            action="store",
+            default=None,
+            dest="nodelist",
+            help="machine name(s) to target Slurm jobs toward",
+            type=str,
+            required=False,
+        )
         parser.add_argument(
             "--exclusive",
             action="store_true",

{lsst_ctrl_execute-29.2025.1000 → lsst_ctrl_execute-29.2025.2100}/python/lsst/ctrl/execute/slurmPlugin.py

@@ -214,6 +214,16 @@ class SlurmPlugin(Allocator):
         else:
             self.defaults["RESERVATION"] = ""
 
+        if self.opts.exclude is not None:
+            self.defaults["EXCLUDE"] = f"#SBATCH --exclude {self.opts.exclude}"
+        else:
+            self.defaults["EXCLUDE"] = ""
+
+        if self.opts.nodelist is not None:
+            self.defaults["NODELIST"] = f"#SBATCH --nodelist {self.opts.nodelist}"
+        else:
+            self.defaults["NODELIST"] = ""
+
         if self.opts.exclusive is not None:
             self.defaults["EXCLUSIVE"] = "#SBATCH --exclusive"
         else:
@@ -291,7 +301,11 @@ class SlurmPlugin(Allocator):
         """Determine and submit the glideins needed from job pressure."""
 
         verbose = self.isVerbose()
+        cpus = self.getCPUs()
         autoCPUs = self.getAutoCPUs()
+        minAutoCPUs = self.getMinAutoCPUs()
+        if cpus >= minAutoCPUs:
+            autoCPUs = cpus
         memoryPerCore = self.getMemoryPerCore()
         memoryLimit = autoCPUs * memoryPerCore
         auser = self.getUserName()

{lsst_ctrl_execute-29.2025.1000 → lsst_ctrl_execute-29.2025.2100}/tests/test_allocatorParser.py

@@ -42,6 +42,10 @@ class TestAllocatorParser(lsst.utils.tests.TestCase):
             "12",
             "-m",
             "00:30:00",
+            "--exclude",
+            "sdfmilan003",
+            "--nodelist",
+            "sdfmilan004",
             "-q",
             "normal",
             "-O",
@@ -57,6 +61,8 @@ class TestAllocatorParser(lsst.utils.tests.TestCase):
         self.assertEqual(args.nodeCount, 64)
         self.assertEqual(args.cpus, 12)
         self.assertEqual(args.maximumWallClock, "00:30:00")
+        self.assertEqual(args.exclude, "sdfmilan003")
+        self.assertEqual(args.nodelist, "sdfmilan004")
         self.assertEqual(args.queue, "normal")
         self.assertEqual(args.outputLog, "outlog")
         self.assertEqual(args.errorLog, "errlog")

{lsst_ctrl_execute-29.2025.1000 → lsst_ctrl_execute-29.2025.2100}/tests/test_slurmPlugin.py

@@ -26,6 +26,7 @@ import sys
 import unittest
 
 import lsst.utils.tests
+from lsst.ctrl.execute.allocator import Allocator
 from lsst.ctrl.execute.allocatorParser import AllocatorParser
 from lsst.ctrl.execute.condorConfig import CondorConfig
 from lsst.ctrl.execute.namedClassFactory import NamedClassFactory
@@ -77,6 +78,25 @@ class SlurmPluginTest(lsst.utils.tests.TestCase):
         self.assertTrue(self.config)
         self.assertTrue(condor_info_file)
 
+        platform = "test1"
+        configuration = CondorConfig()
+        p1 = os.path.join("tests/testfiles", "config_execconfig.py")
+        execConfigName = p1
+        configuration.load(execConfigName)
+        scheduler: Allocator = schedulerClass(platform, args, configuration, condor_info_file)
+        self.assertTrue(scheduler)
+
+        autocpus = scheduler.getAutoCPUs()
+        minautocpus = scheduler.getMinAutoCPUs()
+        cpus = scheduler.getCPUs()
+        nodes = scheduler.getNodes()
+        wallclock = scheduler.getWallClock()
+        self.assertEqual(autocpus, 16)
+        self.assertEqual(minautocpus, 15)
+        self.assertEqual(cpus, 12)
+        self.assertEqual(nodes, 64)
+        self.assertEqual(wallclock, "00:30:00")
+
 
 if __name__ == "__main__":
     lsst.utils.tests.init()

lsst_ctrl_execute-29.2025.2100/tests/testfiles/config_execconfig.py

@@ -0,0 +1,6 @@
+# flake8: noqa
+config.platform.nodeSetRequired = False
+config.platform.localScratch = "/condor_scratch"
+config.platform.fileSystemDomain = "slac.stanford.edu"
+config.platform.scheduler = "slurm"
+config.platform.peakcpus = 120