Fix lint and formatting issues

cvs/monitors/cluster-mon/backend/app/api/cluster.py

@@ -179,7 +179,9 @@ async def get_cluster_status() -> Dict[str, Any]:
         if isinstance(node_data, dict) and "error" not in node_data:
             for gpu_id, gpu_metrics in node_data.items():
                 if isinstance(gpu_metrics, dict):
-                    temp = gpu_metrics.get("Temperature (Sensor junction) (C)") or gpu_metrics.get("Temperature (Sensor edge) (C)")
+                    temp = gpu_metrics.get("Temperature (Sensor junction) (C)") or gpu_metrics.get(
+                        "Temperature (Sensor edge) (C)"
+                    )
                     if temp:
                         total_temp += float(temp)
                         temp_count += 1
@@ -242,22 +244,14 @@ async def get_cluster_health() -> Dict[str, Any]:
         # Create alerts for unhealthy nodes
         if node_status == "unhealthy":
             for issue in issues:
-                health_data["alerts"].append({
-                    "severity": "warning",
-                    "node": node,
-                    "message": issue
-                })
+                health_data["alerts"].append({"severity": "warning", "node": node, "message": issue})
 
     # Check unreachable nodes
     for node in app_state.ssh_manager.unreachable_hosts:
         health_data["nodes"][node] = {
             "status": "unreachable",
             "issues": ["Node is unreachable via SSH"],
         }
-        health_data["alerts"].append({
-            "severity": "critical",
-            "node": node,
-            "message": f"Node {node} is unreachable"
-        })
+        health_data["alerts"].append({"severity": "critical", "node": node, "message": f"Node {node} is unreachable"})
 
     return health_data

cvs/monitors/cluster-mon/backend/app/api/config.py

@@ -7,8 +7,6 @@
 from typing import Dict, Any, List, Optional
 import yaml
 import os
-import signal
-import asyncio
 from pathlib import Path
 
 router = APIRouter()
@@ -75,6 +73,7 @@ async def update_configuration(config: ClusterConfigUpdate) -> Dict[str, Any]:
     Saves nodes to config/nodes.txt and updates cluster.yaml
     """
     import logging
+
     logger = logging.getLogger(__name__)
 
     logger.info("=" * 80)
@@ -148,6 +147,7 @@ async def update_configuration(config: ClusterConfigUpdate) -> Dict[str, Any]:
 
         # SECURITY: Store password in memory only, never persist to disk
         from app.main import app_state
+
         if config.auth_method == "password" and config.password:
             app_state.ssh_password = config.password
         else:
@@ -213,7 +213,9 @@ async def update_configuration(config: ClusterConfigUpdate) -> Dict[str, Any]:
         logger.info("=" * 80)
 
         password_note = ""
-        if config.auth_method == "password" or (config.use_jump_host and config.jump_host and config.jump_host.auth_method == "password"):
+        if config.auth_method == "password" or (
+            config.use_jump_host and config.jump_host and config.jump_host.auth_method == "password"
+        ):
             password_note = " Passwords are stored in memory only."
 
         return {

cvs/monitors/cluster-mon/backend/app/api/logs.py

@@ -29,7 +29,4 @@ async def get_dmesg_errors() -> Dict[str, Any]:
         return logs_data
 
     except Exception as e:
-        raise HTTPException(
-            status_code=500,
-            detail=f"Failed to collect logs: {str(e)}"
-        )
+        raise HTTPException(status_code=500, detail=f"Failed to collect logs: {str(e)}")

cvs/monitors/cluster-mon/backend/app/api/nodes.py

@@ -84,9 +84,9 @@ async def list_nodes() -> List[Dict[str, Any]]:
                             util = gpu_entry.get("GPU use (%)", 0)
                         total_util += float(util) if util else 0
 
-                node_info["avg_gpu_util"] = round(
-                    total_util / node_info["gpu_count"], 2
-                ) if node_info["gpu_count"] > 0 else 0
+                node_info["avg_gpu_util"] = (
+                    round(total_util / node_info["gpu_count"], 2) if node_info["gpu_count"] > 0 else 0
+                )
 
             # Handle dict format (old rocm-smi format)
             elif isinstance(node_util, dict) and 'error' not in node_util:
@@ -97,9 +97,9 @@ async def list_nodes() -> List[Dict[str, Any]]:
                         util = gpu_metrics.get("GPU use (%)", 0)
                         total_util += float(util) if util else 0
 
-                node_info["avg_gpu_util"] = round(
-                    total_util / node_info["gpu_count"], 2
-                ) if node_info["gpu_count"] > 0 else 0
+                node_info["avg_gpu_util"] = (
+                    round(total_util / node_info["gpu_count"], 2) if node_info["gpu_count"] > 0 else 0
+                )
 
         if node in temp_data:
             node_temp = temp_data[node]
@@ -114,7 +114,9 @@ async def list_nodes() -> List[Dict[str, Any]]:
                         temp_data_field = gpu_entry.get("temperature", {})
                         if isinstance(temp_data_field, dict):
                             # Try different sensor names
-                            temp = temp_data_field.get("hotspot", temp_data_field.get("edge", temp_data_field.get("junction", 0)))
+                            temp = temp_data_field.get(
+                                "hotspot", temp_data_field.get("edge", temp_data_field.get("junction", 0))
+                            )
                             if isinstance(temp, dict):
                                 temp = temp.get("value", 0)
                         else:
@@ -131,9 +133,12 @@ async def list_nodes() -> List[Dict[str, Any]]:
                 for gpu_temp in node_temp.values():
                     if isinstance(gpu_temp, dict):
                         # Try junction (hotspot) first, then edge, then memory
-                        temp = gpu_temp.get("Temperature (Sensor junction) (C)",
-                                 gpu_temp.get("Temperature (Sensor edge) (C)",
-                                   gpu_temp.get("Temperature (Sensor memory) (C)", 0)))
+                        temp = gpu_temp.get(
+                            "Temperature (Sensor junction) (C)",
+                            gpu_temp.get(
+                                "Temperature (Sensor edge) (C)", gpu_temp.get("Temperature (Sensor memory) (C)", 0)
+                            ),
+                        )
                         if temp:
                             try:
                                 total_temp += float(temp)
@@ -228,9 +233,12 @@ async def get_node_details(node_id: str) -> Dict[str, Any]:
 
             # Temperature - try junction (hotspot) first, then edge, then memory
             if gpu_id in temp_data and isinstance(temp_data[gpu_id], dict):
-                temp = temp_data[gpu_id].get("Temperature (Sensor junction) (C)",
-                         temp_data[gpu_id].get("Temperature (Sensor edge) (C)",
-                           temp_data[gpu_id].get("Temperature (Sensor memory) (C)", 0)))
+                temp = temp_data[gpu_id].get(
+                    "Temperature (Sensor junction) (C)",
+                    temp_data[gpu_id].get(
+                        "Temperature (Sensor edge) (C)", temp_data[gpu_id].get("Temperature (Sensor memory) (C)", 0)
+                    ),
+                )
                 try:
                     gpu_info["temperature_c"] = float(temp) if temp else 0.0
                 except (ValueError, TypeError):

cvs/monitors/cluster-mon/backend/app/api/packages.py

@@ -38,30 +38,25 @@ def install_package(request: PackageInstallRequest) -> Dict[str, Any]:
         # Get the appropriate installer
         if request.package.lower() == 'lldp':
             from app.installers.lldp_installer import LLDPInstaller
+
             installer = LLDPInstaller(ssh_manager)
         else:
-            raise HTTPException(
-                status_code=400,
-                detail=f"Unsupported package: {request.package}. Supported: lldp"
-            )
+            raise HTTPException(status_code=400, detail=f"Unsupported package: {request.package}. Supported: lldp")
 
         # Run installation (synchronous)
         result = installer.install_package()
 
         return {
             "success": True,
             "message": f"Installation complete: {result['successful']} successful, {result['failed']} failed",
-            **result
+            **result,
         }
 
     except HTTPException:
         raise
     except Exception as e:
         logger.error(f"Package installation failed: {str(e)}", exc_info=True)
-        raise HTTPException(
-            status_code=500,
-            detail=f"Package installation failed: {str(e)}"
-        )
+        raise HTTPException(status_code=500, detail=f"Package installation failed: {str(e)}")
 
 
 @router.get("/status/{package}")
@@ -88,20 +83,18 @@ def get_package_status(package: str) -> Dict[str, Any]:
                 "installed_nodes": [],
                 "not_installed_nodes": [],
                 "status_by_node": {},
-                "note": "SSH not configured yet. Save configuration first."
+                "note": "SSH not configured yet. Save configuration first.",
             }
 
         logger.info(f"Checking package status: {package}")
 
         # Get the appropriate installer
         if package.lower() == 'lldp':
             from app.installers.lldp_installer import LLDPInstaller
+
             installer = LLDPInstaller(ssh_manager)
         else:
-            raise HTTPException(
-                status_code=400,
-                detail=f"Unsupported package: {package}. Supported: lldp"
-            )
+            raise HTTPException(status_code=400, detail=f"Unsupported package: {package}. Supported: lldp")
 
         # Check installation status (synchronous)
         installed_map = installer.check_installation()
@@ -116,17 +109,14 @@ def get_package_status(package: str) -> Dict[str, Any]:
             "not_installed_count": len(not_installed_nodes),
             "installed_nodes": installed_nodes,
             "not_installed_nodes": not_installed_nodes,
-            "status_by_node": installed_map
+            "status_by_node": installed_map,
         }
 
     except HTTPException:
         raise
     except Exception as e:
         logger.error(f"Failed to check package status: {str(e)}", exc_info=True)
-        raise HTTPException(
-            status_code=500,
-            detail=f"Failed to check package status: {str(e)}"
-        )
+        raise HTTPException(status_code=500, detail=f"Failed to check package status: {str(e)}")
 
 
 @router.get("/list")
@@ -141,7 +131,7 @@ async def list_supported_packages() -> Dict[str, Any]:
                 "name": "LLDP Daemon",
                 "description": "Link Layer Discovery Protocol daemon for network topology discovery",
                 "package_name": "lldpd",
-                "check_command": "lldpcli"
+                "check_command": "lldpcli",
             },
             # Add more packages here as we implement them
             # {

cvs/monitors/cluster-mon/backend/app/api/restart.py

@@ -19,6 +19,7 @@ async def restart_backend() -> Dict[str, Any]:
     This triggers a graceful shutdown and restart by executing
     the backend in a new process.
     """
+
     async def do_restart():
         await asyncio.sleep(1)
         # Restart using os.execv to replace current process

cvs/monitors/cluster-mon/backend/app/api/ssh_keys.py

@@ -4,7 +4,6 @@
 
 from fastapi import APIRouter, UploadFile, File, HTTPException
 from typing import Dict, Any
-import os
 from pathlib import Path
 import logging
 
@@ -26,17 +25,15 @@ async def upload_ssh_key(file: UploadFile = File(...)) -> Dict[str, Any]:
         # Only allow common SSH key filenames for security
         allowed_names = ["id_rsa", "id_ed25519", "id_ecdsa", "cluster_id_ed25519", "known_hosts", "config"]
         if file.filename not in allowed_names:
-            raise HTTPException(
-                status_code=400,
-                detail=f"Invalid key filename. Allowed: {', '.join(allowed_names)}"
-            )
+            raise HTTPException(status_code=400, detail=f"Invalid key filename. Allowed: {', '.join(allowed_names)}")
 
         # Create .ssh directory if it doesn't exist
         ssh_dir = Path("/root/.ssh")
         ssh_dir.mkdir(mode=0o700, exist_ok=True)
 
         # Ensure directory has correct ownership
         import os
+
         try:
             os.chown(ssh_dir, 0, 0)  # root:root
             ssh_dir.chmod(0o700)
@@ -52,6 +49,7 @@ async def upload_ssh_key(file: UploadFile = File(...)) -> Dict[str, Any]:
 
         # Set proper permissions and ownership
         import os
+
         if file.filename in ["known_hosts", "config"]:
             key_path.chmod(0o644)
         else:
@@ -76,7 +74,7 @@ async def upload_ssh_key(file: UploadFile = File(...)) -> Dict[str, Any]:
             "message": f"SSH key '{file.filename}' uploaded successfully",
             "filename": file.filename,
             "size": len(content),
-            "path": str(key_path)
+            "path": str(key_path),
         }
 
     except Exception as e:
@@ -93,25 +91,21 @@ async def list_ssh_keys() -> Dict[str, Any]:
         ssh_dir = Path("/root/.ssh")
 
         if not ssh_dir.exists():
-            return {
-                "keys": [],
-                "message": "No SSH keys directory found"
-            }
+            return {"keys": [], "message": "No SSH keys directory found"}
 
         keys = []
         for key_file in ssh_dir.iterdir():
             if key_file.is_file():
                 stat = key_file.stat()
-                keys.append({
-                    "filename": key_file.name,
-                    "size": stat.st_size,
-                    "permissions": oct(stat.st_mode)[-3:],
-                })
+                keys.append(
+                    {
+                        "filename": key_file.name,
+                        "size": stat.st_size,
+                        "permissions": oct(stat.st_mode)[-3:],
+                    }
+                )
 
-        return {
-            "keys": keys,
-            "count": len(keys)
-        }
+        return {"keys": keys, "count": len(keys)}
 
     except Exception as e:
         logger.error(f"Failed to list SSH keys: {e}")
@@ -137,10 +131,7 @@ async def delete_ssh_key(filename: str) -> Dict[str, Any]:
         key_path.unlink()
         logger.info(f"SSH key deleted: {filename}")
 
-        return {
-            "success": True,
-            "message": f"SSH key '{filename}' deleted successfully"
-        }
+        return {"success": True, "message": f"SSH key '{filename}' deleted successfully"}
 
     except HTTPException:
         raise