Project import generated by Copybara.

GitOrigin-RevId: bb4b5dc68657b3c00105b1bf0f49ad93b63612ea

Author: zenotech-ci

epiccli/catalog.py

@@ -78,11 +78,13 @@ def list_instances(project_config):
         for instance in sorted_instances:
             price_per_hour = instance.get('price_per_hour', 0)
             if price_per_hour > 0:
+                efa_str = "EFA Supported" if instance.get('efa_supported') else "No EFA"
                 click.echo(
                     f"- {instance['instance_type']}: "
                     f"{instance['vcpus']} vCPUs, "
                     f"{instance['memory_gb']:.2f} GB Memory, "
-                    f"${instance.get('price_per_hour', 0):.4f}/hr"
+                    f"${instance.get('price_per_hour', 0):.4f}/hr, "
+                    f"{efa_str}"
                 )
 
     except requests.exceptions.RequestException as e:

epiccli/job.py

@@ -28,6 +28,21 @@ def create_job(job_json_file, project_config):
         with open(job_json_file) as f:
             job_data = json.load(f)
 
+        jobs_list = job_data.get("jobs", []) if isinstance(job_data, dict) and "jobs" in job_data else [job_data]
+        for job_item in jobs_list:
+            if isinstance(job_item, dict):
+                tasks = job_item.get("spec", {}).get("tasks", [])
+                for task in tasks:
+                    partitions = int(task.get("partitions", 1))
+                    nodes = int(task.get("nodes", 2)) if task.get("task_distribution") == "node" else 1
+                    
+                    if partitions < nodes:
+                        click.echo(f"Error: Number of partitions ({partitions}) must be greater than or equal to number of nodes ({nodes}).")
+                        return
+                    if partitions % nodes != 0:
+                        click.echo(f"Error: Number of partitions ({partitions}) must be exactly divisible by number of nodes ({nodes}).")
+                        return
+
         response = requests.post(f"{api_url}/job/", headers=headers, json=job_data)
         response.raise_for_status()
         job_response = response.json()

epiccli/main.py

@@ -459,6 +459,9 @@ def create_job_command(job_json_file, project_name):
 
     - `instance_types`: A list of instance types to use for the job. If not
       specified, the default instance types are used. 
+    - `capacity_reservation_id`: (Optional) The ID of a Capacity Reservation (e.g., Capacity Blocks for ML) to target for the job.
+    - `task_distribution`: (Optional) The task distribution strategy (e.g., "core" or "node").
+    - `nodes`: (Optional) The number of nodes to use for the job, typically used when `task_distribution` is "node".
     - Use the `epic catalog list-instances` command to see the available instance types.
 
     .. code-block:: json
@@ -492,15 +495,16 @@ def create_job_command(job_json_file, project_name):
                 "spec": {
                     "app_code": "my-openfoam2212",
                     "tasks": [{"reference": "main-task", 
-                                "partitions": 32, 
+                                "partitions": 64, 
+                                "nodes": 2,
                                 "runtime": 1, 
-                                "task_distribution": "core",
+                                "task_distribution": "node",
                                 "memory_gb": 16,
-                                "instance_types": ["m5.xlarge"] 
+                                "instance_types": ["c5n.18xlarge"] 
                             }]
                 },
                 "input_data": {"path": "v2212/motorBike"},
-                "app_options": {"base_command": "su sudofoam -c '. /usr/lib/openfoam/openfoam2212/etc/bashrc && ls -lta && ./Allclean && ./Allrun'"},
+                "app_options": {"base_command": "su sudofoam -c '. /usr/lib/openfoam/openfoam2212/etc/bashrc && ls -lta && cat /tmp/hostfile && mpirun -np 64 -x PATH -x LD_LIBRARY_PATH -x WM_PROJECT_DIR -x FOAM_SETTINGS --hostfile /tmp/hostfile --use-hwthread-cpus simpleFoam -parallel | tee run.log && reconstructPar -latestTime'"},
                 "cluster": {"queue_code": "batch-single-node"}
             }]
         }