Spaces:
Running
Running
File size: 21,323 Bytes
aedaf74 91495a2 aedaf74 91495a2 aedaf74 91495a2 aedaf74 91495a2 aedaf74 91495a2 aedaf74 91495a2 aedaf74 91495a2 aedaf74 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 | # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
"""
DC-Ops Environment Implementation.
Wires the thermal and power simulations into OpenEnv's Environment interface.
Each step:
1. Parse the agent's command
2. Apply mutations to simulation state
3. Advance simulation by game-time dt (default 60s)
4. Render dashboard observation
5. Compute reward (via multi-objective RewardFunction)
6. Check termination conditions
"""
from __future__ import annotations
from typing import Any, Optional
from uuid import uuid4
from openenv.core.env_server.interfaces import Environment
from openenv.core.env_server.types import State
try:
from ..config import (
ASHRAE_CLASSES,
DatacenterConfig,
PowerConfig,
make_default_datacenter_config,
load_datacenter_config,
)
from ..models import DcOpsAction, DcOpsObservation
from ..actions.parser import AVAILABLE_ACTIONS, CommandResult, parse_command
from ..rendering.dashboard import render_dashboard
from ..simulation.thermal import ThermalAlarm, ThermalSimulation
from ..simulation.power import PowerAlarm, PowerSimulation
from ..scenarios.base import Scenario, ScenarioResult
from ..scenarios.registry import get_scenario, random_scenario
from ..rewards.reward_function import RewardFunction
except ImportError:
from config import (
ASHRAE_CLASSES,
DatacenterConfig,
PowerConfig,
make_default_datacenter_config,
load_datacenter_config,
)
from models import DcOpsAction, DcOpsObservation
from actions.parser import AVAILABLE_ACTIONS, CommandResult, parse_command
from rendering.dashboard import render_dashboard
from simulation.thermal import ThermalAlarm, ThermalSimulation
from simulation.power import PowerAlarm, PowerSimulation
from scenarios.base import Scenario, ScenarioResult
from scenarios.registry import get_scenario, random_scenario
from rewards.reward_function import RewardFunction
# Default episode configuration
DEFAULT_STEP_BUDGET = 15
DEFAULT_GAME_TIME_PER_STEP_S = 60.0 # 1 minute of sim time per agent step
DEFAULT_SIM_DT_S = 1.0 # Physics integration timestep
class DcOpsEnvironment(Environment):
"""Datacenter operations environment for LLM-based RL agents.
The agent observes a text-based monitoring dashboard and issues
natural-language operator commands. The environment simulates
physics-based thermal and power dynamics.
Episode flow:
1. reset() initializes the datacenter and optionally injects a fault
2. step() parses the command, advances simulation, returns dashboard
3. Episode ends on: budget exhaustion, critical failure, escalation, or resolution
"""
SUPPORTS_CONCURRENT_SESSIONS: bool = True
def __init__(self) -> None:
self._state = State(episode_id=str(uuid4()), step_count=0)
self._thermal_sim: ThermalSimulation | None = None
self._power_sim: PowerSimulation | None = None
self._config: DatacenterConfig | None = None
self._scenario: Scenario | None = None
self._reward_fn: RewardFunction | None = None
self._step_budget: int = DEFAULT_STEP_BUDGET
self._game_time_per_step_s: float = DEFAULT_GAME_TIME_PER_STEP_S
self._sim_dt_s: float = DEFAULT_SIM_DT_S
self._alert: str = ""
self._scenario_type: str = ""
self._done: bool = False
self._cumulative_reward: float = 0.0
self._action_history: list[str] = []
self._escalated: bool = False
def reset(
self,
seed: Optional[int] = None,
episode_id: Optional[str] = None,
**kwargs: Any,
) -> DcOpsObservation:
"""Reset the environment and return initial observation.
Kwargs:
scenario (str | Scenario): Scenario ID (e.g. 'A1') or Scenario instance.
If provided, overrides config/alert/step_budget/scenario_type.
If not provided, uses raw kwargs (backward compatible).
config (DatacenterConfig): Custom datacenter configuration.
config_name (str): Built-in config name ("default", "small", "large").
Used when config is not provided (e.g. from WebSocket/HTTP JSON).
step_budget (int): Max steps for the episode.
game_time_per_step_s (float): Simulation time per step.
scenario_type (str): Scenario category label.
alert (str): Initial alert message.
fault_injection (dict): Fault to inject, e.g.
{"type": "crac_fault", "unit_id": "CRAC-3", "fault": "compressor"}
"""
# Episode state
self._state = State(
episode_id=episode_id or str(uuid4()),
step_count=0,
)
self._done = False
self._cumulative_reward = 0.0
self._action_history = []
self._escalated = False
# Resolve scenario
scenario_arg = kwargs.get("scenario")
if isinstance(scenario_arg, str):
self._scenario = get_scenario(scenario_arg)
elif isinstance(scenario_arg, Scenario):
self._scenario = scenario_arg
elif scenario_arg is None and kwargs.get("random_scenario"):
self._scenario = random_scenario(
scenario_type=kwargs.get("scenario_type"),
difficulty=kwargs.get("difficulty"),
seed=seed,
)
else:
self._scenario = None
# Reset scenario mutable state (counters, flags) for episode reuse
if self._scenario:
self._scenario.reset_state()
# Configuration — scenario can modify the base config
# Support config_name (string) from JSON APIs, or config (DatacenterConfig) from Python
config_arg = kwargs.get("config")
config_name = kwargs.get("config_name")
if isinstance(config_arg, DatacenterConfig):
self._config = config_arg
elif config_name and isinstance(config_name, str) and config_name != "default":
self._config = load_datacenter_config(config_name)
else:
self._config = make_default_datacenter_config()
if self._scenario:
self._config = self._scenario.configure(self._config)
# Episode parameters — scenario provides defaults, kwargs can override
if self._scenario:
self._step_budget = kwargs.get("step_budget", self._scenario.step_budget)
self._game_time_per_step_s = kwargs.get(
"game_time_per_step_s", self._scenario.game_time_per_step_s
)
self._scenario_type = kwargs.get("scenario_type", self._scenario.scenario_type)
self._alert = kwargs.get("alert", self._scenario.alert_message)
else:
self._step_budget = kwargs.get("step_budget", DEFAULT_STEP_BUDGET)
self._game_time_per_step_s = kwargs.get("game_time_per_step_s", DEFAULT_GAME_TIME_PER_STEP_S)
self._scenario_type = kwargs.get("scenario_type", "")
self._alert = kwargs.get("alert", "")
self._sim_dt_s = self._config.simulation_dt_s
# Initialize reward function with scenario-type-aware weights
self._reward_fn = RewardFunction(scenario_type=self._scenario_type)
self._reward_fn.reset() # Defensive: ensure clean state
# Initialize simulations
self._thermal_sim = ThermalSimulation(self._config)
# Initialize power sim if config has power infrastructure
if self._config.power and self._config.power.ups_units:
it_load = self._thermal_sim.state.total_it_load_kw
self._power_sim = PowerSimulation(self._config.power, it_load_kw=it_load)
# Wire power state into datacenter state
self._thermal_sim.state.power = self._power_sim.state
else:
self._power_sim = None
# Apply fault injection — scenario or raw kwargs
if self._scenario:
# Warmup FIRST, then inject fault (so DC is at steady-state)
self._warmup_simulation()
self._scenario.inject_fault(self._thermal_sim, self._power_sim)
else:
fault = kwargs.get("fault_injection")
if fault:
self._apply_fault_injection(fault)
self._warmup_simulation()
# Render initial observation
return self._make_observation(action_result="Environment initialized. Awaiting your command.")
def step(
self,
action: DcOpsAction,
timeout_s: Optional[float] = None,
**kwargs: Any,
) -> DcOpsObservation:
"""Execute one agent step.
1. Parse and execute the command
2. Advance simulation by game_time_per_step_s
3. Check for alarms and termination
4. Compute reward via RewardFunction
5. Return observation
"""
if self._done:
return self._make_observation(
action_result="Episode already ended. Call reset().",
reward=0.0,
)
self._state.step_count += 1
self._action_history.append(action.command)
# 1. Parse and execute command
cmd_result = parse_command(
action.command,
self._thermal_sim,
self._power_sim,
)
# Handle escalation
if cmd_result.command_name == "escalate":
self._escalated = True
self._done = True
# Evaluate scenario for procedure penalties
scenario_result: ScenarioResult | None = None
if self._scenario:
scenario_result = self._scenario.evaluate_step(
self._thermal_sim, self._power_sim,
action.command, self._action_history,
self._state.step_count,
)
# Compute base reward components — escalation penalty is handled
# by scenario procedure rules + action_quality, not doubled here
components = self._reward_fn.compute(
self._thermal_sim, self._power_sim, cmd_result,
action.command, self._action_history, scenario_result,
)
reward = components.total
self._cumulative_reward += reward
return self._make_observation(
action_result=cmd_result.message,
reward=reward,
)
# 2. Handle acknowledge_alarm — clear alert before new alarms overwrite
if cmd_result.command_name == "acknowledge_alarm" and cmd_result.success:
self._alert = ""
# 3. Advance simulation
thermal_alarms, power_alarms = self._advance_simulation()
# 4. Build alert from alarms (only new critical/warning alarms override)
self._update_alert(thermal_alarms, power_alarms)
# 5. Evaluate scenario (before reward, so progress is available)
scenario_result = None
if self._scenario:
scenario_result = self._scenario.evaluate_step(
self._thermal_sim, self._power_sim,
action.command, self._action_history,
self._state.step_count,
)
# 6. Compute reward via RewardFunction
components = self._reward_fn.compute(
self._thermal_sim, self._power_sim, cmd_result,
action.command, self._action_history, scenario_result,
)
reward = components.total
self._cumulative_reward += reward
# 7. Check termination
self._check_termination(thermal_alarms, power_alarms)
# 7b. Scenario resolution
if scenario_result and scenario_result.resolved and not self._done:
self._done = True
# Speed bonus: fraction of budget remaining
speed_bonus = (self._step_budget - self._state.step_count) / self._step_budget
reward += speed_bonus
self._cumulative_reward += speed_bonus
if scenario_result.resolution_message:
self._alert = scenario_result.resolution_message
return self._make_observation(
action_result=cmd_result.message,
reward=reward,
)
@property
def state(self) -> State:
return self._state
# -------------------------------------------------------------------
# Internal methods
# -------------------------------------------------------------------
def _warmup_simulation(self, warmup_steps: int = 120) -> None:
"""Run simulation for a brief warmup to reach quasi-steady-state."""
for _ in range(warmup_steps):
self._thermal_sim.step(self._sim_dt_s)
if self._power_sim:
it_load = self._thermal_sim.state.total_it_load_kw
self._power_sim.step(self._sim_dt_s, it_load)
def _advance_simulation(self) -> tuple[list[ThermalAlarm], list[PowerAlarm]]:
"""Advance simulation by game_time_per_step_s seconds."""
n_substeps = int(self._game_time_per_step_s / self._sim_dt_s)
all_thermal_alarms: list[ThermalAlarm] = []
all_power_alarms: list[PowerAlarm] = []
for _ in range(n_substeps):
# Thermal step
thermal_result = self._thermal_sim.step(self._sim_dt_s)
all_thermal_alarms.extend(thermal_result.alarms)
# Power step
if self._power_sim:
it_load = self._thermal_sim.state.total_it_load_kw
power_result = self._power_sim.step(self._sim_dt_s, it_load)
all_power_alarms.extend(power_result.alarms)
# Deduplicate alarms by type (keep most recent)
thermal_alarms = _dedupe_alarms_by_type(all_thermal_alarms)
power_alarms = _dedupe_alarms_by_type(all_power_alarms)
return thermal_alarms, power_alarms
def _update_alert(
self,
thermal_alarms: list[ThermalAlarm],
power_alarms: list[PowerAlarm],
) -> None:
"""Update the active alert string from current alarms."""
critical_messages: list[str] = []
for alarm in thermal_alarms:
if alarm.severity == "critical":
critical_messages.append(alarm.message)
for alarm in power_alarms:
if alarm.severity == "critical":
critical_messages.append(alarm.message)
if critical_messages:
self._alert = " | ".join(critical_messages[:3]) # Limit to 3 alerts
else:
# Check for warnings
warnings = []
for alarm in thermal_alarms:
if alarm.severity == "warning":
warnings.append(alarm.message)
for alarm in power_alarms:
if alarm.severity == "warning":
warnings.append(alarm.message)
if warnings:
self._alert = warnings[0]
else:
self._alert = ""
def _check_termination(
self,
thermal_alarms: list[ThermalAlarm],
power_alarms: list[PowerAlarm],
) -> None:
"""Check if episode should end."""
# Step budget exhausted
if self._state.step_count >= self._step_budget:
self._done = True
return
# Critical thermal failure: any rack above allowable max
for zone in self._thermal_sim.state.zones:
ashrae = ASHRAE_CLASSES.get(zone.ashrae_class)
if not ashrae:
continue
if zone.max_inlet_temp_c > ashrae.allowable_max_c + 5.0:
self._done = True
self._alert = (
f"CRITICAL: Zone {zone.zone_id} inlet temp "
f"{zone.max_inlet_temp_c:.1f}°C exceeds allowable max "
f"{ashrae.allowable_max_c:.1f}°C by >5°C. Emergency shutdown."
)
return
# UPS battery exhausted
if self._power_sim:
for ups in self._power_sim.state.ups_units:
if ups.mode.value == "fault" and ups.battery_soc <= 0:
self._done = True
self._alert = f"CRITICAL: {ups.unit_id} battery exhausted. Unprotected load."
return
def _apply_fault_injection(self, fault: dict) -> None:
"""Apply a fault injection to the simulation.
Supported fault types:
- crac_fault: {"type": "crac_fault", "unit_id": "CRAC-3", "fault": "compressor"}
- utility_loss: {"type": "utility_loss"}
- ups_fault: {"type": "ups_fault", "unit_id": "UPS-1"}
- rack_load_change: {"type": "rack_load_change", "rack_id": "A-01", "load_kw": 15.0}
- outside_temp: {"type": "outside_temp", "temp_c": 42.0}
"""
fault_type = fault.get("type", "")
if fault_type == "crac_fault":
from ..simulation.types import CRACFaultType
unit_id = fault.get("unit_id", "")
fault_name = fault.get("fault", "compressor")
try:
ft = CRACFaultType(fault_name)
except ValueError:
ft = CRACFaultType.COMPRESSOR
self._thermal_sim.inject_crac_fault(unit_id, ft)
elif fault_type == "utility_loss":
if self._power_sim:
self._power_sim.set_utility_available(False)
elif fault_type == "ups_fault":
if self._power_sim:
unit_id = fault.get("unit_id", "")
self._power_sim.inject_ups_fault(unit_id)
elif fault_type == "rack_load_change":
rack_id = fault.get("rack_id", "")
load_kw = fault.get("load_kw", 8.0)
self._thermal_sim.set_rack_load(rack_id, load_kw)
elif fault_type == "outside_temp":
temp_c = fault.get("temp_c", 35.0)
self._thermal_sim.set_outside_temp(temp_c)
def _make_observation(
self,
action_result: str = "",
reward: float = 0.0,
) -> DcOpsObservation:
"""Build the observation to return to the agent."""
dashboard = render_dashboard(
self._thermal_sim.state,
alert=self._alert,
step=self._state.step_count,
max_steps=self._step_budget,
scenario_type=self._scenario_type,
)
steps_remaining = max(0, self._step_budget - self._state.step_count)
# Build metadata with structured data
dc_state = self._thermal_sim.state
metadata = {
"sim_time_s": dc_state.sim_time_s,
"total_it_load_kw": dc_state.total_it_load_kw,
"total_cooling_power_kw": dc_state.total_cooling_power_kw,
"pue": dc_state.pue,
"outside_temp_c": dc_state.outside_temp_c,
"cumulative_reward": self._cumulative_reward,
"zones": {},
}
for zone in dc_state.zones:
metadata["zones"][zone.zone_id] = {
"cold_aisle_temp_c": zone.cold_aisle_temp_c,
"hot_aisle_temp_c": zone.hot_aisle_temp_c,
"max_inlet_temp_c": zone.max_inlet_temp_c,
"total_it_load_kw": zone.total_it_load_kw,
}
if self._power_sim:
power = self._power_sim.state
metadata["power"] = {
"utility_available": power.utility_available,
"on_generator": power.on_generator,
"total_ups_loss_kw": power.total_ups_loss_kw,
"total_pdu_loss_kw": power.total_pdu_loss_kw,
}
for ups in power.ups_units:
metadata["power"][ups.unit_id] = {
"mode": ups.mode.value,
"battery_soc": ups.battery_soc,
"load_fraction": ups.load_fraction,
"efficiency": ups.efficiency,
}
if self._scenario:
metadata["scenario"] = {
"id": self._scenario.scenario_id,
"name": self._scenario.name,
"difficulty": self._scenario.difficulty,
}
# Use scenario-specific actions if defined, otherwise all actions
actions = AVAILABLE_ACTIONS
if self._scenario and self._scenario.available_actions is not None:
actions = self._scenario.available_actions
return DcOpsObservation(
dashboard=dashboard,
available_actions=actions,
alert=self._alert,
scenario_type=self._scenario_type,
steps_remaining=steps_remaining,
action_result=action_result,
done=self._done,
reward=reward,
metadata=metadata,
)
def _dedupe_alarms_by_type(alarms: list) -> list:
"""Keep only the last alarm of each (component, alarm_type) pair."""
seen: dict[tuple[str, str], Any] = {}
for alarm in alarms:
key = (getattr(alarm, "component", ""), getattr(alarm, "alarm_type", ""))
seen[key] = alarm
return list(seen.values())
|