Spaces:
Paused
Paused
File size: 25,765 Bytes
98cacb0 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 | import numpy as np
import pandas as pd
import json
import time
import asyncio
from typing import List, Dict, Any, Optional, Union, Callable
from dataclasses import dataclass
from app.core.event_hub import event_hub
from app.deps import get_vector_db
from sentence_transformers import SentenceTransformer
import logging
from datetime import datetime, timedelta
from enum import Enum
from app.core.sre_logging import emit_vector_log
logger = logging.getLogger(__name__)
class VectorStoreEventType(Enum):
"""Pub/sub event types for vector storage lifecycle"""
UPSERT_STARTED = "vector.upsert.started"
UPSERT_COMPLETED = "vector.upsert.completed"
UPSERT_FAILED = "vector.upsert.failed"
SEARCH_QUERIED = "vector.search.queried"
CACHE_WARMED = "vector.cache.warmed"
VSS_FALLBACK = "vector.vss.fallback"
@dataclass
class VectorMetrics:
"""SRE monitoring metrics for vector operations"""
org_id: str
operation: str
duration_ms: float
vector_count: int
redis_latency_ms: float = 0
vss_latency_ms: float = 0
cost_usd: float = 0.0 # Estimated cost per 1000 vectors
error: Optional[str] = None
pipeline_used: bool = False
class VectorService:
"""
π§ Einstein's semantic memory with VSS acceleration
TCP Redis features: True pipelines, pub/sub, zero rate limits
SRE mindset: Metrics, circuit breakers, real-time monitoring
"""
# ====== Singleton model cache ======
_global_model_cache = {}
_model_lock = asyncio.Lock()
_default_model_name = "all-MiniLM-L6-v2"
# ====== SRE: Circuit breaker state ======
_redis_circuit_breaker = {
"failure_count": 0,
"last_failure_time": None,
"is_open": False,
"threshold": 5, # Open after 5 failures
"reset_timeout": 300 # Reset after 5 minutes
}
# ====== Cost tracking ======
# Upstash: $0.20 per 100k commands | TCP Redis: $0
COST_PER_COMMAND_UPSTASH = 0.000002 # $0.20 / 100,000
COST_PER_COMMAND_TCP = 0.0
def __init__(self, org_id: str):
self.org_id = org_id
self.vector_conn = get_vector_db(org_id)
self._model = None
self._metrics_callbacks: List[Callable[[VectorMetrics], None]] = []
# ====== SRE: Metrics collection ======
def add_metrics_callback(self, callback: Callable[[VectorMetrics], None]):
"""Register callback for real-time metrics (e.g., Prometheus)"""
self._metrics_callbacks.append(callback)
def _emit_metrics(self, metrics: VectorMetrics):
"""Notify all registered callbacks (analytics worker, etc.)"""
for callback in self._metrics_callbacks:
try:
callback(metrics)
except Exception as e:
logger.error(f"[METRICS] β Callback failed: {e}")
def _record_operation(self, operation: str, start_time: float,
vector_count: int = 0, **kwargs):
"""Helper to record metrics in SRE format"""
duration_ms = (time.time() - start_time) * 1000
# Estimate cost
cost_per_call = (self.COST_PER_COMMAND_UPSTASH if event_hub.is_rest_api
else self.COST_PER_COMMAND_TCP)
estimated_cost = (vector_count or kwargs.get('commands', 0)) * cost_per_call
metrics = VectorMetrics(
org_id=self.org_id,
operation=operation,
duration_ms=duration_ms,
vector_count=vector_count,
cost_usd=estimated_cost,
pipeline_used=kwargs.get('pipeline_used', False),
redis_latency_ms=kwargs.get('redis_latency', 0),
vss_latency_ms=kwargs.get('vss_latency', 0),
error=kwargs.get('error')
)
self._emit_metrics(metrics)
# Log in SRE format (structured logging)
log_data = {
"event": "vector_operation",
"org_id": self.org_id,
"operation": operation,
"duration_ms": round(duration_ms, 2),
"vector_count": vector_count,
"cost_usd": round(estimated_cost, 6),
"pipeline_used": metrics.pipeline_used,
"redis_type": "upstash" if event_hub.is_rest_api else "tcp"
}
if metrics.error:
log_data["error"] = metrics.error
logger.error(f"[METRICS] {json.dumps(log_data)}")
else:
logger.info(f"[METRICS] {json.dumps(log_data)}")
# ====== SRE: Circuit breaker ======
def _check_circuit_breaker(self) -> bool:
"""Check if Redis circuit is open (too many failures)"""
state = self._redis_circuit_breaker
if not state["is_open"]:
return True
# Check if enough time has passed to try again
if state["last_failure_time"]:
elapsed = time.time() - state["last_failure_time"]
if elapsed > state["reset_timeout"]:
logger.warning("[CIRCUIT] π Closing breaker, trying again...")
state["is_open"] = False
state["failure_count"] = 0
return True
logger.error("[CIRCUIT] π΄ Circuit breaker OPEN, skipping Redis")
return False
def _record_redis_failure(self, error: str):
"""Track failures for circuit breaker"""
state = self._redis_circuit_breaker
state["failure_count"] += 1
state["last_failure_time"] = time.time()
if state["failure_count"] >= state["threshold"]:
state["is_open"] = True
logger.critical(f"[CIRCUIT] π΄ Breaker opened! {state['failure_count']} failures")
def _record_redis_success(self):
"""Reset failure count on success"""
state = self._redis_circuit_breaker
if state["failure_count"] > 0:
logger.info(f"[CIRCUIT] β
Resetting failure count (was {state['failure_count']})")
state["failure_count"] = 0
# ====== Pub/Sub event emission ======
def _publish_vector_event(self, event_type: VectorStoreEventType,
data: Dict[str, Any]):
"""Publish events to Redis pub/sub for real-time monitoring"""
try:
channel = f"vector:events:{self.org_id}"
payload = {
"type": event_type.value,
"timestamp": datetime.utcnow().isoformat(),
"org_id": self.org_id,
"data": data
}
# Fire and forget - don't block on pub/sub
asyncio.create_task(
asyncio.to_thread(
event_hub.publish,
channel,
json.dumps(payload)
)
)
logger.debug(f"[PUBSUB] π‘ Published {event_type.value}")
except Exception as e:
logger.error(f"[PUBSUB] β Failed to publish event: {e}")
# ====== Embedding generation (unchanged core logic) ======
async def _get_or_load_model(self) -> SentenceTransformer:
async with self._model_lock:
if self._default_model_name in self._global_model_cache:
logger.debug(f"[Vector] Using cached model: {self._default_model_name}")
return self._global_model_cache[self._default_model_name]
logger.info(f"[Vector] Loading model: {self._default_model_name}")
model = await asyncio.to_thread(
SentenceTransformer,
self._default_model_name,
device="cpu"
)
self._global_model_cache[self._default_model_name] = model
logger.info(f"[Vector] β
Model cached globally")
return model
def _embed_sync(self, text: str, model: SentenceTransformer) -> List[float]:
if not text or not text.strip():
dim = model.get_sentence_embedding_dimension()
return [0.0] * dim
embedding = model.encode(
text,
convert_to_tensor=False,
normalize_embeddings=True
)
return embedding.tolist()
async def embed(self, text: str) -> List[float]:
if not isinstance(text, str):
raise TypeError(f"Text must be string, got {type(text)}")
model = await self._get_or_load_model()
return await asyncio.to_thread(self._embed_sync, text, model)
async def embed_batch(self, texts: List[str], batch_size: int = 100) -> List[List[float]]:
if not texts:
logger.warning("[Vector] Empty text list")
return []
texts = [t for t in texts if t and t.strip()]
if not texts:
return []
model = await self._get_or_load_model()
embeddings = []
total_batches = (len(texts) + batch_size - 1) // batch_size
for i in range(0, len(texts), batch_size):
batch = texts[i:i + batch_size]
batch_embeddings = await asyncio.to_thread(
lambda batch_texts: [self._embed_sync(t, model) for t in batch_texts],
batch
)
embeddings.extend(batch_embeddings)
if (i // batch_size + 1) % 5 == 0:
logger.debug(f"[Embed] Batch {i//batch_size + 1}/{total_batches}")
emit_vector_log("info", f"β
Generated {len(embeddings)} embeddings",
org_id=self.org_id, vector_count=len(embeddings))
return embeddings
# ====== REFACTORED: TCP Redis pipeline + pub/sub ======
async def _upsert_redis(
self,
embeddings: List[List[float]],
metadata: List[Dict[str, Any]],
namespace: str
) -> bool:
"""
π TCP Redis: True pipeline (0ms latency, zero cost)
Upstash: Sequential with rate limiting
"""
start_time = time.time()
# SRE: Check circuit breaker
if not self._check_circuit_breaker():
logger.error("[UPSERT] π΄ Circuit open, skipping Redis")
self._record_operation(
"upsert_redis", start_time, vector_count=len(embeddings),
error="circuit_breaker_open"
)
return False
# Strategic: Store only hot vectors (100 max)
max_vectors = min(100, len(embeddings))
if len(embeddings) > 100:
logger.info(f"[UPSERT] π Truncating {len(embeddings)} β {max_vectors} vectors for hot cache")
try:
# π― Check pipeline support (TCP vs Upstash)
pipe = event_hub.pipeline()
if pipe and not event_hub.is_rest_api:
# β
**TCP REDIS: True pipeline - 1 command, 10ms total**
for idx in range(max_vectors):
key = f"vector:{namespace}:{idx}:{int(time.time())}"
pipe.setex(key, 86400, json.dumps({
"embedding": embeddings[idx],
"metadata": metadata[idx],
"org_id": self.org_id
}))
# Execute pipeline in thread pool
redis_start = time.time()
await asyncio.to_thread(pipe.execute)
redis_latency = (time.time() - redis_start) * 1000
self._record_redis_success()
self._record_operation(
"upsert_redis", start_time, vector_count=max_vectors,
pipeline_used=True, redis_latency=redis_latency
)
# π **PUB/SUB: Broadcast completion event**
self._publish_vector_event(
VectorStoreEventType.UPSERT_COMPLETED,
{
"namespace": namespace,
"vectors_stored": max_vectors,
"storage": "redis_hot",
"latency_ms": round(redis_latency, 2)
}
)
logger.info(f"[β
VECTOR] Redis PIPELINE: {max_vectors} vectors in {redis_latency:.2f}ms")
return True
else:
# β **UPSTASH: Sequential with rate limiting**
logger.warning("[UPSERT] β οΈ Pipeline not supported, using sequential")
for idx in range(max_vectors):
key = f"vector:{namespace}:{idx}:{int(time.time())}"
redis_start = time.time()
await asyncio.to_thread(
event_hub.setex,
key,
86400,
json.dumps({
"embedding": embeddings[idx],
"metadata": metadata[idx],
"org_id": self.org_id
})
)
redis_latency = (time.time() - redis_start) * 1000
await asyncio.sleep(0.01) # Rate limit
# Emit per-vector event for granular monitoring
self._publish_vector_event(
VectorStoreEventType.UPSERT_COMPLETED,
{
"namespace": namespace,
"vector_id": idx,
"storage": "redis_hot_sequential",
"latency_ms": round(redis_latency, 2)
}
)
logger.info(f"[β
VECTOR] Redis SEQUENTIAL: {max_vectors} vectors (rate-limited)")
return True
except Exception as e:
self._record_redis_failure(str(e))
self._record_operation(
"upsert_redis", start_time, vector_count=max_vectors,
error=str(e)
)
self._publish_vector_event(
VectorStoreEventType.UPSERT_FAILED,
{
"namespace": namespace,
"error": str(e),
"vector_count": max_vectors
}
)
emit_vector_log("error", f"β Redis error: {e}", error=str(e))
return False
# ====== Existing methods (polished with metrics) ======
async def upsert_embeddings(
self,
embeddings: List[List[float]],
metadata: List[Dict[str, Any]],
namespace: str
) -> bool:
"""Store in Redis + VSS with full observability"""
start_time = time.time()
try:
# π **PUB/SUB: Start event**
self._publish_vector_event(
VectorStoreEventType.UPSERT_STARTED,
{
"namespace": namespace,
"total_vectors": len(embeddings),
"hot_vectors": min(100, len(embeddings))
}
)
# Run both stores concurrently
redis_task = self._upsert_redis(embeddings, metadata, namespace)
vss_start = time.time()
vss_task = asyncio.to_thread(self._upsert_vss, embeddings, metadata, namespace)
redis_success, _ = await asyncio.gather(redis_task, vss_task)
vss_latency = (time.time() - vss_start) * 1000
self._record_operation(
"dual_upsert", start_time, vector_count=len(embeddings),
vss_latency=vss_latency
)
if redis_success:
logger.info(f"[β
VECTOR] Dual-store complete: {len(embeddings)} vectors")
else:
logger.warning("[β οΈ VECTOR] Redis failed, VSS succeeded (graceful degradation)")
return True
except Exception as e:
self._record_operation(
"upsert_embeddings", start_time, vector_count=len(embeddings),
error=str(e)
)
logger.error(f"[β VECTOR] Dual upsert failed: {e}")
return False
def _upsert_vss(self, embeddings, metadata, namespace):
"""Store in DuckDB VSS (cold storage)"""
try:
import pandas as pd
records = []
for idx, (emb, meta) in enumerate(zip(embeddings, metadata)):
content = " ".join([str(v) for v in meta.values() if v])[:1000]
records.append({
"id": f"{namespace}:{idx}:{int(time.time())}",
"org_id": self.org_id,
"content": content,
"embedding": emb,
"entity_type": namespace.split(":")[0],
"created_at": datetime.now().isoformat(),
})
if not records:
return
records_df = pd.DataFrame(records)
self.vector_conn.execute("""
INSERT INTO vector_store.embeddings
(id, org_id, content, embedding, entity_type, created_at)
SELECT id, org_id, content,
embedding::FLOAT[384],
entity_type, created_at
FROM records_df
ON CONFLICT (id) DO UPDATE SET
embedding = EXCLUDED.embedding,
content = EXCLUDED.content,
created_at = EXCLUDED.created_at
""")
logger.info(f"[β
VECTOR] VSS: Stored {len(records_df)} vectors")
except Exception as e:
logger.error(f"[β VECTOR] VSS error: {e}", exc_info=True)
async def semantic_search(self, query_embedding: List[float],
top_k: int = 10, min_score: float = 0.7,
days_back: int = 30) -> List[Dict]:
"""
π Search with full observability and pub/sub events
"""
start_time = time.time()
try:
# Try Redis hot cache first
redis_start = time.time()
redis_results = await self._search_redis(query_embedding, top_k, min_score)
redis_latency = (time.time() - redis_start) * 1000
if redis_results:
self._record_operation(
"search_redis", start_time, vector_count=len(redis_results),
redis_latency=redis_latency
)
self._publish_vector_event(
VectorStoreEventType.SEARCH_QUERIED,
{
"source": "redis",
"results": len(redis_results),
"latency_ms": round(redis_latency, 2),
"fallback_to_vss": False
}
)
return redis_results
# Fallback to VSS
logger.info("[SEARCH] Cache miss, querying VSS...")
vss_start = time.time()
vss_results = self._search_vss(query_embedding, top_k, min_score, days_back)
vss_latency = (time.time() - vss_start) * 1000
self._record_operation(
"search_vss", start_time, vector_count=len(vss_results),
vss_latency=vss_latency
)
self._publish_vector_event(
VectorStoreEventType.VSS_FALLBACK,
{
"source": "vss",
"results": len(vss_results),
"latency_ms": round(vss_latency, 2),
"cache_warm_triggered": len(vss_results) > 0
}
)
# Warm cache with VSS results
if vss_results:
asyncio.create_task(self._warm_cache(vss_results))
return vss_results
except Exception as e:
self._record_operation(
"semantic_search", start_time, vector_count=0,
error=str(e)
)
logger.error(f"[SEARCH] Error: {e}")
return []
async def _search_redis(self, query_emb: List[float], top_k: int, min_score: float) -> List[Dict]:
"""Search Redis with circuit breaker protection"""
if not self._check_circuit_breaker():
logger.warning("[SEARCH] π΄ Circuit open, skipping Redis")
return []
try:
pattern = f"vector:{self.org_id}:*"
keys = await asyncio.to_thread(event_hub.keys, pattern)
keys = keys[:1000] # Limit scan
results = []
query_np = np.array(query_emb, dtype=np.float32)
for key in keys:
data = await asyncio.to_thread(event_hub.get_key, key)
if not data:
continue
try:
vec_data = json.loads(data)
emb = np.array(vec_data["embedding"], dtype=np.float32)
similarity = np.dot(query_np, emb) / (
np.linalg.norm(query_np) * np.linalg.norm(emb) + 1e-9
)
if similarity >= min_score:
results.append({
"score": float(similarity),
"metadata": vec_data["metadata"],
"source": "redis"
})
except Exception:
continue
self._record_redis_success()
return sorted(results, key=lambda x: x["score"], reverse=True)[:top_k]
except Exception as e:
self._record_redis_failure(str(e))
logger.error(f"[SEARCH] Redis error: {e}")
return []
def _search_vss(self, query_emb: List[float], top_k: int, min_score: float, days_back: int) -> List[Dict]:
"""Search DuckDB VSS"""
try:
cutoff = (datetime.now() - timedelta(days=days_back)).isoformat()
results = self.vector_conn.execute("""
SELECT id, content, embedding, created_at,
array_cosine_similarity(embedding, ?::FLOAT[384]) as similarity
FROM vector_store.embeddings
WHERE org_id = ?
AND entity_type = ?
AND created_at >= ?
AND similarity >= ?
ORDER BY similarity DESC
LIMIT ?
""", [query_emb, self.org_id, "sales", cutoff, min_score, top_k]).fetchall()
return [{
"score": float(r[4]),
"metadata": {
"id": r[0],
"content": r[1],
"created_at": r[3].isoformat() if r[3] else None
},
"source": "vss"
} for r in results]
except Exception as e:
logger.error(f"[SEARCH] VSS error: {e}")
return []
async def _warm_cache(self, results: List[Dict]):
"""Warm Redis with VSS results (non-blocking)"""
try:
pipe = event_hub.pipeline()
if not pipe:
return # Can't warm cache if no pipeline
for r in results[:10]: # Warm top 10 only
pipe.setex(
f"vector:warm:{int(time.time())}:{r['metadata']['id']}",
86400,
json.dumps(r)
)
await asyncio.to_thread(pipe.execute)
logger.info(f"[WARM] π₯ Cached {len(results[:10])} vectors to Redis")
self._publish_vector_event(
VectorStoreEventType.CACHE_WARMED,
{
"vectors_warmed": len(results[:10]),
"source": "vss_to_redis"
}
)
except Exception as e:
logger.error(f"[WARM] β Failed: {e}")
# ---- Background Cleanup Worker (with SRE metrics) ----
def cleanup_expired_vectors():
"""π§Ή Daily cleanup with monitoring"""
try:
start_time = time.time()
vector_conn = get_vector_db()
deleted = vector_conn.execute("""
DELETE FROM vector_store.embeddings
WHERE created_at <= (CURRENT_TIMESTAMP - INTERVAL 30 DAY)
RETURNING COUNT(*) as count
""").fetchone()
duration_ms = (time.time() - start_time) * 1000
if deleted and deleted[0] > 0:
logger.info(f"[CLEANUP] ποΈ Deleted {deleted[0]} vectors in {duration_ms:.2f}ms")
# Publish cleanup event
asyncio.create_task(
event_hub.publish(
"vector:cleanup:events",
json.dumps({
"type": "cleanup.completed",
"deleted_count": deleted[0] if deleted else 0,
"duration_ms": round(duration_ms, 2)
})
)
)
except Exception as e:
logger.error(f"[CLEANUP] β Error: {e}", exc_info=True) |