Merge pull request #850 from linsword13/fix-slurm-wm

Fix slurm batch_submit self-reference

lib/ramble/ramble/test/workflow_manager_functionality/slurm_workflow_manager.py

@@ -21,6 +21,46 @@
 )
 
 
+def test_slurm_workflow_default():
+    workspace_name = "test_slurm_workflow_default"
+
+    test_config = """
+ramble:
+  variants:
+    workflow_manager: slurm
+  variables:
+    processes_per_node: 1
+    n_nodes: 1
+  applications:
+    hostname:
+      workloads:
+        local:
+          experiments:
+            test_default: {}
+"""
+    with ramble.workspace.create(workspace_name) as ws:
+        ws.write()
+        config_path = os.path.join(ws.config_dir, ramble.workspace.config_file_name)
+        with open(config_path, "w+") as f:
+            f.write(test_config)
+        ws._re_read()
+        workspace("setup", "--dry-run", global_args=["-D", ws.root])
+
+        path = os.path.join(ws.experiment_dir, "hostname", "local", "test_default")
+        files = [f for f in os.listdir(path) if os.path.isfile(os.path.join(path, f))]
+        assert "batch_submit" in files
+        assert "batch_query" in files
+        assert "batch_cancel" in files
+        assert "batch_wait" in files
+        with open(os.path.join(path, "batch_submit")) as f:
+            content = f.read()
+            assert "slurm_experiment_sbatch" in content
+            assert "execute_experiment" not in content
+            assert ".slurm_job" in content
+            assert "sbatch" in content
+            assert "batch_submit" not in content
+
+
 def test_slurm_workflow():
     workspace_name = "test_slurm_workflow"
 

var/ramble/repos/builtin/workflow_managers/slurm/workflow_manager.py

@@ -108,7 +108,7 @@ def _batch_submit_vars(self):
                     "`sbatch` is missing in the given `batch_submit` command"
                 )
         else:
-            batch_submit_script = vars["batch_submit"]
+            batch_submit_script = vars["slurm_experiment_sbatch"]
             batch_submit_cmd = f"sbatch {batch_submit_script}"
         return {
             "batch_submit_cmd": batch_submit_cmd,
@@ -151,7 +151,6 @@ def template_render_vars(self):
             ("#SBATCH --gpus-per-node {gpus_per_node}"),
         ]
         partition = expander.expand_var_name("slurm_partition")
-        self._check_partition(partition)
         if partition:
             pragmas.append("#SBATCH -p {slurm_partition}")
         extra_headers = (
@@ -165,33 +164,6 @@ def template_render_vars(self):
             "workflow_hostfile_cmd": self.runner.get_hostfile_cmd(),
         }
 
-    def _check_partition(self, partition):
-        """Warns about potential issues of the slurm_partition config
-
-        Only gives out warning as the user may be relying on a custom
-        execute template that contains the relevant partition info.
-        """
-        try:
-            partition_prop = self.runner.get_partitions()
-        except RunnerError:
-            return
-        if partition_prop is None:
-            return
-
-        partitions = partition_prop["partitions"]
-        if partition not in partitions:
-            default_partition = partition_prop["default_partition"]
-            if default_partition is not None and not partition:
-                logger.info(
-                    "`slurm_partition` is not given, "
-                    f"using default partition {default_partition}"
-                )
-            else:
-                logger.warn(
-                    "Missing valid `slurm_partition` setting. "
-                    f"It should be one of {partitions}"
-                )
-
     def get_status(self, workspace):
         expander = self.app_inst.expander
         run_dir = expander.expand_var_name("experiment_run_dir")