wanderlust.ai / src /wanderlust_ai /core /fallback_chains.py
BlakeL's picture
Upload 115 files
3f9f85b verified
"""
Fallback Chains for API Failures
This module implements intelligent fallback mechanisms that provide alternative
data sources and strategies when primary APIs fail.
"""
import asyncio
import time
import math
from datetime import datetime, timedelta
from typing import Dict, List, Optional, Any, Callable, Union, Tuple
from dataclasses import dataclass, field
from enum import Enum
import logging
from functools import wraps
from .error_categorization import CategorizedError, ErrorCategory, ErrorType
from .retry_strategies import RetryConfig, RetryStrategy
class FallbackStrategy(str, Enum):
"""Different fallback strategies."""
SEQUENTIAL = "sequential" # Try fallbacks in order
PARALLEL = "parallel" # Try all fallbacks simultaneously
INTELLIGENT = "intelligent" # Choose fallback based on error type
CACHED_ONLY = "cached_only" # Only use cached data
PARTIAL_RESULTS = "partial_results" # Return partial results if available
class DataSource(str, Enum):
"""Available data sources."""
TAVILY = "tavily"
SERPAPI = "serpapi"
CACHE = "cache"
STATIC_DATA = "static_data"
USER_PREFERENCES = "user_preferences"
HISTORICAL_DATA = "historical_data"
@dataclass
class FallbackSource:
"""Definition of a fallback data source."""
name: str
source_type: DataSource
operation: Callable
priority: int = 1 # Lower number = higher priority
timeout_seconds: float = 30.0
success_rate: float = 1.0 # Tracked dynamically
last_success: Optional[datetime] = None
last_failure: Optional[datetime] = None
failure_count: int = 0
success_count: int = 0
is_available: bool = True
metadata: Dict[str, Any] = field(default_factory=dict)
def get_reliability_score(self) -> float:
"""Calculate reliability score for this source."""
total_attempts = self.success_count + self.failure_count
if total_attempts == 0:
return 1.0
base_success_rate = self.success_count / total_attempts
# Penalize recent failures
recency_penalty = 0.0
if self.last_failure and self.last_success:
if self.last_failure > self.last_success:
time_since_failure = (datetime.now() - self.last_failure).total_seconds()
# Reduce penalty over time (exponential decay)
recency_penalty = 0.3 * math.exp(-time_since_failure / 3600) # 1 hour half-life
return max(0.0, base_success_rate - recency_penalty)
@dataclass
class FallbackResult:
"""Result of a fallback operation."""
success: bool
data: Any = None
source_name: str = ""
source_type: DataSource = DataSource.CACHE
execution_time_seconds: float = 0.0
error: Optional[Exception] = None
metadata: Dict[str, Any] = field(default_factory=dict)
fallback_chain: List[str] = field(default_factory=list)
@dataclass
class FallbackChain:
"""Definition of a fallback chain for a specific operation."""
operation_name: str
strategy: FallbackStrategy
sources: List[FallbackSource]
max_execution_time: float = 60.0
min_success_rate: float = 0.5
cache_fallback_enabled: bool = True
partial_results_enabled: bool = True
metadata: Dict[str, Any] = field(default_factory=dict)
class FallbackChainManager:
"""
Manager for fallback chains that coordinates multiple data sources
and implements intelligent fallback strategies.
"""
def __init__(self):
self.logger = logging.getLogger(__name__)
self._chains: Dict[str, FallbackChain] = {}
self._source_statistics: Dict[str, Dict[str, Any]] = {}
self._cache_store: Dict[str, Any] = {}
self._cache_timestamps: Dict[str, datetime] = {}
self._cache_ttl_seconds: Dict[str, float] = {}
def register_chain(self, chain: FallbackChain):
"""Register a fallback chain."""
self._chains[chain.operation_name] = chain
# Initialize source statistics
for source in chain.sources:
if source.name not in self._source_statistics:
self._source_statistics[source.name] = {
"total_attempts": 0,
"successful_attempts": 0,
"failed_attempts": 0,
"avg_execution_time": 0.0,
"last_updated": datetime.now()
}
def get_or_create_chain(self, operation_name: str,
sources: List[FallbackSource],
strategy: FallbackStrategy = FallbackStrategy.SEQUENTIAL) -> FallbackChain:
"""Get existing chain or create a new one."""
if operation_name not in self._chains:
chain = FallbackChain(
operation_name=operation_name,
strategy=strategy,
sources=sources
)
self.register_chain(chain)
return self._chains[operation_name]
def cache_result(self, key: str, data: Any, ttl_seconds: float = 3600):
"""Cache a result for fallback use."""
self._cache_store[key] = data
self._cache_timestamps[key] = datetime.now()
self._cache_ttl_seconds[key] = ttl_seconds
def get_cached_result(self, key: str) -> Optional[Any]:
"""Get cached result if available and not expired."""
if key not in self._cache_store:
return None
# Check if expired
if key in self._cache_timestamps and key in self._cache_ttl_seconds:
age_seconds = (datetime.now() - self._cache_timestamps[key]).total_seconds()
if age_seconds > self._cache_ttl_seconds[key]:
# Remove expired cache
del self._cache_store[key]
del self._cache_timestamps[key]
del self._cache_ttl_seconds[key]
return None
return self._cache_store[key]
async def execute_fallback_chain(self, operation_name: str,
context: Optional[Dict[str, Any]] = None,
cache_key: Optional[str] = None) -> FallbackResult:
"""
Execute a fallback chain for an operation.
Args:
operation_name: Name of the operation
context: Context for the operation
cache_key: Key for caching results
Returns:
FallbackResult with data from the best available source
"""
if operation_name not in self._chains:
return FallbackResult(
success=False,
error=Exception(f"No fallback chain registered for {operation_name}")
)
chain = self._chains[operation_name]
context = context or {}
# Try cache first if enabled
if chain.cache_fallback_enabled and cache_key:
cached_data = self.get_cached_result(cache_key)
if cached_data:
self.logger.info(f"Using cached data for {operation_name}")
return FallbackResult(
success=True,
data=cached_data,
source_name="cache",
source_type=DataSource.CACHE,
fallback_chain=["cache"]
)
# Execute based on strategy
if chain.strategy == FallbackStrategy.SEQUENTIAL:
return await self._execute_sequential_fallback(chain, context, cache_key)
elif chain.strategy == FallbackStrategy.PARALLEL:
return await self._execute_parallel_fallback(chain, context, cache_key)
elif chain.strategy == FallbackStrategy.INTELLIGENT:
return await self._execute_intelligent_fallback(chain, context, cache_key)
elif chain.strategy == FallbackStrategy.CACHED_ONLY:
return await self._execute_cached_only_fallback(chain, context, cache_key)
else:
return await self._execute_sequential_fallback(chain, context, cache_key)
async def _execute_sequential_fallback(self, chain: FallbackChain,
context: Dict[str, Any],
cache_key: Optional[str]) -> FallbackResult:
"""Execute fallback sources sequentially."""
fallback_chain = []
partial_results = []
# Sort sources by priority and reliability
sorted_sources = sorted(chain.sources, key=lambda s: (s.priority, -s.get_reliability_score()))
for source in sorted_sources:
if not source.is_available:
continue
fallback_chain.append(source.name)
start_time = time.time()
try:
# Execute source with timeout
if asyncio.iscoroutinefunction(source.operation):
data = await asyncio.wait_for(
source.operation(**context),
timeout=source.timeout_seconds
)
else:
# For sync functions, run in thread
loop = asyncio.get_event_loop()
data = await loop.run_in_executor(
None,
lambda: source.operation(**context)
)
execution_time = time.time() - start_time
# Record success
self._record_source_result(source, True, execution_time)
# Cache result if key provided
if cache_key:
self.cache_result(cache_key, data)
return FallbackResult(
success=True,
data=data,
source_name=source.name,
source_type=source.source_type,
execution_time_seconds=execution_time,
fallback_chain=fallback_chain
)
except Exception as error:
execution_time = time.time() - start_time
self._record_source_result(source, False, execution_time)
# Store partial results if available
if hasattr(error, 'partial_results') and error.partial_results:
partial_results.extend(error.partial_results)
self.logger.warning(f"Fallback source {source.name} failed: {error}")
continue
# All sources failed - return partial results if available
if chain.partial_results_enabled and partial_results:
return FallbackResult(
success=True,
data=partial_results,
source_name="partial_results",
source_type=DataSource.CACHE,
fallback_chain=fallback_chain,
metadata={"partial_results": True}
)
return FallbackResult(
success=False,
error=Exception(f"All fallback sources failed for {chain.operation_name}"),
fallback_chain=fallback_chain
)
async def _execute_parallel_fallback(self, chain: FallbackChain,
context: Dict[str, Any],
cache_key: Optional[str]) -> FallbackResult:
"""Execute fallback sources in parallel."""
fallback_chain = []
available_sources = [s for s in chain.sources if s.is_available]
if not available_sources:
return FallbackResult(
success=False,
error=Exception("No available fallback sources"),
fallback_chain=[]
)
# Create tasks for all sources
tasks = []
for source in available_sources:
fallback_chain.append(source.name)
task = asyncio.create_task(self._execute_single_source(source, context))
tasks.append((source, task))
try:
# Wait for first successful result
for source, task in tasks:
try:
result = await asyncio.wait_for(task, timeout=source.timeout_seconds)
if result.success:
# Cancel other tasks
for _, other_task in tasks:
if other_task != task and not other_task.done():
other_task.cancel()
# Cache result if key provided
if cache_key:
self.cache_result(cache_key, result.data)
return FallbackResult(
success=True,
data=result.data,
source_name=source.name,
source_type=source.source_type,
execution_time_seconds=result.execution_time_seconds,
fallback_chain=fallback_chain
)
except asyncio.TimeoutError:
self._record_source_result(source, False, source.timeout_seconds)
continue
except Exception as error:
self._record_source_result(source, False, 0.0)
continue
# No sources succeeded
return FallbackResult(
success=False,
error=Exception("All parallel fallback sources failed"),
fallback_chain=fallback_chain
)
finally:
# Ensure all tasks are cleaned up
for _, task in tasks:
if not task.done():
task.cancel()
async def _execute_intelligent_fallback(self, chain: FallbackChain,
context: Dict[str, Any],
cache_key: Optional[str]) -> FallbackResult:
"""Execute intelligent fallback based on error analysis."""
# For now, use sequential with intelligent source ordering
# In a more sophisticated implementation, this could analyze
# the specific error type and choose the most appropriate source
# Sort by reliability score and priority
sorted_sources = sorted(chain.sources, key=lambda s: (
s.priority,
-s.get_reliability_score(),
-s.success_rate
))
# Update chain sources order
chain.sources = sorted_sources
return await self._execute_sequential_fallback(chain, context, cache_key)
async def _execute_cached_only_fallback(self, chain: FallbackChain,
context: Dict[str, Any],
cache_key: Optional[str]) -> FallbackResult:
"""Execute cached-only fallback."""
if cache_key:
cached_data = self.get_cached_result(cache_key)
if cached_data:
return FallbackResult(
success=True,
data=cached_data,
source_name="cache",
source_type=DataSource.CACHE,
fallback_chain=["cache"]
)
return FallbackResult(
success=False,
error=Exception("No cached data available"),
fallback_chain=[]
)
async def _execute_single_source(self, source: FallbackSource,
context: Dict[str, Any]) -> FallbackResult:
"""Execute a single fallback source."""
start_time = time.time()
try:
if asyncio.iscoroutinefunction(source.operation):
data = await source.operation(**context)
else:
loop = asyncio.get_event_loop()
data = await loop.run_in_executor(None, lambda: source.operation(**context))
execution_time = time.time() - start_time
self._record_source_result(source, True, execution_time)
return FallbackResult(
success=True,
data=data,
source_name=source.name,
source_type=source.source_type,
execution_time_seconds=execution_time
)
except Exception as error:
execution_time = time.time() - start_time
self._record_source_result(source, False, execution_time)
return FallbackResult(
success=False,
error=error,
source_name=source.name,
source_type=source.source_type,
execution_time_seconds=execution_time
)
def _record_source_result(self, source: FallbackSource, success: bool, execution_time: float):
"""Record the result of a source execution."""
if success:
source.success_count += 1
source.last_success = datetime.now()
else:
source.failure_count += 1
source.last_failure = datetime.now()
# Update global statistics
if source.name not in self._source_statistics:
self._source_statistics[source.name] = {
"total_attempts": 0,
"successful_attempts": 0,
"failed_attempts": 0,
"avg_execution_time": 0.0,
"last_updated": datetime.now()
}
stats = self._source_statistics[source.name]
stats["total_attempts"] += 1
if success:
stats["successful_attempts"] += 1
else:
stats["failed_attempts"] += 1
# Update average execution time
total_time = stats["avg_execution_time"] * (stats["total_attempts"] - 1)
stats["avg_execution_time"] = (total_time + execution_time) / stats["total_attempts"]
stats["last_updated"] = datetime.now()
# Update source availability based on failure rate
total_attempts = source.success_count + source.failure_count
if total_attempts >= 5: # Only after sufficient data
failure_rate = source.failure_count / total_attempts
source.is_available = failure_rate < 0.8 # Mark unavailable if >80% failure rate
def get_source_statistics(self, source_name: Optional[str] = None) -> Dict[str, Any]:
"""Get statistics for fallback sources."""
if source_name:
return self._source_statistics.get(source_name, {})
return self._source_statistics
def get_chain_statistics(self, operation_name: Optional[str] = None) -> Dict[str, Any]:
"""Get statistics for fallback chains."""
if operation_name and operation_name in self._chains:
chain = self._chains[operation_name]
return {
"operation_name": chain.operation_name,
"strategy": chain.strategy.value,
"sources": [
{
"name": s.name,
"source_type": s.source_type.value,
"priority": s.priority,
"success_rate": s.get_reliability_score(),
"is_available": s.is_available,
"success_count": s.success_count,
"failure_count": s.failure_count
}
for s in chain.sources
]
}
else:
return {
"chains": list(self._chains.keys()),
"total_chains": len(self._chains)
}
# Convenience functions for creating common fallback chains
def create_api_fallback_chain(operation_name: str, primary_api: Callable,
fallback_apis: List[Callable],
cache_fallback: bool = True) -> FallbackChain:
"""Create a fallback chain for API operations."""
sources = []
# Primary API
sources.append(FallbackSource(
name="primary_api",
source_type=DataSource.TAVILY,
operation=primary_api,
priority=1,
timeout_seconds=30.0
))
# Fallback APIs
for i, api in enumerate(fallback_apis):
source_type = DataSource.SERPAPI if i == 0 else DataSource.STATIC_DATA
sources.append(FallbackSource(
name=f"fallback_api_{i+1}",
source_type=source_type,
operation=api,
priority=i+2,
timeout_seconds=30.0
))
# Cache fallback
if cache_fallback:
sources.append(FallbackSource(
name="cache",
source_type=DataSource.CACHE,
operation=lambda **kwargs: None, # Will be handled by cache logic
priority=999,
timeout_seconds=1.0
))
return FallbackChain(
operation_name=operation_name,
strategy=FallbackStrategy.SEQUENTIAL,
sources=sources,
cache_fallback_enabled=cache_fallback
)
def create_search_fallback_chain(operation_name: str,
tavily_search: Callable,
serpapi_search: Callable,
cached_search: Callable) -> FallbackChain:
"""Create a fallback chain specifically for search operations."""
sources = [
FallbackSource(
name="tavily_search",
source_type=DataSource.TAVILY,
operation=tavily_search,
priority=1,
timeout_seconds=15.0
),
FallbackSource(
name="serpapi_search",
source_type=DataSource.SERPAPI,
operation=serpapi_search,
priority=2,
timeout_seconds=20.0
),
FallbackSource(
name="cached_search",
source_type=DataSource.CACHE,
operation=cached_search,
priority=3,
timeout_seconds=5.0
)
]
return FallbackChain(
operation_name=operation_name,
strategy=FallbackStrategy.INTELLIGENT,
sources=sources,
cache_fallback_enabled=True,
partial_results_enabled=True
)
# Global fallback chain manager
_global_fallback_manager: Optional[FallbackChainManager] = None
def get_global_fallback_manager() -> FallbackChainManager:
"""Get the global fallback chain manager instance."""
global _global_fallback_manager
if _global_fallback_manager is None:
_global_fallback_manager = FallbackChainManager()
return _global_fallback_manager
# Decorator for fallback chains
def with_fallback_chain(operation_name: str, chain: Optional[FallbackChain] = None):
"""Decorator to add fallback chain to a function."""
def decorator(func: Callable) -> Callable:
@wraps(func)
async def async_wrapper(*args, **kwargs):
manager = get_global_fallback_manager()
# If no chain provided, try to get existing one
if chain is None:
if operation_name not in manager._chains:
# If no chain exists, just execute the function normally
return await func(*args, **kwargs)
fallback_chain = manager._chains[operation_name]
else:
fallback_chain = chain
# Create cache key from function arguments
cache_key = f"{operation_name}:{hash(str(args) + str(sorted(kwargs.items())))}"
# Execute fallback chain
context = {"args": args, "kwargs": kwargs, "function": func}
result = await manager.execute_fallback_chain(
operation_name, context, cache_key
)
if result.success:
return result.data
else:
raise result.error
@wraps(func)
def sync_wrapper(*args, **kwargs):
# For sync functions, run in event loop
async def async_func():
return await async_wrapper(*args, **kwargs)
try:
loop = asyncio.get_event_loop()
return loop.run_until_complete(async_func())
except RuntimeError:
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
try:
return loop.run_until_complete(async_func())
finally:
loop.close()
if asyncio.iscoroutinefunction(func):
return async_wrapper
else:
return sync_wrapper
return decorator