Merge pull request #92 from LSSTDESC/in2p3-update

Update site for in2p3 for their new SLURM launcher

ceci/minirunner.py

@@ -9,6 +9,7 @@
 import subprocess
 import time
 from timeit import default_timer
+import sys
 
 # Constant indicators
 COMPLETE = 0
@@ -279,6 +280,7 @@ def _launch(self, job, alloc):
                 EVENT_LAUNCH,
                 {"job": job, "stdout": stdout_file, "process": p, "nodes": alloc},
             )
+        sys.stdout.flush()
 
     def _ready_jobs(self):
         # Find jobs ready to be run now
@@ -366,7 +368,7 @@ def _check_completed(self):
                 completed_jobs.append(job)
                 for node in alloc:
                     node.free()
-
+        sys.stdout.flush()
         self.running = continuing_jobs
 
         for job in completed_jobs:

ceci/sites/ccin2p3.py

@@ -7,6 +7,7 @@
 
 class CCParallel(Site):
     """Object representing execution in the local environment, e.g. a laptop."""
+    default_mpi_command = "mpirun -n"
 
     def command(self, cmd, sec):
         """Generate a complete command line to be run with the specified execution variables.
@@ -74,17 +75,11 @@ def configure_for_parsl(self):  # pylint: disable=no-self-use
 
     def configure_for_mini(self):
         """Utility function to setup self for local execution"""
-        total_cores = int(os.environ["NSLOTS"])
-        cores_per_node = 16  # seems to be the case
-        nodes = total_cores // cores_per_node
-        last_node_codes = total_cores % cores_per_node
+        nodes = int(os.environ.get("SLURM_JOB_NUM_NODES", "1"))
+        cores_per_node = int(os.environ.get("SLURM_CPUS_ON_NODE", "1"))
 
         nodes = [Node(f"Node_{i}", cores_per_node) for i in range(nodes)]
 
-        if last_node_codes:
-            i = len(nodes)
-            nodes.append(Node(f"Node_{i}", last_node_codes))
-
         self.info["nodes"] = nodes
 
     def configure_for_cwl(self):