Update app.py
Browse files
app.py
CHANGED
|
@@ -2,9 +2,9 @@
|
|
| 2 |
import logging
|
| 3 |
import uuid
|
| 4 |
from datetime import datetime, timezone
|
| 5 |
-
from typing import Dict, Optional
|
| 6 |
|
| 7 |
-
from fastapi import FastAPI
|
| 8 |
from fastapi.middleware.cors import CORSMiddleware
|
| 9 |
from fastapi.openapi.docs import get_swagger_ui_html, get_redoc_html
|
| 10 |
from fastapi.responses import RedirectResponse
|
|
@@ -16,6 +16,21 @@ 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 |
logging.basicConfig(level=logging.INFO)
|
| 20 |
logger = logging.getLogger(__name__)
|
| 21 |
|
|
@@ -35,7 +50,20 @@ risk_engine = RiskEngine()
|
|
| 35 |
faiss_index = create_faiss_index(dim=MemoryConstants.VECTOR_DIM)
|
| 36 |
memory = RAGGraphMemory(faiss_index)
|
| 37 |
|
| 38 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
decision_history = []
|
| 40 |
|
| 41 |
# ========================= PYDANTIC MODELS =========================
|
|
@@ -107,21 +135,105 @@ async def get_history():
|
|
| 107 |
"""Return the last 10 decisions."""
|
| 108 |
return decision_history[-10:]
|
| 109 |
|
| 110 |
-
@fastapi_app.post("/v1/incidents/evaluate"
|
| 111 |
async def evaluate_incident(request: EvaluateRequest):
|
| 112 |
"""
|
| 113 |
-
Evaluate an incident
|
| 114 |
-
|
|
|
|
| 115 |
"""
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 125 |
|
| 126 |
@fastapi_app.post("/v1/feedback")
|
| 127 |
async def record_outcome(decision_id: str, success: bool):
|
|
@@ -225,10 +337,9 @@ with gr.Blocks(title="ARF v4 Demo") as demo:
|
|
| 225 |
history_btn.click(fn=lambda: decision_history[-10:], outputs=decision_output)
|
| 226 |
|
| 227 |
# ========================= Mount Gradio and Add Documentation Routes =========================
|
| 228 |
-
# Mount Gradio at "/api" – this means Gradio will handle all requests starting with "/api".
|
| 229 |
app = gr.mount_gradio_app(fastapi_app, demo, path="/api")
|
| 230 |
|
| 231 |
-
# Add documentation routes at "/docs"
|
| 232 |
@app.get("/docs", include_in_schema=False)
|
| 233 |
async def swagger_ui():
|
| 234 |
return get_swagger_ui_html(
|
|
@@ -247,7 +358,6 @@ async def redoc_ui():
|
|
| 247 |
async def openapi():
|
| 248 |
return fastapi_app.openapi()
|
| 249 |
|
| 250 |
-
# Optional redirect from /api/docs to /docs for backward compatibility.
|
| 251 |
@app.get("/api/docs", include_in_schema=False)
|
| 252 |
async def redirect_docs():
|
| 253 |
return RedirectResponse(url="/docs")
|
|
|
|
| 2 |
import logging
|
| 3 |
import uuid
|
| 4 |
from datetime import datetime, timezone
|
| 5 |
+
from typing import Dict, Optional, List
|
| 6 |
|
| 7 |
+
from fastapi import FastAPI, HTTPException
|
| 8 |
from fastapi.middleware.cors import CORSMiddleware
|
| 9 |
from fastapi.openapi.docs import get_swagger_ui_html, get_redoc_html
|
| 10 |
from fastapi.responses import RedirectResponse
|
|
|
|
| 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 governance loop and healing intent
|
| 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 |
|
|
|
|
| 50 |
faiss_index = create_faiss_index(dim=MemoryConstants.VECTOR_DIM)
|
| 51 |
memory = RAGGraphMemory(faiss_index)
|
| 52 |
|
| 53 |
+
# Create policy engine and cost estimator (use default implementations)
|
| 54 |
+
policy_engine = PolicyEngine() # Will need policies loaded if any
|
| 55 |
+
cost_estimator = CostEstimator() # Default estimator
|
| 56 |
+
|
| 57 |
+
# Initialize the governance loop
|
| 58 |
+
governance_loop = GovernanceLoop(
|
| 59 |
+
policy_engine=policy_engine,
|
| 60 |
+
cost_estimator=cost_estimator,
|
| 61 |
+
risk_engine=risk_engine,
|
| 62 |
+
memory=memory,
|
| 63 |
+
enable_epistemic=True,
|
| 64 |
+
)
|
| 65 |
+
|
| 66 |
+
# In‑memory storage for demo purposes (used by /v1/history and /v1/feedback)
|
| 67 |
decision_history = []
|
| 68 |
|
| 69 |
# ========================= PYDANTIC MODELS =========================
|
|
|
|
| 135 |
"""Return the last 10 decisions."""
|
| 136 |
return decision_history[-10:]
|
| 137 |
|
| 138 |
+
@fastapi_app.post("/v1/incidents/evaluate")
|
| 139 |
async def evaluate_incident(request: EvaluateRequest):
|
| 140 |
"""
|
| 141 |
+
Evaluate an incident by converting it into an infrastructure intent
|
| 142 |
+
and running it through the full governance loop. Returns a complete
|
| 143 |
+
HealingIntent with risk assessment, similar incidents, and recommended actions.
|
| 144 |
"""
|
| 145 |
+
try:
|
| 146 |
+
# Map the incident to a DeployConfigurationIntent (as an example)
|
| 147 |
+
# You can change the mapping logic based on your needs
|
| 148 |
+
intent = DeployConfigurationIntent(
|
| 149 |
+
service_name=request.service_name,
|
| 150 |
+
change_scope="single_instance", # default
|
| 151 |
+
deployment_target=Environment.DEV, # assume dev for now
|
| 152 |
+
configuration=request.metrics,
|
| 153 |
+
requester="system",
|
| 154 |
+
provenance={"source": "incident_evaluation", "event_type": request.event_type, "severity": request.severity},
|
| 155 |
+
)
|
| 156 |
+
|
| 157 |
+
# Run through governance loop
|
| 158 |
+
healing_intent: HealingIntent = governance_loop.run(
|
| 159 |
+
intent=intent,
|
| 160 |
+
context={
|
| 161 |
+
"incident_metadata": {
|
| 162 |
+
"service_name": request.service_name,
|
| 163 |
+
"event_type": request.event_type,
|
| 164 |
+
"severity": request.severity,
|
| 165 |
+
"metrics": request.metrics,
|
| 166 |
+
}
|
| 167 |
+
},
|
| 168 |
+
)
|
| 169 |
+
|
| 170 |
+
# Serialize the healing intent to a dictionary suitable for JSON response
|
| 171 |
+
# We'll use the full dict (including OSS context) for the frontend.
|
| 172 |
+
response_dict = healing_intent.to_dict(include_oss_context=True)
|
| 173 |
+
|
| 174 |
+
# Add any extra fields expected by the frontend that might not be in HealingIntent
|
| 175 |
+
# The frontend's EvaluateResponse includes fields like base_risk, weight, etc.
|
| 176 |
+
# We can compute these from the risk contributions if needed.
|
| 177 |
+
# For simplicity, we'll map the healing intent fields to the expected shape.
|
| 178 |
+
# The frontend expects:
|
| 179 |
+
# - risk_score (already present)
|
| 180 |
+
# - epistemic_uncertainty (from confidence_distribution)
|
| 181 |
+
# - confidence_interval (from confidence_distribution)
|
| 182 |
+
# - risk_contributions (from risk_factors)
|
| 183 |
+
# - similar_incidents (already present)
|
| 184 |
+
# - recommended_actions (from action or alternative_actions)
|
| 185 |
+
# - explanation (from justification)
|
| 186 |
+
# - policy_violations (already present)
|
| 187 |
+
# - requires_escalation (based on recommended_action)
|
| 188 |
+
#
|
| 189 |
+
# We'll construct a response that matches the frontend's EvaluateResponse type.
|
| 190 |
+
|
| 191 |
+
# Compute confidence interval if confidence_distribution exists
|
| 192 |
+
confidence_interval = None
|
| 193 |
+
if healing_intent.confidence_distribution:
|
| 194 |
+
dist = healing_intent.confidence_distribution
|
| 195 |
+
confidence_interval = [dist.get("p5", 0.0), dist.get("p95", 1.0)]
|
| 196 |
+
else:
|
| 197 |
+
# Fallback based on risk_score (e.g., 90% CI width 0.1)
|
| 198 |
+
confidence_interval = [
|
| 199 |
+
max(0.0, healing_intent.risk_score - 0.05),
|
| 200 |
+
min(1.0, healing_intent.risk_score + 0.05),
|
| 201 |
+
]
|
| 202 |
+
|
| 203 |
+
# Convert risk_factors to list of RiskContribution objects
|
| 204 |
+
risk_contributions = []
|
| 205 |
+
if healing_intent.risk_factors:
|
| 206 |
+
for factor, contribution in healing_intent.risk_factors.items():
|
| 207 |
+
risk_contributions.append({"factor": factor, "contribution": contribution})
|
| 208 |
+
|
| 209 |
+
# Convert similar_incidents (list of dicts) – already in correct format? The frontend expects
|
| 210 |
+
# each incident to have fields: incident_id, component, severity, timestamp, metrics, similarity_score, outcome_success.
|
| 211 |
+
# HealingIntent's similar_incidents might have different structure; we can pass as-is if matches.
|
| 212 |
+
# If not, we need to transform. We'll assume they are compatible or simply pass.
|
| 213 |
+
|
| 214 |
+
# Determine if escalation is required
|
| 215 |
+
requires_escalation = healing_intent.recommended_action == RecommendedAction.ESCALATE
|
| 216 |
+
|
| 217 |
+
# Build the response
|
| 218 |
+
response = {
|
| 219 |
+
"risk_score": healing_intent.risk_score,
|
| 220 |
+
"epistemic_uncertainty": healing_intent.confidence_distribution.get("std", 0.05) if healing_intent.confidence_distribution else 0.05,
|
| 221 |
+
"confidence_interval": confidence_interval,
|
| 222 |
+
"risk_contributions": risk_contributions,
|
| 223 |
+
"similar_incidents": healing_intent.similar_incidents or [],
|
| 224 |
+
"recommended_actions": healing_intent.alternative_actions or [],
|
| 225 |
+
"explanation": healing_intent.justification,
|
| 226 |
+
"policy_violations": healing_intent.policy_violations or [],
|
| 227 |
+
"requires_escalation": requires_escalation,
|
| 228 |
+
# Also include raw healing intent for debugging (optional)
|
| 229 |
+
"_full_healing_intent": healing_intent.to_dict(include_oss_context=False),
|
| 230 |
+
}
|
| 231 |
+
|
| 232 |
+
return response
|
| 233 |
+
|
| 234 |
+
except Exception as e:
|
| 235 |
+
logger.exception("Error in evaluate_incident")
|
| 236 |
+
raise HTTPException(status_code=500, detail=str(e))
|
| 237 |
|
| 238 |
@fastapi_app.post("/v1/feedback")
|
| 239 |
async def record_outcome(decision_id: str, success: bool):
|
|
|
|
| 337 |
history_btn.click(fn=lambda: decision_history[-10:], outputs=decision_output)
|
| 338 |
|
| 339 |
# ========================= Mount Gradio and Add Documentation Routes =========================
|
|
|
|
| 340 |
app = gr.mount_gradio_app(fastapi_app, demo, path="/api")
|
| 341 |
|
| 342 |
+
# Add documentation routes at "/docs"
|
| 343 |
@app.get("/docs", include_in_schema=False)
|
| 344 |
async def swagger_ui():
|
| 345 |
return get_swagger_ui_html(
|
|
|
|
| 358 |
async def openapi():
|
| 359 |
return fastapi_app.openapi()
|
| 360 |
|
|
|
|
| 361 |
@app.get("/api/docs", include_in_schema=False)
|
| 362 |
async def redirect_docs():
|
| 363 |
return RedirectResponse(url="/docs")
|