Update app.py
Browse files
app.py
CHANGED
|
@@ -16,28 +16,23 @@ from agentic_reliability_framework.core.governance.risk_engine import RiskEngine
|
|
| 16 |
from agentic_reliability_framework.runtime.memory import create_faiss_index, RAGGraphMemory
|
| 17 |
from agentic_reliability_framework.runtime.memory.constants import MemoryConstants
|
| 18 |
|
| 19 |
-
# Additional imports for
|
| 20 |
-
from agentic_reliability_framework.core.governance.governance_loop import GovernanceLoop
|
| 21 |
from agentic_reliability_framework.core.governance.policy_engine import PolicyEngine
|
| 22 |
from agentic_reliability_framework.core.governance.cost_estimator import CostEstimator
|
| 23 |
from agentic_reliability_framework.core.governance.intents import (
|
| 24 |
DeployConfigurationIntent,
|
| 25 |
Environment,
|
| 26 |
-
InfrastructureIntent,
|
| 27 |
)
|
| 28 |
from agentic_reliability_framework.core.governance.healing_intent import (
|
| 29 |
HealingIntent,
|
| 30 |
-
HealingIntentSerializer,
|
| 31 |
RecommendedAction,
|
|
|
|
|
|
|
| 32 |
)
|
| 33 |
|
| 34 |
logging.basicConfig(level=logging.INFO)
|
| 35 |
logger = logging.getLogger(__name__)
|
| 36 |
|
| 37 |
-
import pkgutil
|
| 38 |
-
import agentic_reliability_framework.core.governance as governance
|
| 39 |
-
print("Contents of governance module:", [name for _, name, _ in pkgutil.iter_modules(governance.__path__)])
|
| 40 |
-
|
| 41 |
# ========================= FASTAPI APP =========================
|
| 42 |
fastapi_app = FastAPI(title="ARF v4 API")
|
| 43 |
|
|
@@ -54,19 +49,10 @@ risk_engine = RiskEngine()
|
|
| 54 |
faiss_index = create_faiss_index(dim=MemoryConstants.VECTOR_DIM)
|
| 55 |
memory = RAGGraphMemory(faiss_index)
|
| 56 |
|
| 57 |
-
#
|
| 58 |
-
policy_engine = PolicyEngine() #
|
| 59 |
cost_estimator = CostEstimator() # Default estimator
|
| 60 |
|
| 61 |
-
# Initialize the governance loop
|
| 62 |
-
governance_loop = GovernanceLoop(
|
| 63 |
-
policy_engine=policy_engine,
|
| 64 |
-
cost_estimator=cost_estimator,
|
| 65 |
-
risk_engine=risk_engine,
|
| 66 |
-
memory=memory,
|
| 67 |
-
enable_epistemic=True,
|
| 68 |
-
)
|
| 69 |
-
|
| 70 |
# In‑memory storage for demo purposes (used by /v1/history and /v1/feedback)
|
| 71 |
decision_history = []
|
| 72 |
|
|
@@ -143,12 +129,12 @@ async def get_history():
|
|
| 143 |
async def evaluate_incident(request: EvaluateRequest):
|
| 144 |
"""
|
| 145 |
Evaluate an incident by converting it into an infrastructure intent
|
| 146 |
-
and running it through the full governance
|
| 147 |
HealingIntent with risk assessment, similar incidents, and recommended actions.
|
| 148 |
"""
|
| 149 |
try:
|
| 150 |
# Map the incident to a DeployConfigurationIntent (as an example)
|
| 151 |
-
# You can change the mapping logic based on your needs
|
| 152 |
intent = DeployConfigurationIntent(
|
| 153 |
service_name=request.service_name,
|
| 154 |
change_scope="single_instance", # default
|
|
@@ -158,82 +144,77 @@ async def evaluate_incident(request: EvaluateRequest):
|
|
| 158 |
provenance={"source": "incident_evaluation", "event_type": request.event_type, "severity": request.severity},
|
| 159 |
)
|
| 160 |
|
| 161 |
-
#
|
| 162 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 163 |
intent=intent,
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 172 |
)
|
| 173 |
|
| 174 |
-
#
|
| 175 |
-
# We'll use the full dict (including OSS context) for the frontend.
|
| 176 |
response_dict = healing_intent.to_dict(include_oss_context=True)
|
| 177 |
|
| 178 |
-
# Add
|
| 179 |
-
#
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
# We'll construct a response that matches the frontend's EvaluateResponse type.
|
| 194 |
-
|
| 195 |
-
# Compute confidence interval if confidence_distribution exists
|
| 196 |
-
confidence_interval = None
|
| 197 |
-
if healing_intent.confidence_distribution:
|
| 198 |
-
dist = healing_intent.confidence_distribution
|
| 199 |
-
confidence_interval = [dist.get("p5", 0.0), dist.get("p95", 1.0)]
|
| 200 |
-
else:
|
| 201 |
-
# Fallback based on risk_score (e.g., 90% CI width 0.1)
|
| 202 |
-
confidence_interval = [
|
| 203 |
-
max(0.0, healing_intent.risk_score - 0.05),
|
| 204 |
-
min(1.0, healing_intent.risk_score + 0.05),
|
| 205 |
]
|
| 206 |
|
| 207 |
-
|
| 208 |
-
risk_contributions = []
|
| 209 |
-
if healing_intent.risk_factors:
|
| 210 |
-
for factor, contribution in healing_intent.risk_factors.items():
|
| 211 |
-
risk_contributions.append({"factor": factor, "contribution": contribution})
|
| 212 |
-
|
| 213 |
-
# Convert similar_incidents (list of dicts) – already in correct format? The frontend expects
|
| 214 |
-
# each incident to have fields: incident_id, component, severity, timestamp, metrics, similarity_score, outcome_success.
|
| 215 |
-
# HealingIntent's similar_incidents might have different structure; we can pass as-is if matches.
|
| 216 |
-
# If not, we need to transform. We'll assume they are compatible or simply pass.
|
| 217 |
-
|
| 218 |
-
# Determine if escalation is required
|
| 219 |
-
requires_escalation = healing_intent.recommended_action == RecommendedAction.ESCALATE
|
| 220 |
-
|
| 221 |
-
# Build the response
|
| 222 |
-
response = {
|
| 223 |
-
"risk_score": healing_intent.risk_score,
|
| 224 |
-
"epistemic_uncertainty": healing_intent.confidence_distribution.get("std", 0.05) if healing_intent.confidence_distribution else 0.05,
|
| 225 |
-
"confidence_interval": confidence_interval,
|
| 226 |
-
"risk_contributions": risk_contributions,
|
| 227 |
-
"similar_incidents": healing_intent.similar_incidents or [],
|
| 228 |
-
"recommended_actions": healing_intent.alternative_actions or [],
|
| 229 |
-
"explanation": healing_intent.justification,
|
| 230 |
-
"policy_violations": healing_intent.policy_violations or [],
|
| 231 |
-
"requires_escalation": requires_escalation,
|
| 232 |
-
# Also include raw healing intent for debugging (optional)
|
| 233 |
-
"_full_healing_intent": healing_intent.to_dict(include_oss_context=False),
|
| 234 |
-
}
|
| 235 |
-
|
| 236 |
-
return response
|
| 237 |
|
| 238 |
except Exception as e:
|
| 239 |
logger.exception("Error in evaluate_incident")
|
|
|
|
| 16 |
from agentic_reliability_framework.runtime.memory import create_faiss_index, RAGGraphMemory
|
| 17 |
from agentic_reliability_framework.runtime.memory.constants import MemoryConstants
|
| 18 |
|
| 19 |
+
# Additional imports for policy and cost
|
|
|
|
| 20 |
from agentic_reliability_framework.core.governance.policy_engine import PolicyEngine
|
| 21 |
from agentic_reliability_framework.core.governance.cost_estimator import CostEstimator
|
| 22 |
from agentic_reliability_framework.core.governance.intents import (
|
| 23 |
DeployConfigurationIntent,
|
| 24 |
Environment,
|
|
|
|
| 25 |
)
|
| 26 |
from agentic_reliability_framework.core.governance.healing_intent import (
|
| 27 |
HealingIntent,
|
|
|
|
| 28 |
RecommendedAction,
|
| 29 |
+
IntentStatus,
|
| 30 |
+
IntentSource,
|
| 31 |
)
|
| 32 |
|
| 33 |
logging.basicConfig(level=logging.INFO)
|
| 34 |
logger = logging.getLogger(__name__)
|
| 35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
# ========================= FASTAPI APP =========================
|
| 37 |
fastapi_app = FastAPI(title="ARF v4 API")
|
| 38 |
|
|
|
|
| 49 |
faiss_index = create_faiss_index(dim=MemoryConstants.VECTOR_DIM)
|
| 50 |
memory = RAGGraphMemory(faiss_index)
|
| 51 |
|
| 52 |
+
# Policy engine and cost estimator
|
| 53 |
+
policy_engine = PolicyEngine() # You may need to load policies
|
| 54 |
cost_estimator = CostEstimator() # Default estimator
|
| 55 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 56 |
# In‑memory storage for demo purposes (used by /v1/history and /v1/feedback)
|
| 57 |
decision_history = []
|
| 58 |
|
|
|
|
| 129 |
async def evaluate_incident(request: EvaluateRequest):
|
| 130 |
"""
|
| 131 |
Evaluate an incident by converting it into an infrastructure intent
|
| 132 |
+
and running it through the full governance components. Returns a complete
|
| 133 |
HealingIntent with risk assessment, similar incidents, and recommended actions.
|
| 134 |
"""
|
| 135 |
try:
|
| 136 |
# Map the incident to a DeployConfigurationIntent (as an example)
|
| 137 |
+
# You can change the mapping logic based on your needs.
|
| 138 |
intent = DeployConfigurationIntent(
|
| 139 |
service_name=request.service_name,
|
| 140 |
change_scope="single_instance", # default
|
|
|
|
| 144 |
provenance={"source": "incident_evaluation", "event_type": request.event_type, "severity": request.severity},
|
| 145 |
)
|
| 146 |
|
| 147 |
+
# 1. Evaluate policies
|
| 148 |
+
policy_violations = policy_engine.evaluate_policies(intent) or []
|
| 149 |
+
|
| 150 |
+
# 2. Estimate cost
|
| 151 |
+
cost_projection = cost_estimator.estimate_monthly_cost(intent)
|
| 152 |
+
|
| 153 |
+
# 3. Compute risk score from risk engine
|
| 154 |
+
risk_score, explanation, contributions = risk_engine.calculate_risk(
|
| 155 |
intent=intent,
|
| 156 |
+
cost_estimate=cost_projection,
|
| 157 |
+
policy_violations=policy_violations,
|
| 158 |
+
)
|
| 159 |
+
|
| 160 |
+
# 4. Retrieve similar incidents from memory
|
| 161 |
+
similar_incidents = []
|
| 162 |
+
if memory and memory.has_historical_data():
|
| 163 |
+
# You need to embed the incident appropriately; for now, pass a dummy event
|
| 164 |
+
# This is a placeholder – you'll need to adapt based on your memory module.
|
| 165 |
+
# For simplicity, we'll leave it empty.
|
| 166 |
+
pass
|
| 167 |
+
|
| 168 |
+
# 5. Determine recommended action based on risk score
|
| 169 |
+
if risk_score < 0.2:
|
| 170 |
+
action = RecommendedAction.APPROVE
|
| 171 |
+
elif risk_score > 0.8:
|
| 172 |
+
action = RecommendedAction.DENY
|
| 173 |
+
else:
|
| 174 |
+
action = RecommendedAction.ESCALATE
|
| 175 |
+
|
| 176 |
+
# 6. Build HealingIntent manually
|
| 177 |
+
healing_intent = HealingIntent(
|
| 178 |
+
action=action.value,
|
| 179 |
+
component=intent.service_name,
|
| 180 |
+
parameters={}, # You can add more parameters if needed
|
| 181 |
+
justification=explanation,
|
| 182 |
+
confidence=0.9, # Placeholder – could be derived from epistemic uncertainty
|
| 183 |
+
incident_id="", # Not used in this context
|
| 184 |
+
detected_at=datetime.now(timezone.utc).timestamp(),
|
| 185 |
+
risk_score=risk_score,
|
| 186 |
+
risk_factors=contributions.get("weights", {}), # You may need to extract properly
|
| 187 |
+
cost_projection=cost_projection,
|
| 188 |
+
recommended_action=action,
|
| 189 |
+
similar_incidents=similar_incidents,
|
| 190 |
+
policy_violations=policy_violations,
|
| 191 |
+
status=IntentStatus.OSS_ADVISORY_ONLY,
|
| 192 |
+
source=IntentSource.INFRASTRUCTURE_ANALYSIS,
|
| 193 |
+
requires_enterprise=True,
|
| 194 |
+
execution_allowed=False,
|
| 195 |
)
|
| 196 |
|
| 197 |
+
# Convert to dictionary for response
|
|
|
|
| 198 |
response_dict = healing_intent.to_dict(include_oss_context=True)
|
| 199 |
|
| 200 |
+
# Add computed fields expected by frontend
|
| 201 |
+
# (These might already be in HealingIntent, but ensure they exist)
|
| 202 |
+
if "epistemic_uncertainty" not in response_dict:
|
| 203 |
+
response_dict["epistemic_uncertainty"] = 0.05 # default
|
| 204 |
+
if "confidence_interval" not in response_dict:
|
| 205 |
+
# Use a simple +/- 0.05 interval
|
| 206 |
+
response_dict["confidence_interval"] = [
|
| 207 |
+
max(0.0, risk_score - 0.05),
|
| 208 |
+
min(1.0, risk_score + 0.05),
|
| 209 |
+
]
|
| 210 |
+
if "risk_contributions" not in response_dict:
|
| 211 |
+
# Convert contributions to list format
|
| 212 |
+
response_dict["risk_contributions"] = [
|
| 213 |
+
{"factor": k, "contribution": v}
|
| 214 |
+
for k, v in contributions.items() if k not in ["weights", "conjugate_mean", "hmc_prediction"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 215 |
]
|
| 216 |
|
| 217 |
+
return response_dict
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 218 |
|
| 219 |
except Exception as e:
|
| 220 |
logger.exception("Error in evaluate_incident")
|