conductor(checkpoint): Checkpoint end of Phase 1

scripts/cluster_status/nomad_client.py

@@ -1,5 +1,6 @@
 import subprocess
 import re
+import sys
 
 def get_node_map():
     """
@@ -18,8 +19,14 @@ def get_node_map():
                 if len(parts) >= 4:
                     node_map[parts[0]] = parts[3]
         return node_map
+    except FileNotFoundError:
+        print("Warning: 'nomad' binary not found in PATH.", file=sys.stderr)
+        return {}
+    except subprocess.CalledProcessError as e:
+        print(f"Warning: Failed to query Nomad nodes: {e}", file=sys.stderr)
+        return {}
     except Exception as e:
-        print(f"Error getting node map: {e}")
+        print(f"Error getting node map: {e}", file=sys.stderr)
         return {}
 
 def get_allocation_id(node_name, job_id):
@@ -57,8 +64,10 @@ def get_allocation_id(node_name, job_id):
                                 return l.split("=")[1].strip()
                         return alloc_id
                         
+    except FileNotFoundError:
+        return None # Warning already printed by get_node_map likely
     except Exception as e:
-        print(f"Error getting allocation ID: {e}")
+        print(f"Error getting allocation ID: {e}", file=sys.stderr)
         
     return None
 
@@ -81,7 +90,23 @@ def get_allocation_logs(alloc_id, tail=20):
             )
             return result.stdout
     except Exception as e:
-        return f"Error fetching logs: {e}"
+        # Don't print stack trace, just the error
+        return f"Nomad Error: {str(e)}"
+
+def exec_command(alloc_id, command, task="navidrome"):
+    """
+    Executes a command inside a specific allocation and task.
+    """
+    try:
+        args = ["nomad", "alloc", "exec", "-task", task, alloc_id] + command
+        result = subprocess.run(
+            args,
+            capture_output=True, text=True, check=True
+        )
+        return result.stdout
+    except Exception as e:
+        # Don't print stack trace, just return error string
+        return f"Nomad Error: {str(e)}"
 
 def restart_allocation(alloc_id):
     """
@@ -94,5 +119,5 @@ def restart_allocation(alloc_id):
         )
         return True
     except Exception as e:
-        print(f"Error restarting allocation: {e}")
+        print(f"Error restarting allocation: {e}", file=sys.stderr)
         return False