File size: 9,555 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 | """
Evaluation Worker / Consumer for Job Queue
Provides worker functionality for processing evaluation jobs from the queue.
Handles job execution, checkpointing, and progress reporting.
"""
import asyncio
import os
import socket
import uuid
from dataclasses import dataclass
from datetime import datetime
from typing import Optional
from backend.core.config import settings
from backend.logging.logger import get_logger
from .job_schema import (
JobProgressUpdate,
JobStatus,
EvaluationJob,
)
from .producer import _job_queue, get_job_producer
from .status_tracker import get_status_tracker
logger = get_logger("queue.consumer", component="queue")
@dataclass
class WorkerConfig:
"""Configuration for evaluation worker."""
worker_id: str
max_concurrent_jobs: int = 1
job_timeout_seconds: int = 3600
heartbeat_interval_seconds: int = 30
enable_checkpointing: bool = True
class EvaluationWorker:
"""
Worker that processes evaluation jobs from the queue.
Responsibilities:
- Poll queue for new jobs
- Execute evaluation jobs
- Handle checkpointing
- Report progress
- Manage job lifecycle
"""
def __init__(self, config: Optional[WorkerConfig] = None):
self.config = config or WorkerConfig(
worker_id=f"worker-{socket.gethostname()}-{os.getpid()}",
)
self._status_tracker = get_status_tracker()
self._producer = get_job_producer()
self._active_jobs: dict[uuid.UUID, asyncio.Task] = {}
self._running = False
self._current_job: Optional[EvaluationJob] = None
async def start(self) -> None:
"""Start the worker."""
self._running = True
logger.info(
"Worker started",
worker_id=self.config.worker_id,
max_concurrent=self.config.max_concurrent_jobs,
)
while self._running:
try:
# Poll for jobs
await self._poll_and_process()
# Brief sleep to prevent CPU spinning
await asyncio.sleep(1)
except asyncio.CancelledError:
logger.info("Worker cancelled", worker_id=self.config.worker_id)
break
except Exception as e:
logger.error(
"Worker error",
worker_id=self.config.worker_id,
error=str(e),
)
await asyncio.sleep(5)
# Cancel active jobs
for job_id, task in self._active_jobs.items():
if not task.done():
task.cancel()
logger.info("Cancelled active job", job_id=str(job_id))
logger.info("Worker stopped", worker_id=self.config.worker_id)
async def stop(self) -> None:
"""Stop the worker."""
self._running = False
async def _poll_and_process(self) -> None:
"""Poll queue and process available jobs."""
# Check if we can accept more jobs
if len(self._active_jobs) >= self.config.max_concurrent_jobs:
return
# Find a queued job
for job in _job_queue:
if job.status == JobStatus.QUEUED:
# Check if already being processed
if job.job_id in self._active_jobs:
continue
# Start processing the job
await self._process_job(job)
break
async def _process_job(self, job: EvaluationJob) -> None:
"""Process a single evaluation job."""
job_id_str = str(job.job_id)
try:
# Mark job as started
self._current_job = job
await self._status_tracker.start_job(job.job_id, self.config.worker_id)
job.status = JobStatus.RUNNING
job.started_at = datetime.utcnow()
logger.info(
"Processing job",
job_id=job_id_str,
worker_id=self.config.worker_id,
model=job.model_name,
)
# Create task for async processing
task = asyncio.create_task(self._execute_job(job))
self._active_jobs[job.job_id] = task
# Wait for completion
await task
# Job completed successfully
logger.info(
"Job completed",
job_id=job_id_str,
worker_id=self.config.worker_id,
)
except asyncio.CancelledError:
logger.info("Job cancelled", job_id=job_id_str)
await self._status_tracker.fail_job(
job.job_id,
"Job cancelled by worker",
)
except Exception as e:
logger.error(
"Job failed",
job_id=job_id_str,
error=str(e),
)
await self._status_tracker.fail_job(
job.job_id,
str(e),
)
finally:
# Clean up
self._active_jobs.pop(job.job_id, None)
self._current_job = None
# Remove from queue
_job_queue[:] = [j for j in _job_queue if j.job_id != job.job_id]
async def _execute_job(self, job: EvaluationJob) -> None:
"""Execute the evaluation job."""
# Import orchestrator here to avoid circular imports
from backend.core.orchestrator import (
EvaluationInput,
EvaluationOrchestrator,
)
# Get metadata
metadata = job.metadata or {}
mutation_depth = metadata.get("mutation_depth", 2)
attack_types = metadata.get("attack_types", ["jailbreak"])
max_concurrency = metadata.get("max_concurrency", 4)
# Create evaluation input
eval_input = EvaluationInput(
model_name=job.model_name,
model_version=job.model_version,
dataset_name=job.dataset_name,
dataset_version=job.dataset_version,
mutation_depth=mutation_depth,
attack_types=attack_types,
max_concurrency=max_concurrency,
)
# Create orchestrator
orchestrator = EvaluationOrchestrator()
# Track progress for checkpointing
checkpoint_interval = job.checkpoint_interval
completed_samples = 0
failed_samples = 0
# For checkpointing - we need to hook into the orchestrator
# This is a simplified version - in production, you'd have more sophisticated checkpointing
# Run evaluation
output = await orchestrator.start_run(eval_input)
# Wait for completion (the orchestrator runs asynchronously)
# In a real implementation, we'd need to track progress periodically
# Mark job as complete
await self._status_tracker.complete_job(
job.job_id,
output.composite_score,
output.metrics,
)
job.status = JobStatus.COMPLETED
job.completed_at = datetime.utcnow()
job.composite_score = output.composite_score
job.metrics = output.metrics
job.progress = 100.0
# Update total/completed samples
if output.metrics:
job.total_samples = output.metrics.get("total_samples", 0)
job.completed_samples = output.metrics.get("successful_samples", 0)
job.failed_samples = output.metrics.get("failed_samples", 0)
async def get_current_job_status(self) -> Optional[dict]:
"""Get status of current job being processed."""
if self._current_job is None:
return None
job = self._current_job
return {
"job_id": str(job.job_id),
"status": job.status.value,
"progress": job.progress,
"completed_samples": job.completed_samples,
"total_samples": job.total_samples,
}
def get_worker_status(self) -> dict:
"""Get worker status."""
return {
"worker_id": self.config.worker_id,
"running": self._running,
"active_jobs": len(self._active_jobs),
"max_concurrent_jobs": self.config.max_concurrent_jobs,
}
# Global worker instance
_worker: Optional[EvaluationWorker] = None
def get_worker(config: Optional[WorkerConfig] = None) -> EvaluationWorker:
"""Get the global worker instance."""
global _worker
if _worker is None:
_worker = EvaluationWorker(config)
return _worker
async def start_worker() -> EvaluationWorker:
"""Start the worker and return it."""
worker = get_worker()
asyncio.create_task(worker.start())
return worker
async def stop_worker() -> None:
"""Stop the worker."""
global _worker
if _worker is not None:
await _worker.stop()
_worker = None
__all__ = [
"WorkerConfig",
"EvaluationWorker",
"get_worker",
"start_worker",
"stop_worker",
]
|