Spaces:
Running
Running
File size: 18,815 Bytes
c452421 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 | """Core environment implementing the OpenEnv step() / reset() / state() API.
This module owns all mutable episode state. It is deliberately a single-
episode, per-session environment — the FastAPI layer maintains one instance
per session ID, ensuring concurrent agents never share state.
"""
from __future__ import annotations
import copy
import random
import time
from collections import deque
from datetime import datetime, timezone
from typing import Any, Deque, Dict, List, Optional, Tuple
from src.graders import grade
from src.models import (
Action,
ActionType,
EnvironmentState,
GraderResult,
IncidentSeverity,
IncidentStatus,
Observation,
Reward,
ServiceMetrics,
StepResult,
)
from src.rewards import compute_step_reward
from src.scenarios import Scenario, apply_blast_radius, get_scenario
class IncidentResponseEnv:
"""Incident Response Triage environment.
Lifecycle:
env = IncidentResponseEnv()
obs = env.reset("severity_classification")
while not done:
result = env.step(action)
obs, reward, done, info = result.observation, result.reward, result.done, result.info
grader_result = env.grade()
"""
# How many per-step metric snapshots to retain per service
_TSDB_MAX_SAMPLES: int = 64
def __init__(self) -> None:
self._scenario: Optional[Scenario] = None
self._task_id: Optional[str] = None
self._step: int = 0
self._done: bool = True
self._cumulative_reward: float = 0.0
# Progressive state
self._investigated: List[str] = []
self._severity_classified: Optional[IncidentSeverity] = None
self._diagnosis: Optional[str] = None
self._remediations: List[str] = []
self._escalations: List[str] = []
self._communications: List[str] = []
self._actions_history: List[Dict[str, Any]] = []
self._incident_status: IncidentStatus = IncidentStatus.OPEN
self._last_message: str = ""
# Logs / metrics revealed so far
self._revealed_logs: Dict[str, list] = {}
self._revealed_metrics: Dict[str, Any] = {}
# TSDB ring buffer: service -> deque of (unix_timestamp, ServiceMetrics)
# Populated after every step so /prometheus/query_range returns real history.
self._metric_history: Dict[str, Deque[Tuple[float, ServiceMetrics]]] = {}
# ------------------------------------------------------------------
# reset()
# ------------------------------------------------------------------
def reset(self, task_id: str, variant_seed: int = 0) -> Observation:
"""Reset the environment for a new episode on the given task.
Args:
task_id: Task to run.
variant_seed: Scenario variant index (default 0 = primary scenario).
"""
scenario = get_scenario(task_id, variant_seed=variant_seed)
self._scenario = scenario
self._task_id = task_id
self._step = 0
self._done = False
self._cumulative_reward = 0.0
self._investigated = []
self._severity_classified = None
self._diagnosis = None
self._remediations = []
self._escalations = []
self._communications = []
self._actions_history = []
self._incident_status = IncidentStatus.OPEN
self._last_message = "Incident opened. Review the alerts and begin your investigation."
self._revealed_logs = {}
self._revealed_metrics = {}
self._metric_history = {}
# Record step-0 snapshot so range queries have at least one data point
self._record_metric_snapshot()
return self._build_observation()
# ------------------------------------------------------------------
# step()
# ------------------------------------------------------------------
def step(self, action: Action) -> StepResult:
"""Process one agent action and return the result."""
if self._done:
raise RuntimeError("Episode is done. Call reset() first.")
if self._scenario is None:
raise RuntimeError("Environment not initialised. Call reset() first.")
self._step += 1
scenario = self._scenario
# Record action
self._actions_history.append(action.model_dump())
# Process action effects
self._process_action(action, scenario)
# Compute reward
reward = compute_step_reward(
action=action,
scenario=scenario,
step_number=self._step,
already_investigated=self._investigated,
already_classified=self._severity_classified,
already_diagnosed=self._diagnosis,
already_remediated=self._remediations,
already_escalated=self._escalations,
already_communicated=self._communications,
actions_history=self._actions_history,
)
self._cumulative_reward += reward.value
# Apply action state changes (after reward so duplicates are penalised first)
self._apply_state_changes(action, scenario)
# Record metric snapshot into TSDB ring buffer
self._record_metric_snapshot()
# Check episode termination
done = self._check_done(scenario)
self._done = done
obs = self._build_observation()
info: Dict[str, Any] = {
"cumulative_reward": round(self._cumulative_reward, 4),
"steps_remaining": max(0, scenario.max_steps - self._step),
}
if done:
info["grader"] = self.grade().model_dump()
return StepResult(observation=obs, reward=reward, done=done, info=info)
# ------------------------------------------------------------------
# state()
# ------------------------------------------------------------------
def state(self) -> EnvironmentState:
"""Return the full internal state snapshot."""
return EnvironmentState(
task_id=self._task_id or "",
scenario_id=self._scenario.scenario_id if self._scenario else "",
step_number=self._step,
max_steps=self._scenario.max_steps if self._scenario else 0,
incident_status=self._incident_status,
done=self._done,
cumulative_reward=round(self._cumulative_reward, 4),
total_steps_taken=self._step,
alerts=list(self._scenario.initial_alerts) if self._scenario else [],
actions_history=copy.deepcopy(self._actions_history),
severity_classified=self._severity_classified,
diagnosis=self._diagnosis,
remediations_applied=list(self._remediations),
escalations_made=list(self._escalations),
communications_sent=list(self._communications),
investigated_services=list(self._investigated),
)
# ------------------------------------------------------------------
# grade()
# ------------------------------------------------------------------
def grade(self) -> GraderResult:
"""Grade the current episode. Can be called mid-episode or after done."""
if self._scenario is None or self._task_id is None:
raise RuntimeError("No episode in progress.")
return grade(self._task_id, self.state(), self._scenario)
def live_metrics(self) -> Dict[str, ServiceMetrics]:
"""Return service metrics with blast-radius degradation at the current step.
Safe to call at any point (including before any actions are taken).
Returns an empty dict when no episode is in progress.
This is the same numerical data the agent would eventually see via
investigate actions, but served here without consuming an action slot —
analogous to a Prometheus scrape that is always available passively.
"""
if self._scenario is None:
return {}
return apply_blast_radius(self._scenario, self._step)
def metric_history(
self,
start: float,
end: float,
step_seconds: float = 1.0,
) -> Dict[str, List[Tuple[float, ServiceMetrics]]]:
"""Return per-service metric history in the [start, end] time window.
This powers the Prometheus-compatible ``/prometheus/query_range`` endpoint.
Each entry is a ``(unix_timestamp, ServiceMetrics)`` tuple, sampled once
per environment step. The ``step_seconds`` parameter is accepted for
API compatibility but does not resample — the ring buffer already stores
one sample per episode step.
Returns an empty dict when no episode is in progress.
"""
if not self._metric_history:
return {}
result: Dict[str, List[Tuple[float, ServiceMetrics]]] = {}
for svc, dq in self._metric_history.items():
samples = [(ts, m) for ts, m in dq if start <= ts <= end]
if samples:
result[svc] = samples
return result
# ------------------------------------------------------------------
# Private helpers
# ------------------------------------------------------------------
def _record_metric_snapshot(self) -> None:
"""Append the current blast-radius metrics to the per-service ring buffer."""
if self._scenario is None:
return
ts = time.time()
live = apply_blast_radius(self._scenario, self._step)
for svc, metrics in live.items():
if svc not in self._metric_history:
self._metric_history[svc] = deque(maxlen=self._TSDB_MAX_SAMPLES)
self._metric_history[svc].append((ts, metrics))
def _process_action(self, action: Action, scenario: Scenario) -> None:
"""Process action effects (messages, state transitions)."""
if action.action_type == ActionType.INVESTIGATE:
target = (action.target or "").strip()
if target not in scenario.available_services:
self._last_message = (
f"Unknown service '{target}'. "
f"Available: {', '.join(scenario.available_services)}"
)
elif target in self._investigated:
self._last_message = f"Already investigated {target}. Logs and metrics available."
else:
# Reveal logs (always static — logs are historical records)
if target in scenario.service_logs:
self._revealed_logs[target] = [
entry.model_dump() for entry in scenario.service_logs[target]
]
# Reveal LIVE metrics with blast-radius degradation applied
live_metrics = apply_blast_radius(scenario, self._step)
if target in live_metrics:
self._revealed_metrics[target] = live_metrics[target].model_dump()
self._last_message = (
f"Investigation of {target} complete. Logs and live metrics now available."
)
if self._incident_status == IncidentStatus.OPEN:
self._incident_status = IncidentStatus.INVESTIGATING
elif action.action_type == ActionType.CLASSIFY:
severity_str = action.parameters.get("severity", "")
try:
sev = IncidentSeverity(severity_str)
self._last_message = f"Incident classified as {sev.value}."
except ValueError:
self._last_message = (
f"Invalid severity '{severity_str}'. Use P1, P2, P3, or P4."
)
elif action.action_type == ActionType.DIAGNOSE:
root_cause = action.parameters.get("root_cause", "")
target_svc = (action.target or "").strip()
self._last_message = (
f"Diagnosis recorded: root cause in {target_svc} — {root_cause[:120]}"
)
elif action.action_type == ActionType.REMEDIATE:
rem_action = action.parameters.get("action", "")
target_svc = (action.target or "").strip()
if not rem_action or not target_svc:
self._last_message = "Remediation requires 'action' parameter and 'target' service."
else:
self._last_message = (
f"Remediation '{rem_action}' applied to {target_svc}."
)
if self._incident_status in (IncidentStatus.OPEN, IncidentStatus.INVESTIGATING):
self._incident_status = IncidentStatus.MITIGATING
elif action.action_type == ActionType.ESCALATE:
team = (action.target or "").strip()
priority = action.parameters.get("priority", "high")
message = action.parameters.get("message", "")
self._last_message = (
f"Escalated to {team} (priority: {priority}). "
f"Message: {message[:80]}"
)
elif action.action_type == ActionType.COMMUNICATE:
channel = (action.target or "status_page").strip()
message = action.parameters.get("message", "")
self._last_message = (
f"Status update posted to {channel}: {message[:100]}"
)
def _apply_state_changes(self, action: Action, scenario: Scenario) -> None:
"""Persist state changes after reward is computed."""
if action.action_type == ActionType.INVESTIGATE:
target = (action.target or "").strip()
if target in scenario.available_services and target not in self._investigated:
self._investigated.append(target)
elif action.action_type == ActionType.CLASSIFY:
severity_str = action.parameters.get("severity", "")
try:
self._severity_classified = IncidentSeverity(severity_str)
except ValueError:
pass
elif action.action_type == ActionType.DIAGNOSE:
if self._diagnosis is None:
self._diagnosis = action.parameters.get("root_cause", "")
elif action.action_type == ActionType.REMEDIATE:
rem_action = action.parameters.get("action", "")
target_svc = (action.target or "").strip()
if rem_action and target_svc:
key = f"{rem_action}:{target_svc}"
if key not in self._remediations:
self._remediations.append(key)
elif action.action_type == ActionType.ESCALATE:
team = (action.target or "").strip()
if team and team not in self._escalations:
self._escalations.append(team)
elif action.action_type == ActionType.COMMUNICATE:
message = action.parameters.get("message", "")
if message:
self._communications.append(message[:200])
def _check_done(self, scenario: Scenario) -> bool:
"""Episode ends when max steps reached or incident resolved."""
if self._step >= scenario.max_steps:
self._last_message += " [Episode ended: max steps reached.]"
return True
# For easy task: done once classified
if scenario.task_id == "severity_classification" and self._severity_classified is not None:
# Give agent a chance to investigate first, but if classified, we're done
# Actually let them keep going for a few more steps if they want
if self._step >= 2 or self._severity_classified is not None:
# Check if the last action was classify
if (self._actions_history and
self._actions_history[-1].get("action_type") == ActionType.CLASSIFY.value):
self._incident_status = IncidentStatus.RESOLVED
self._last_message += " [Episode complete: severity classified.]"
return True
# For medium: done once diagnosed AND remediated
if scenario.task_id == "root_cause_analysis":
if self._diagnosis and len(self._remediations) > 0:
self._incident_status = IncidentStatus.RESOLVED
self._last_message += " [Episode complete: diagnosed and remediated.]"
return True
# For hard: done once all major actions taken
if scenario.task_id == "full_incident_management":
has_severity = self._severity_classified is not None
has_diagnosis = self._diagnosis is not None
has_remediation = len(self._remediations) > 0
has_escalation = len(self._escalations) > 0
has_communication = len(self._communications) > 0
if all([has_severity, has_diagnosis, has_remediation, has_escalation, has_communication]):
self._incident_status = IncidentStatus.RESOLVED
self._last_message += " [Episode complete: full incident management workflow completed.]"
return True
return False
def _build_observation(self) -> Observation:
"""Construct the current observation from internal state."""
scenario = self._scenario
if scenario is None:
raise RuntimeError("No scenario loaded.")
# Reconstruct typed logs from revealed data
from src.models import LogEntry, ServiceMetrics
typed_logs = {}
for svc, entries in self._revealed_logs.items():
typed_logs[svc] = [LogEntry(**e) for e in entries]
typed_metrics = {}
for svc, m in self._revealed_metrics.items():
typed_metrics[svc] = ServiceMetrics(**m)
return Observation(
incident_id=scenario.incident_id,
timestamp=datetime.now(timezone.utc).isoformat(),
step_number=self._step,
max_steps=scenario.max_steps,
task_id=scenario.task_id,
task_description=scenario.description,
alerts=list(scenario.initial_alerts),
available_services=list(scenario.available_services),
investigated_services=list(self._investigated),
logs=typed_logs,
metrics=typed_metrics,
incident_status=self._incident_status,
severity_classified=self._severity_classified,
diagnosis=self._diagnosis,
actions_taken=[
f"Step {i+1}: {a.get('action_type', '?')} → {a.get('target', 'N/A')}"
for i, a in enumerate(self._actions_history)
],
remediations_applied=list(self._remediations),
escalations_made=list(self._escalations),
communications_sent=list(self._communications),
message=self._last_message,
)
|