Cascade / cascade /diagnostics /execution_monitor.py
tostido's picture
feat(diagnostics): Add code bug exposure system v0.6.0
1754798
"""
CASCADE Execution Monitor - Monitor live code execution.
Wraps Python execution similar to how cascade.observe wraps processes.
Captures execution flow, exceptions, and anomalies in real-time.
"""
import sys
import time
import threading
import traceback
import functools
from typing import Any, Dict, List, Optional, Callable, Set, Tuple
from dataclasses import dataclass, field
from contextlib import contextmanager
import queue
from cascade.core.event import Event, CausationLink
from cascade.core.graph import CausationGraph
from cascade.core.adapter import SymbioticAdapter
@dataclass
class ExecutionFrame:
"""A frame in the execution trace."""
frame_id: str
function_name: str
file_path: str
line_number: int
local_vars: Dict[str, str] # Sanitized string representations
timestamp: float
duration_ms: Optional[float] = None
exception: Optional[str] = None
@dataclass
class Anomaly:
"""An execution anomaly detected during monitoring."""
anomaly_type: str
description: str
severity: str
frame_id: str
timestamp: float
context: Dict[str, Any] = field(default_factory=dict)
class ExecutionMonitor:
"""
Monitor live code execution and capture anomalies.
Uses sys.settrace to capture every function call, return, and exception.
Integrates with cascade-lattice's causation graph for tracing.
Usage:
monitor = ExecutionMonitor()
with monitor.monitoring():
# Your code here
result = my_function()
# After execution:
monitor.get_anomalies()
monitor.get_execution_trace()
"""
def __init__(self,
capture_locals: bool = True,
max_depth: int = 100,
exclude_modules: Optional[Set[str]] = None):
self.capture_locals = capture_locals
self.max_depth = max_depth
self.exclude_modules = exclude_modules or {
'cascade', 'threading', 'queue', 'logging',
'importlib', '_frozen_importlib', 'posixpath', 'genericpath',
}
# Execution tracking
self.frames: List[ExecutionFrame] = []
self.anomalies: List[Anomaly] = []
self.call_stack: List[str] = []
# Causation graph
self.graph = CausationGraph()
self.adapter = SymbioticAdapter()
# Thresholds for anomaly detection
self.slow_threshold_ms = 100
self.deep_recursion_threshold = 50
# State
self._monitoring = False
self._lock = threading.RLock()
self._frame_counter = 0
self._function_times: Dict[str, List[float]] = {}
self._prev_trace = None
@contextmanager
def monitoring(self):
"""Context manager for monitoring execution."""
self.start()
try:
yield self
finally:
self.stop()
def start(self):
"""Start execution monitoring."""
if self._monitoring:
return
self._monitoring = True
self._prev_trace = sys.gettrace()
sys.settrace(self._trace_calls)
threading.settrace(self._trace_calls)
def stop(self):
"""Stop execution monitoring."""
if not self._monitoring:
return
self._monitoring = False
sys.settrace(self._prev_trace)
threading.settrace(None)
# Analyze collected data for anomalies
self._analyze_for_anomalies()
def _trace_calls(self, frame, event, arg):
"""Trace function for sys.settrace."""
if not self._monitoring:
return None
# Filter out excluded modules
code = frame.f_code
module = frame.f_globals.get('__name__', '')
if any(module.startswith(exc) for exc in self.exclude_modules):
return None
# Check depth
if len(self.call_stack) > self.max_depth:
return None
try:
if event == 'call':
self._handle_call(frame, code)
elif event == 'return':
self._handle_return(frame, code, arg)
elif event == 'exception':
self._handle_exception(frame, code, arg)
except Exception:
# Don't let tracing errors affect the program
pass
return self._trace_calls
def _handle_call(self, frame, code):
"""Handle function call event."""
with self._lock:
self._frame_counter += 1
frame_id = f"frame_{self._frame_counter}"
# Capture local variables
local_vars = {}
if self.capture_locals:
for name, value in list(frame.f_locals.items())[:20]:
local_vars[name] = self._sanitize_value(value)
exec_frame = ExecutionFrame(
frame_id=frame_id,
function_name=code.co_name,
file_path=code.co_filename,
line_number=frame.f_lineno,
local_vars=local_vars,
timestamp=time.time(),
)
self.frames.append(exec_frame)
# Create causation link to caller
if self.call_stack:
caller_id = self.call_stack[-1]
link = CausationLink(
from_event=caller_id,
to_event=frame_id,
causation_type="call",
strength=1.0,
explanation=f"Called {code.co_name}"
)
self.graph.add_link(link)
self.call_stack.append(frame_id)
# Track function timing
func_key = f"{code.co_filename}:{code.co_name}"
if func_key not in self._function_times:
self._function_times[func_key] = []
def _handle_return(self, frame, code, return_value):
"""Handle function return event."""
with self._lock:
if not self.call_stack:
return
frame_id = self.call_stack.pop()
# Find the frame and update duration
for exec_frame in reversed(self.frames):
if exec_frame.frame_id == frame_id:
exec_frame.duration_ms = (time.time() - exec_frame.timestamp) * 1000
# Track timing
func_key = f"{code.co_filename}:{code.co_name}"
if func_key in self._function_times:
self._function_times[func_key].append(exec_frame.duration_ms)
break
def _handle_exception(self, frame, code, arg):
"""Handle exception event."""
exc_type, exc_value, exc_tb = arg
with self._lock:
frame_id = self.call_stack[-1] if self.call_stack else f"frame_{self._frame_counter}"
# Update frame with exception
for exec_frame in reversed(self.frames):
if exec_frame.frame_id == frame_id:
exec_frame.exception = f"{exc_type.__name__}: {exc_value}"
break
# Record as anomaly
self.anomalies.append(Anomaly(
anomaly_type="exception",
description=f"{exc_type.__name__}: {exc_value}",
severity="error",
frame_id=frame_id,
timestamp=time.time(),
context={
"exception_type": exc_type.__name__,
"exception_message": str(exc_value),
"function": code.co_name,
"file": code.co_filename,
"line": frame.f_lineno,
},
))
def _analyze_for_anomalies(self):
"""Analyze collected data for additional anomalies."""
# Check for slow functions
for func_key, times in self._function_times.items():
if times:
avg_time = sum(times) / len(times)
max_time = max(times)
if max_time > self.slow_threshold_ms:
self.anomalies.append(Anomaly(
anomaly_type="slow_execution",
description=f"Slow function: {func_key} (max: {max_time:.1f}ms, avg: {avg_time:.1f}ms)",
severity="warning",
frame_id="",
timestamp=time.time(),
context={
"function": func_key,
"max_time_ms": max_time,
"avg_time_ms": avg_time,
"call_count": len(times),
},
))
# Check for deep recursion
max_depth = max((len(self.call_stack) for f in self.frames), default=0)
if max_depth > self.deep_recursion_threshold:
self.anomalies.append(Anomaly(
anomaly_type="deep_recursion",
description=f"Deep call stack detected: {max_depth} frames",
severity="warning",
frame_id="",
timestamp=time.time(),
context={"max_depth": max_depth},
))
# Check for repeated exceptions
exception_counts: Dict[str, int] = {}
for anomaly in self.anomalies:
if anomaly.anomaly_type == "exception":
exc_type = anomaly.context.get("exception_type", "unknown")
exception_counts[exc_type] = exception_counts.get(exc_type, 0) + 1
for exc_type, count in exception_counts.items():
if count > 3:
self.anomalies.append(Anomaly(
anomaly_type="repeated_exception",
description=f"{exc_type} occurred {count} times",
severity="error",
frame_id="",
timestamp=time.time(),
context={"exception_type": exc_type, "count": count},
))
def _sanitize_value(self, value: Any, max_len: int = 100) -> str:
"""Convert value to safe string representation."""
try:
if value is None:
return "None"
# Numpy arrays
if hasattr(value, 'shape'):
return f"<array {value.shape}>"
# Tensors
if hasattr(value, 'size') and callable(value.size):
return f"<tensor {value.size()}>"
# Large collections
if isinstance(value, (list, dict, set)):
s = str(value)
if len(s) > max_len:
return s[:max_len] + "..."
return s
# Strings
if isinstance(value, str):
if len(value) > max_len:
return value[:max_len] + "..."
return repr(value)
# Primitives
if isinstance(value, (int, float, bool)):
return str(value)
# Fallback
return f"<{type(value).__name__}>"
except Exception:
return "<error>"
# =========================================================================
# QUERIES
# =========================================================================
def get_anomalies(self, severity: Optional[str] = None) -> List[Anomaly]:
"""Get detected anomalies, optionally filtered by severity."""
if severity:
return [a for a in self.anomalies if a.severity == severity]
return list(self.anomalies)
def get_execution_trace(self) -> List[ExecutionFrame]:
"""Get the execution trace."""
return list(self.frames)
def get_call_graph(self) -> Dict[str, List[str]]:
"""Get the call graph as adjacency list."""
graph: Dict[str, List[str]] = {}
for link in self.graph._links:
if link.from_event not in graph:
graph[link.from_event] = []
graph[link.from_event].append(link.to_event)
return graph
def get_hotspots(self, top_n: int = 10) -> List[Tuple[str, float, int]]:
"""Get the hottest functions by total time spent."""
totals: Dict[str, Tuple[float, int]] = {}
for func_key, times in self._function_times.items():
if times:
totals[func_key] = (sum(times), len(times))
sorted_funcs = sorted(totals.items(), key=lambda x: x[1][0], reverse=True)
return [(func, total, count) for func, (total, count) in sorted_funcs[:top_n]]
# =========================================================================
# REPORTING
# =========================================================================
def get_summary(self) -> Dict[str, Any]:
"""Get monitoring summary."""
return {
"total_frames": len(self.frames),
"total_anomalies": len(self.anomalies),
"anomalies_by_type": self._count_by_key(self.anomalies, lambda a: a.anomaly_type),
"anomalies_by_severity": self._count_by_key(self.anomalies, lambda a: a.severity),
"functions_traced": len(self._function_times),
}
def _count_by_key(self, items, key_func) -> Dict[str, int]:
"""Count items by key function."""
counts: Dict[str, int] = {}
for item in items:
key = key_func(item)
counts[key] = counts.get(key, 0) + 1
return counts
def get_report(self) -> str:
"""Generate a human-readable report."""
lines = [
"EXECUTION MONITORING REPORT",
"=" * 60,
f"Frames captured: {len(self.frames)}",
f"Functions traced: {len(self._function_times)}",
f"Anomalies detected: {len(self.anomalies)}",
"",
]
# Anomalies by severity
if self.anomalies:
lines.append("ANOMALIES")
lines.append("-" * 40)
severity_icons = {"critical": "🔴", "error": "❌", "warning": "⚠️", "info": "ℹ️"}
for anomaly in sorted(self.anomalies, key=lambda a:
["critical", "error", "warning", "info"].index(a.severity)
if a.severity in ["critical", "error", "warning", "info"] else 99):
icon = severity_icons.get(anomaly.severity, "•")
lines.append(f" {icon} [{anomaly.anomaly_type}] {anomaly.description}")
lines.append("")
# Hotspots
hotspots = self.get_hotspots(5)
if hotspots:
lines.append("PERFORMANCE HOTSPOTS")
lines.append("-" * 40)
for func, total_ms, count in hotspots:
avg_ms = total_ms / count if count else 0
lines.append(f" {func}")
lines.append(f" Total: {total_ms:.1f}ms | Calls: {count} | Avg: {avg_ms:.1f}ms")
lines.append("")
return "\n".join(lines)
# =============================================================================
# CONVENIENCE DECORATORS
# =============================================================================
def monitor(func: Callable = None, **kwargs) -> Callable:
"""
Decorator to monitor a function's execution.
Usage:
@monitor
def my_function():
...
# Access monitoring results
my_function._monitor_results
"""
def decorator(fn: Callable) -> Callable:
@functools.wraps(fn)
def wrapper(*args, **call_kwargs):
monitor = ExecutionMonitor(**kwargs)
with monitor.monitoring():
result = fn(*args, **call_kwargs)
# Attach results to function
wrapper._monitor_results = {
"anomalies": monitor.get_anomalies(),
"summary": monitor.get_summary(),
"report": monitor.get_report(),
}
return result
wrapper._monitor_results = None
return wrapper
if func is not None:
return decorator(func)
return decorator