Spaces:
Runtime error
Runtime error
File size: 15,226 Bytes
c1dc8ac | 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 | """Distributed processing module for scaling to 5M+ samples using Ray."""
import logging
import time
import hashlib
from typing import List, Dict, Any, Optional, Iterator
from pathlib import Path
import numpy as np
from PIL import Image
import io
import pickle
from concurrent.futures import ThreadPoolExecutor, as_completed
try:
import ray
RAY_AVAILABLE = True
except ImportError:
RAY_AVAILABLE = False
ray = None
from .base import PipelineStage, CharacterAttributes, ProcessingResult
from .input_loader import DatasetItem
logger = logging.getLogger(__name__)
if RAY_AVAILABLE:
@ray.remote
class DistributedWorker:
"""Ray actor for distributed character attribute extraction."""
def __init__(self, pipeline_config: Dict[str, Any]):
"""Initialize worker with pipeline configuration."""
self.pipeline_config = pipeline_config
self.pipeline = None
self._initialize_pipeline()
def _initialize_pipeline(self):
"""Initialize the character extraction pipeline on worker."""
try:
from character_pipeline import create_pipeline
self.pipeline = create_pipeline(self.pipeline_config)
logger.info(f"Worker {ray.get_runtime_context().get_worker_id()} initialized")
except Exception as e:
logger.error(f"Failed to initialize pipeline on worker: {e}")
raise
def process_batch(self, items: List[DatasetItem]) -> List[ProcessingResult]:
"""Process a batch of items on this worker."""
results = []
for item in items:
try:
start_time = time.time()
result = self.pipeline.extract_from_image(item.image_path)
processing_time = time.time() - start_time
results.append(ProcessingResult(
item_id=item.item_id,
attributes=result,
success=True,
processing_time=processing_time
))
except Exception as e:
logger.warning(f"Failed to process {item.item_id}: {e}")
results.append(ProcessingResult(
item_id=item.item_id,
attributes=CharacterAttributes(),
success=False,
error_message=str(e)
))
return results
def get_worker_stats(self) -> Dict[str, Any]:
"""Get worker statistics and health status."""
return {
'worker_id': ray.get_runtime_context().get_worker_id(),
'node_id': ray.get_runtime_context().get_node_id(),
'memory_usage': ray.cluster_resources().get('memory', 0),
'cpu_usage': ray.cluster_resources().get('CPU', 0)
}
else:
class DistributedWorker:
"""Fallback worker when Ray is not available."""
def __init__(self, pipeline_config: Dict[str, Any]):
raise ImportError("Ray is not available. Install with: pip install ray[default]")
class DistributedProcessor(PipelineStage):
"""Distributed processor for handling large-scale character extraction."""
def __init__(self, config: Optional[Dict[str, Any]] = None):
super().__init__("DistributedProcessor", config)
# Configuration
self.num_workers = config.get('num_workers', 4) if config else 4
self.batch_size = config.get('batch_size', 32) if config else 32
self.max_retries = config.get('max_retries', 3) if config else 3
self.checkpoint_interval = config.get('checkpoint_interval', 1000) if config else 1000
# Ray configuration
self.ray_config = config.get('ray_config', {}) if config else {}
# State
self.workers = []
self.is_initialized = False
def initialize_cluster(self) -> bool:
"""Initialize Ray cluster and workers."""
if not RAY_AVAILABLE:
self.logger.error("Ray is not available. Install with: pip install ray[default]")
return False
try:
if not ray.is_initialized():
ray.init(**self.ray_config)
# Create distributed workers
pipeline_config = self.config.get('pipeline_config', {})
self.workers = [
DistributedWorker.remote(pipeline_config)
for _ in range(self.num_workers)
]
# Test worker initialization
health_checks = ray.get([worker.get_worker_stats.remote() for worker in self.workers])
self.logger.info(f"Initialized {len(health_checks)} workers successfully")
self.is_initialized = True
return True
except Exception as e:
self.logger.error(f"Failed to initialize Ray cluster: {e}")
return False
def process_large_dataset(self, dataset_path: str, output_path: str) -> Dict[str, Any]:
"""Process large dataset with distributed workers."""
if not self.is_initialized:
if not self.initialize_cluster():
raise RuntimeError("Failed to initialize distributed cluster")
start_time = time.time()
total_processed = 0
total_successful = 0
checkpoints = []
try:
# Load dataset in streaming fashion
dataset_stream = self._create_dataset_stream(dataset_path)
# Process in distributed batches
batch_futures = []
current_batch = []
for item in dataset_stream:
current_batch.append(item)
if len(current_batch) >= self.batch_size:
# Distribute batch to available worker
worker = self.workers[len(batch_futures) % len(self.workers)]
future = worker.process_batch.remote(current_batch.copy())
batch_futures.append(future)
current_batch = []
# Process completed batches
if len(batch_futures) >= self.num_workers * 2:
completed_results = self._collect_completed_batches(batch_futures)
total_processed += sum(len(results) for results in completed_results)
total_successful += sum(
sum(1 for r in results if r.success)
for results in completed_results
)
# Save checkpoint
if total_processed % self.checkpoint_interval == 0:
checkpoint = self._save_checkpoint(completed_results, output_path, total_processed)
checkpoints.append(checkpoint)
self.logger.info(f"Checkpoint saved: {total_processed} items processed")
# Process remaining batch
if current_batch:
worker = self.workers[0]
future = worker.process_batch.remote(current_batch)
batch_futures.append(future)
# Collect all remaining results
if batch_futures:
remaining_results = ray.get(batch_futures)
total_processed += sum(len(results) for results in remaining_results)
total_successful += sum(
sum(1 for r in results if r.success)
for results in remaining_results
)
# Final checkpoint
final_checkpoint = self._save_checkpoint(remaining_results, output_path, total_processed)
checkpoints.append(final_checkpoint)
processing_time = time.time() - start_time
return {
'total_processed': total_processed,
'total_successful': total_successful,
'success_rate': total_successful / total_processed if total_processed > 0 else 0,
'processing_time': processing_time,
'throughput': total_processed / processing_time if processing_time > 0 else 0,
'checkpoints': checkpoints,
'num_workers': len(self.workers)
}
except Exception as e:
self.logger.error(f"Distributed processing failed: {e}")
raise
def _create_dataset_stream(self, dataset_path: str) -> Iterator[DatasetItem]:
"""Create streaming iterator for large dataset."""
dataset_path = Path(dataset_path)
if dataset_path.is_file():
# Single file - assume it's a list of image paths
with open(dataset_path, 'r') as f:
for line_num, line in enumerate(f):
image_path = line.strip()
if image_path and Path(image_path).exists():
yield DatasetItem(
item_id=f"item_{line_num}",
image_path=image_path
)
else:
# Directory - iterate through image files
image_extensions = {'.jpg', '.jpeg', '.png', '.bmp', '.tiff'}
for idx, image_path in enumerate(dataset_path.rglob('*')):
if image_path.suffix.lower() in image_extensions:
yield DatasetItem(
item_id=f"item_{idx}",
image_path=str(image_path)
)
def _collect_completed_batches(self, batch_futures: List) -> List[List[ProcessingResult]]:
"""Collect completed batch results and remove from futures list."""
completed_results = []
ready_futures, batch_futures[:] = ray.wait(batch_futures, num_returns=len(batch_futures), timeout=0)
if ready_futures:
completed_results = ray.get(ready_futures)
return completed_results
def _save_checkpoint(self, results: List[List[ProcessingResult]], output_path: str, total_processed: int) -> str:
"""Save processing checkpoint."""
checkpoint_path = f"{output_path}_checkpoint_{total_processed}.pkl"
# Flatten results
flat_results = []
for batch_results in results:
flat_results.extend(batch_results)
with open(checkpoint_path, 'wb') as f:
pickle.dump(flat_results, f)
return checkpoint_path
def estimate_scalability(self, sample_size: int = 1000) -> Dict[str, Any]:
"""Estimate processing capabilities for 5M scale."""
if not self.is_initialized:
if not self.initialize_cluster():
raise RuntimeError("Failed to initialize distributed cluster")
# Create sample dataset
sample_items = [
DatasetItem(item_id=f"sample_{i}", image_path="sample_image.jpg")
for i in range(sample_size)
]
# Measure processing time
start_time = time.time()
# Distribute sample processing
batch_size = min(self.batch_size, sample_size // self.num_workers)
futures = []
for i in range(0, sample_size, batch_size):
batch = sample_items[i:i + batch_size]
worker = self.workers[i // batch_size % len(self.workers)]
future = worker.process_batch.remote(batch)
futures.append(future)
# Collect results
results = ray.get(futures)
processing_time = time.time() - start_time
# Calculate metrics
total_items = sum(len(batch_results) for batch_results in results)
throughput = total_items / processing_time if processing_time > 0 else 0
# Estimate 5M scale
estimated_5m_time = 5_000_000 / throughput if throughput > 0 else float('inf')
estimated_5m_hours = estimated_5m_time / 3600
# Memory estimation
cluster_resources = ray.cluster_resources()
total_memory_gb = cluster_resources.get('memory', 0) / (1024**3)
total_cpus = cluster_resources.get('CPU', 0)
return {
'sample_size': sample_size,
'processing_time': processing_time,
'throughput_per_second': throughput,
'estimated_5m_processing_time_hours': estimated_5m_hours,
'estimated_5m_processing_time_days': estimated_5m_hours / 24,
'cluster_resources': {
'total_memory_gb': total_memory_gb,
'total_cpus': total_cpus,
'num_workers': len(self.workers)
},
'scalability_recommendations': self._generate_scalability_recommendations(throughput, total_memory_gb, total_cpus)
}
def _generate_scalability_recommendations(self, throughput: float, memory_gb: float, cpus: int) -> List[str]:
"""Generate recommendations for scaling to 5M samples."""
recommendations = []
if throughput < 100: # Less than 100 items/second
recommendations.append("Consider increasing worker count or optimizing pipeline")
if memory_gb < 32:
recommendations.append("Increase cluster memory for better caching and model loading")
if cpus < 16:
recommendations.append("Add more CPU cores for parallel processing")
recommendations.extend([
"Implement data sharding across multiple storage systems",
"Use Redis cluster for distributed caching",
"Consider GPU acceleration for CLIP model inference",
"Implement progressive loading to reduce memory footprint",
"Use Apache Beam or Spark for even larger scale processing"
])
return recommendations
def shutdown(self):
"""Shutdown distributed cluster."""
if self.is_initialized and RAY_AVAILABLE:
ray.shutdown()
self.is_initialized = False
self.logger.info("Distributed cluster shutdown complete")
def process(self, input_data: Any) -> Any:
"""Process input using distributed workers."""
if not RAY_AVAILABLE:
raise RuntimeError("Ray is not available. Install with: pip install ray[default]")
if isinstance(input_data, str):
# Assume it's a dataset path
return self.process_large_dataset(input_data, "distributed_output")
else:
raise ValueError("DistributedProcessor expects dataset path as input")
def validate_input(self, input_data: Any) -> bool:
"""Validate input data."""
return isinstance(input_data, str) and Path(input_data).exists() |