Production agents must handle failures: tool timeouts, API errors, LLM inconsistencies. Implement retries with exponential backoff, fallback tools, and graceful degradation to maintain service quality.
Retry with Exponential Backoff
Code Snippet
import time
from functools import wraps
def retry_with_backoff(max_retries=3, base_delay=1):
def decorator(func):
@wraps(func)
def wrapper(*args, **kwargs):
for attempt in range(max_retries):
try:
return func(*args, **kwargs)
except Exception as e:
if attempt == max_retries - 1:
raise
delay = base_delay * (2 ** attempt)
print(f"Attempt {attempt+1} failed. Retrying in {delay}s...")
time.sleep(delay)
return wrapper
return decorator
@retry_with_backoff(max_retries=3)
def call_external_api():
# API call that might fail
pass
Fallback Tools
Code Snippet
# Primary tool
search_tool = StructuredTool.from_function(func=search_api)
# Fallback: cached results
fallback_search = StructuredTool.from_function(func=search_cache)
def robust_search(query):
try:
return search_tool.invoke({"query": query})
except Exception as e:
print(f"Search API failed: {e}. Falling back to cache.")
return fallback_search.invoke({"query": query})
Graceful Degradation
Code Snippet
# Try detailed information
try:
result = get_detailed_weather(city)
except:
# Fallback to basic info
result = get_basic_weather(city)
return result + " (Note: detailed forecast unavailable)"
Conclusion
Error handling separates fragile demos from production systems. Retries, fallbacks, and graceful degradation keep agents running when components fail. Understanding failure modes and mitigation strategies is essential. Next: deploying agents at scale with proper monitoring and observability.