File size: 20,193 Bytes
1a4aa87 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 | """
Worker Registry for Distributed Worker Coordination
Provides worker registration, heartbeat management, and worker health tracking.
Handles worker lifecycle: REGISTERED -> ACTIVE -> DEGRADED -> OFFLINE
"""
import socket
import uuid
from datetime import datetime, timedelta
from typing import Dict, List, Optional
from sqlalchemy import select, update
from sqlalchemy.ext.asyncio import AsyncSession
from backend.db.models import Worker
from backend.db.session import get_db_context
from backend.logging.logger import get_logger
from .job_schema import JobStatus
from .worker_schema import (
GPUInfo,
HeartbeatRequest,
HeartbeatResponse,
WorkerRegistrationRequest,
WorkerRegistrationResponse,
WorkerStatus,
WorkerStatusResponse,
WorkerListResponse,
WorkerMetricsResponse,
)
logger = get_logger("queue.worker_registry", component="queue")
# Configuration
DEFAULT_HEARTBEAT_INTERVAL = 30 # seconds
DEFAULT_HEARTBEAT_TIMEOUT = 120 # seconds (worker marked offline after this)
class WorkerRegistry:
"""
Manages worker registration, heartbeat, and health monitoring.
Responsibilities:
- Worker registration and deregistration
- Heartbeat processing
- Worker health tracking
- Offline worker detection
- Load factor calculation
"""
def __init__(self):
self._cache: Dict[str, Worker] = {}
self._last_cleanup = datetime.utcnow()
self._cleanup_interval = 60 # Run cleanup every 60 seconds
def _generate_worker_id(self, hostname: str) -> str:
"""Generate a unique worker ID based on hostname and UUID."""
return f"{hostname}-{uuid.uuid4().hex[:8]}"
async def register_worker(
self,
request: WorkerRegistrationRequest,
) -> WorkerRegistrationResponse:
"""
Register a new worker in the cluster.
Args:
request: Worker registration request
Returns:
Worker registration response with assigned worker_id
"""
try:
worker_id = self._generate_worker_id(request.hostname)
# Calculate total GPU memory if GPU info provided
gpu_memory_total = request.gpu_memory_total
if request.gpu_info:
gpu_memory_total = sum(gpu.memory_total_mb for gpu in request.gpu_info)
gpu_count = request.gpu_count
if request.gpu_info:
gpu_count = len(request.gpu_info)
async with get_db_context() as session:
# Create worker record
worker = Worker(
worker_id=worker_id,
hostname=request.hostname,
gpu_count=gpu_count,
gpu_memory_total=gpu_memory_total,
gpu_memory_used=0,
status=WorkerStatus.ACTIVE.value,
last_heartbeat=datetime.utcnow(),
active_jobs=0,
max_concurrent_jobs=request.max_concurrent_jobs,
capabilities=request.capabilities,
worker_metadata=request.worker_metadata,
)
session.add(worker)
await session.commit()
logger.info(
"Worker registered",
worker_id=worker_id,
hostname=request.hostname,
gpu_count=gpu_count,
)
# Cache the worker
self._cache[worker_id] = worker
return WorkerRegistrationResponse(
worker_id=worker_id,
status=WorkerStatus.ACTIVE,
registered_at=worker.registered_at,
heartbeat_interval=DEFAULT_HEARTBEAT_INTERVAL,
heartbeat_timeout=DEFAULT_HEARTBEAT_TIMEOUT,
)
except Exception as e:
logger.error(
"Failed to register worker",
error=str(e),
hostname=request.hostname,
)
raise
async def heartbeat(
self,
request: HeartbeatRequest,
) -> HeartbeatResponse:
"""
Process worker heartbeat and update worker status.
Args:
request: Heartbeat request with worker status
Returns:
Heartbeat response
"""
try:
worker_id = request.worker_id
async with get_db_context() as session:
# Get worker from database
stmt = select(Worker).where(Worker.worker_id == worker_id)
result = await session.execute(stmt)
worker = result.scalar_one_or_none()
if worker is None:
logger.warning(
"Heartbeat from unknown worker",
worker_id=worker_id,
)
raise ValueError(f"Worker {worker_id} not found")
# Update worker status based on heartbeat
now = datetime.utcnow()
worker.last_heartbeat = now
# Update optional fields if provided
if request.gpu_usage is not None:
worker.gpu_usage_percent = request.gpu_usage
if request.gpu_memory_used is not None:
worker.gpu_memory_used = request.gpu_memory_used
if request.active_jobs is not None:
worker.active_jobs = request.active_jobs
# Update GPU info if provided
if request.gpu_info:
total_used = sum(gpu.memory_used_mb for gpu in request.gpu_info)
worker.gpu_memory_used = total_used
avg_util = sum(gpu.utilization_percent for gpu in request.gpu_info) / len(request.gpu_info)
worker.gpu_usage_percent = avg_util
# Update status if provided
if request.status is not None:
worker.status = request.status.value
# Determine if worker should be degraded
if worker.gpu_usage_percent > 90:
worker.status = WorkerStatus.DEGRADED.value
elif worker.active_jobs >= worker.max_concurrent_jobs:
worker.status = WorkerStatus.DEGRADED.value
elif worker.status == WorkerStatus.OFFLINE.value:
# Reactivate if was offline
worker.status = WorkerStatus.ACTIVE.value
await session.commit()
# Update cache
self._cache[worker_id] = worker
logger.debug(
"Heartbeat processed",
worker_id=worker_id,
active_jobs=worker.active_jobs,
gpu_usage=worker.gpu_usage_percent,
)
return HeartbeatResponse(
worker_id=worker_id,
status=WorkerStatus(worker.status),
timestamp=now,
assigned_jobs=worker.active_jobs,
)
except ValueError:
raise
except Exception as e:
logger.error(
"Failed to process heartbeat",
worker_id=request.worker_id,
error=str(e),
)
raise
async def get_worker_status(
self,
worker_id: str,
) -> Optional[WorkerStatusResponse]:
"""
Get worker status by ID.
Args:
worker_id: Worker ID
Returns:
Worker status response or None if not found
"""
# Check cache first
if worker_id in self._cache:
worker = self._cache[worker_id]
return self._build_worker_status_response(worker)
# Fetch from database
try:
async with get_db_context() as session:
stmt = select(Worker).where(Worker.worker_id == worker_id)
result = await session.execute(stmt)
worker = result.scalar_one_or_none()
if worker is None:
return None
return self._build_worker_status_response(worker)
except Exception as e:
logger.error(
"Failed to get worker status",
worker_id=worker_id,
error=str(e),
)
return None
async def list_workers(
self,
status_filter: Optional[WorkerStatus] = None,
) -> WorkerListResponse:
"""
List all workers in the cluster.
Args:
status_filter: Optional status filter
Returns:
Worker list response
"""
try:
async with get_db_context() as session:
stmt = select(Worker).order_by(Worker.registered_at.desc())
if status_filter:
stmt = stmt.where(Worker.status == status_filter.value)
result = await session.execute(stmt)
workers = result.scalars().all()
# Update cache
for worker in workers:
self._cache[worker.worker_id] = worker
# Build responses
worker_responses = [
self._build_worker_status_response(w) for w in workers
]
active_count = sum(
1 for w in workers if w.status == WorkerStatus.ACTIVE.value
)
offline_count = sum(
1 for w in workers if w.status == WorkerStatus.OFFLINE.value
)
return WorkerListResponse(
workers=worker_responses,
total=len(workers),
active=active_count,
offline=offline_count,
)
except Exception as e:
logger.error(
"Failed to list workers",
error=str(e),
)
return WorkerListResponse(workers=[], total=0, active=0, offline=0)
async def get_worker_metrics(self) -> WorkerMetricsResponse:
"""
Get cluster-wide worker metrics.
Returns:
Worker metrics response
"""
try:
async with get_db_context() as session:
stmt = select(Worker)
result = await session.execute(stmt)
workers = result.scalars().all()
total_workers = len(workers)
active_workers = sum(
1 for w in workers if w.status == WorkerStatus.ACTIVE.value
)
offline_workers = sum(
1 for w in workers if w.status == WorkerStatus.OFFLINE.value
)
degraded_workers = sum(
1 for w in workers if w.status == WorkerStatus.DEGRADED.value
)
total_gpus = sum(w.gpu_count for w in workers)
total_gpu_memory = sum(w.gpu_memory_total for w in workers)
used_gpu_memory = sum(w.gpu_memory_used for w in workers)
total_active_jobs = sum(w.active_jobs for w in workers)
# Calculate average load factor
if workers:
load_factors = [
w.active_jobs / w.max_concurrent_jobs
for w in workers
if w.max_concurrent_jobs > 0
]
avg_load = sum(load_factors) / len(load_factors) if load_factors else 0.0
else:
avg_load = 0.0
# Get queue length (pending jobs)
from backend.queue.producer import _job_queue
queue_length = sum(
1 for j in _job_queue
if j.status == JobStatus.QUEUED
)
return WorkerMetricsResponse(
total_workers=total_workers,
active_workers=active_workers,
offline_workers=offline_workers,
degraded_workers=degraded_workers,
total_gpus=total_gpus,
total_gpu_memory_mb=total_gpu_memory,
used_gpu_memory_mb=used_gpu_memory,
total_active_jobs=total_active_jobs,
average_load_factor=avg_load,
queue_length=queue_length,
)
except Exception as e:
logger.error(
"Failed to get worker metrics",
error=str(e),
)
return WorkerMetricsResponse(
total_workers=0,
active_workers=0,
offline_workers=0,
degraded_workers=0,
total_gpus=0,
total_gpu_memory_mb=0,
used_gpu_memory_mb=0,
total_active_jobs=0,
average_load_factor=0.0,
queue_length=0,
)
async def cleanup_offline_workers(self) -> int:
"""
Mark workers as offline if they haven't sent heartbeat within timeout.
Returns:
Number of workers marked offline
"""
try:
now = datetime.utcnow()
timeout_threshold = now - timedelta(seconds=DEFAULT_HEARTBEAT_TIMEOUT)
async with get_db_context() as session:
# Find workers that haven't sent heartbeat
stmt = (
update(Worker)
.where(Worker.last_heartbeat < timeout_threshold)
.where(Worker.status != WorkerStatus.OFFLINE.value)
.values(status=WorkerStatus.OFFLINE.value)
)
result = await session.execute(stmt)
await session.commit()
count = result.rowcount
if count > 0:
logger.info(
"Marked workers as offline",
count=count,
)
return count
except Exception as e:
logger.error(
"Failed to cleanup offline workers",
error=str(e),
)
return 0
async def requeue_worker_jobs(self, worker_id: str) -> int:
"""
Requeue jobs from an offline worker.
Args:
worker_id: Worker ID
Returns:
Number of jobs requeued
"""
try:
from backend.queue.producer import _job_queue
requeued_count = 0
for job in _job_queue:
if job.worker_id == worker_id and job.status == JobStatus.RUNNING:
job.status = JobStatus.QUEUED
job.worker_id = None
job.started_at = None
requeued_count += 1
logger.info(
"Requeued job from offline worker",
job_id=str(job.job_id),
worker_id=worker_id,
)
return requeued_count
except Exception as e:
logger.error(
"Failed to requeue worker jobs",
worker_id=worker_id,
error=str(e),
)
return 0
def _build_worker_status_response(self, worker: Worker) -> WorkerStatusResponse:
"""Build worker status response from worker model."""
# Calculate load factor
load_factor = (
worker.active_jobs / worker.max_concurrent_jobs
if worker.max_concurrent_jobs > 0
else 0.0
)
# Calculate uptime
uptime_seconds = int(
(datetime.utcnow() - worker.registered_at).total_seconds()
)
return WorkerStatusResponse(
worker_id=worker.worker_id,
hostname=worker.hostname,
status=WorkerStatus(worker.status),
gpu_count=worker.gpu_count,
gpu_memory_total=worker.gpu_memory_total,
gpu_memory_used=worker.gpu_memory_used,
gpu_usage_percent=worker.gpu_usage_percent,
active_jobs=worker.active_jobs,
max_concurrent_jobs=worker.max_concurrent_jobs,
load_factor=load_factor,
last_heartbeat=worker.last_heartbeat,
registered_at=worker.registered_at,
capabilities=worker.capabilities,
uptime_seconds=uptime_seconds,
)
async def get_available_workers(
self,
gpu_required: int = 0,
) -> List[Worker]:
"""
Get available workers that can accept new jobs.
Args:
gpu_required: Number of GPUs required (0 for CPU-only)
Returns:
List of available workers sorted by load factor
"""
try:
async with get_db_context() as session:
# Get active workers with capacity
stmt = (
select(Worker)
.where(Worker.status.in_([
WorkerStatus.ACTIVE.value,
WorkerStatus.DEGRADED.value,
]))
.where(Worker.active_jobs < Worker.max_concurrent_jobs)
)
if gpu_required > 0:
stmt = stmt.where(Worker.gpu_count >= gpu_required)
result = await session.execute(stmt)
workers = result.scalars().all()
# Sort by load factor (least loaded first)
sorted_workers = sorted(
workers,
key=lambda w: w.active_jobs / w.max_concurrent_jobs
if w.max_concurrent_jobs > 0
else 0.0
)
return list(sorted_workers)
except Exception as e:
logger.error(
"Failed to get available workers",
error=str(e),
)
return []
# Global instance
_worker_registry: Optional[WorkerRegistry] = None
def get_worker_registry() -> WorkerRegistry:
"""Get the global worker registry instance."""
global _worker_registry
if _worker_registry is None:
_worker_registry = WorkerRegistry()
return _worker_registry
__all__ = [
"WorkerRegistry",
"get_worker_registry",
"DEFAULT_HEARTBEAT_INTERVAL",
"DEFAULT_HEARTBEAT_TIMEOUT",
]
|