Add ComfyUI output capture for crash debugging

- Add background thread to read ComfyUI stdout in real-time
- Store last 200 lines in circular buffer
- Echo output to RunPod logs with [ComfyUI] prefix
- Include last 100 lines in error responses for debugging
- Add comfyui_output field to error responses

This will help diagnose why ComfyUI crashes during generation.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

handler.py

@@ -20,6 +20,8 @@ import base64
 import uuid
 import subprocess
 import signal
+import threading
+from collections import deque
 import requests
 from pathlib import Path
 import runpod
@@ -42,8 +44,36 @@ NODE_STEPS = "150"
 NODE_SPLIT_STEP = "151"
 NODE_SAVE_VIDEO = "117"
 
-# Global ComfyUI process
+# Global ComfyUI process and output capture
 comfyui_process = None
+comfyui_output_buffer = deque(maxlen=200)  # Keep last 200 lines
+comfyui_output_lock = threading.Lock()
+comfyui_reader_thread = None
+
+
+def _read_comfyui_output():
+    """Background thread to read ComfyUI stdout and store in buffer."""
+    global comfyui_process, comfyui_output_buffer
+    while comfyui_process is not None:
+        try:
+            line = comfyui_process.stdout.readline()
+            if line:
+                decoded = line.decode('utf-8', errors='replace').rstrip()
+                with comfyui_output_lock:
+                    comfyui_output_buffer.append(decoded)
+                print(f"[ComfyUI] {decoded}")  # Echo to RunPod logs
+            elif comfyui_process.poll() is not None:
+                # Process ended
+                break
+        except Exception:
+            break
+
+
+def get_comfyui_output(last_n: int = 50) -> list:
+    """Get the last N lines of ComfyUI output."""
+    with comfyui_output_lock:
+        lines = list(comfyui_output_buffer)
+    return lines[-last_n:] if len(lines) > last_n else lines
 
 
 class JobLogger:
@@ -68,7 +98,7 @@ class JobLogger:
 
 def start_comfyui(logger: JobLogger = None):
     """Start ComfyUI server if not already running."""
-    global comfyui_process
+    global comfyui_process, comfyui_reader_thread, comfyui_output_buffer
 
     def log(msg):
         if logger:
@@ -80,6 +110,10 @@ def start_comfyui(logger: JobLogger = None):
         log("ComfyUI server already running")
         return True
 
+    # Clear output buffer for fresh start
+    with comfyui_output_lock:
+        comfyui_output_buffer.clear()
+
     log("Starting ComfyUI server...")
 
     comfyui_process = subprocess.Popen(
@@ -95,6 +129,10 @@ def start_comfyui(logger: JobLogger = None):
         preexec_fn=os.setsid if hasattr(os, 'setsid') else None
     )
 
+    # Start background thread to capture output
+    comfyui_reader_thread = threading.Thread(target=_read_comfyui_output, daemon=True)
+    comfyui_reader_thread.start()
+
     # Wait for server to be ready
     start_time = time.time()
     last_status_time = start_time
@@ -600,13 +638,21 @@ def handler(job: dict) -> dict:
 
     except TimeoutError as e:
         logger.log(f"ERROR: Timeout - {str(e)}")
-        return {"error": str(e), "status": "timeout", "logs": logger.get_logs()}
+        comfyui_logs = get_comfyui_output(100)
+        logger.log(f"ComfyUI output (last 100 lines):")
+        for line in comfyui_logs:
+            logger.log(f"  {line}")
+        return {"error": str(e), "status": "timeout", "logs": logger.get_logs(), "comfyui_output": comfyui_logs}
     except Exception as e:
         import traceback
         tb = traceback.format_exc()
         logger.log(f"ERROR: {str(e)}")
         logger.log(f"Traceback:\n{tb}")
-        return {"error": str(e), "status": "error", "logs": logger.get_logs()}
+        comfyui_logs = get_comfyui_output(100)
+        logger.log(f"ComfyUI output (last 100 lines):")
+        for line in comfyui_logs:
+            logger.log(f"  {line}")
+        return {"error": str(e), "status": "error", "logs": logger.get_logs(), "comfyui_output": comfyui_logs}
 
 
 # RunPod serverless entry point