File size: 46,544 Bytes
dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 7c8b011 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 7c8b011 dbb04e4 7c8b011 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 c3a3710 dbb04e4 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 | """
Tiered Memory Management (Phase 3.5+)
=====================================
Manages memory lifecycle across HOT, WARM, and COLD tiers based on Long-Term Potentiation (LTP).
Tiers:
- HOT (RAM): Fast access, limited capacity. Stores most relevant memories.
- WARM (Qdrant/Disk): Larger capacity, slightly slower access.
- COLD (Archive): Unlimited capacity, slow access. Compressed JSONL.
Logic:
- New memories start in HOT.
- `consolidate()` moves memories between tiers based on LTP strength and hysteresis.
- Promote: WARM -> HOT if `ltp > threshold + delta`
- Demote: HOT -> WARM if `ltp < threshold - delta`
- Archive: WARM -> COLD if `ltp < archive_threshold` (or age)
All vectors use BinaryHDV (packed uint8 arrays).
"""
import gzip
import json
from datetime import datetime, timezone
from itertools import islice
from pathlib import Path
from typing import Dict, List, Optional, Tuple, TYPE_CHECKING, Any
if TYPE_CHECKING:
from .qdrant_store import QdrantStore
import asyncio
import functools
import numpy as np
from loguru import logger
from .binary_hdv import BinaryHDV
from .config import HAIMConfig, get_config
from .node import MemoryNode
from .exceptions import (
MnemoCoreError,
StorageError,
CircuitOpenError,
DataCorruptionError,
)
try:
import faiss
FAISS_AVAILABLE = True
except ImportError:
FAISS_AVAILABLE = False
# Phase 4.0: HNSW index manager (replaces raw FAISS management)
try:
from .hnsw_index import HNSWIndexManager
HNSW_AVAILABLE = True
except ImportError:
HNSW_AVAILABLE = False
class TierManager:
"""
Manages memory storage across tiered hierarchy.
Uses BinaryHDV exclusively for efficient storage and computation.
"""
def __init__(
self,
config: Optional[HAIMConfig] = None,
qdrant_store: Optional["QdrantStore"] = None,
):
"""
Initialize TierManager with optional dependency injection.
Args:
config: Configuration object. If None, uses global get_config().
qdrant_store: QdrantStore instance. If None, will not use Qdrant.
"""
self.config = config or get_config()
# Initialization guard
self._initialized: bool = False
# Async lock - created eagerly; asyncio.Lock() is safe to construct
# outside a running loop in Python 3.10+ (loop binding is deferred).
self.lock: asyncio.Lock = asyncio.Lock()
# HOT Tier: In-memory dictionary
self.hot: Dict[str, MemoryNode] = {}
# Phase 13.2: O(1) inverted index for episodic chain – previous_id → node_id.
# Maintained in sync with self.hot so get_next_in_chain() avoids O(N) scans.
self._next_chain: Dict[str, str] = {}
# WARM Tier: Qdrant (injected) or fallback to filesystem
self.qdrant = qdrant_store
self.use_qdrant = qdrant_store is not None
self.warm_path = None
if not self.use_qdrant:
self.warm_path = Path(self.config.paths.warm_mmap_dir)
self.warm_path.mkdir(parents=True, exist_ok=True)
# COLD Tier path
self.cold_path = Path(self.config.paths.cold_archive_dir)
self.cold_path.mkdir(parents=True, exist_ok=True)
# Phase 4.0: HNSW/FAISS Index for HOT Tier (Binary)
# HNSWIndexManager auto-selects Flat (small N) or HNSW (large N)
cfg = self.config
if HNSW_AVAILABLE:
self._hnsw = HNSWIndexManager(
dimension=cfg.dimensionality,
m=getattr(cfg.qdrant, "hnsw_m", 32),
ef_construction=getattr(cfg.qdrant, "hnsw_ef_construct", 200),
ef_search=64,
)
logger.info(
f"Phase 4.0 HNSWIndexManager initialised for HOT tier "
f"(dim={cfg.dimensionality}, M={getattr(cfg.qdrant, 'hnsw_m', 32)})"
)
else:
self._hnsw = None
# Legacy FAISS fields kept for backward-compat (unused when HNSW available)
self.faiss_index = None
self.faiss_id_map: Dict[int, str] = {}
self.node_id_to_faiss_id: Dict[str, int] = {}
self._next_faiss_id = 1
if not HNSW_AVAILABLE and FAISS_AVAILABLE:
self._init_faiss()
def _init_faiss(self):
"""Initialize FAISS binary index (legacy path, used when hnsw_index unavailable)."""
dim = self.config.dimensionality
base_index = faiss.IndexBinaryFlat(dim)
self.faiss_index = faiss.IndexBinaryIDMap(base_index)
logger.info(f"Initialized FAISS flat binary index for HOT tier (dim={dim})")
async def get_hot_snapshot(self) -> List[MemoryNode]:
"""Return a snapshot of values in HOT tier safely."""
async with self.lock:
return list(self.hot.values())
async def get_hot_recent(self, n: int) -> List[MemoryNode]:
"""Get the most recent n memories from HOT tier efficiently."""
async with self.lock:
try:
recent_keys = list(islice(reversed(self.hot), n))
nodes = [self.hot[k] for k in reversed(recent_keys)]
return nodes
except Exception:
all_nodes = list(self.hot.values())
return all_nodes[-n:]
async def initialize(self):
"""Async initialization for Qdrant collections."""
if self._initialized:
return
if self.use_qdrant and self.qdrant:
try:
await self.qdrant.ensure_collections()
except Exception as e:
logger.error(f"Failed to ensure Qdrant collections: {e}")
self.use_qdrant = False
self.warm_path = Path(self.config.paths.warm_mmap_dir)
self.warm_path.mkdir(parents=True, exist_ok=True)
self._initialized = True
async def _run_in_thread(self, func, *args, **kwargs):
"""Run blocking function in thread pool."""
loop = asyncio.get_running_loop()
return await loop.run_in_executor(None, functools.partial(func, *args, **kwargs))
async def add_memory(self, node: MemoryNode):
"""Add a new memory node. New memories are always HOT initially."""
node.tier = "hot"
# Delta 67.4: Ensure mutual exclusion.
# If the node exists in WARM, remove it before adding to HOT.
await self._delete_from_warm(node.id)
# Phase 1: Add to HOT tier under lock (no I/O)
victim_to_evict = None
async with self.lock:
self.hot[node.id] = node
self._add_to_faiss(node)
# Maintain inverted chain index (Fix 7: O(1) get_next_in_chain)
if node.previous_id:
self._next_chain[node.previous_id] = node.id
# Check if we need to evict - decide under lock, execute outside
if len(self.hot) > self.config.tiers_hot.max_memories:
# Use unified eviction logic, protecting the new node
victim_to_evict = self._prepare_eviction_from_hot(exclude_node_id=node.id)
# Phase 2: Perform I/O outside lock
if victim_to_evict:
await self._save_to_warm(victim_to_evict)
def _add_to_faiss(self, node: MemoryNode):
"""Add node to the ANN index (HNSW preferred, legacy flat as fallback)."""
# Phase 4.0: delegate to HNSWIndexManager
if self._hnsw is not None:
self._hnsw.add(node.id, node.hdv.data)
return
# Legacy FAISS flat path
if not self.faiss_index:
return
try:
fid = self._next_faiss_id
self._next_faiss_id += 1
vec = np.expand_dims(node.hdv.data, axis=0)
ids = np.array([fid], dtype='int64')
self.faiss_index.add_with_ids(vec, ids)
self.faiss_id_map[fid] = node.id
self.node_id_to_faiss_id[node.id] = fid
except Exception as e:
logger.error(f"Failed to add node {node.id} to FAISS: {e}")
async def get_memory(self, node_id: str) -> Optional[MemoryNode]:
"""Retrieve memory by ID from any tier."""
# Check HOT
demote_candidate = None
result_node = None
async with self.lock:
if node_id in self.hot:
node = self.hot[node_id]
node.access()
# check if node should be demoted
if self._should_demote(node):
# Node will be demoted, mark it as warm immediately to prevent TOCTOU
# This ensures concurrent readers see the correct upcoming state
node.tier = "warm"
demote_candidate = node
result_node = node
# If demotion is needed, save to WARM first, then remove from HOT
# This occurs outside the lock to allow concurrency, but the node
# is already marked as "warm" (graceful degradation if save fails)
if demote_candidate:
logger.info(f"Demoting {demote_candidate.id} to WARM (LTP: {demote_candidate.ltp_strength:.4f})")
# Step 1: Save to WARM (I/O outside lock)
await self._save_to_warm(demote_candidate)
# Step 2: Remove from HOT (under lock)
async with self.lock:
# Double-check: it might have been accessed again or removed
if demote_candidate.id in self.hot:
del self.hot[demote_candidate.id]
self._remove_from_faiss(demote_candidate.id)
# node.tier is already "warm"
if result_node:
return result_node
# Check WARM (Qdrant or Disk)
warm_node = await self._load_from_warm(node_id)
if warm_node:
warm_node.tier = "warm"
warm_node.access()
# Check promotion (pure function, no lock needed)
if self._should_promote(warm_node):
await self._promote_to_hot(warm_node)
return warm_node
# Fix 1: Fall back to COLD tier (read-only archive scan).
cold_node = await self._load_from_cold(node_id)
if cold_node:
logger.debug(f"Retrieved {node_id} from COLD tier.")
return cold_node
async def get_memories_batch(self, node_ids: List[str]) -> List[Optional[MemoryNode]]:
"""
Retrieve multiple memories concurrently.
Preserves input order and returns None for missing/error cases.
"""
if not node_ids:
return []
unique_ids = list(dict.fromkeys(node_ids))
tasks = [self.get_memory(nid) for nid in unique_ids]
raw_results = await asyncio.gather(*tasks, return_exceptions=True)
result_by_id: Dict[str, Optional[MemoryNode]] = {}
for nid, result in zip(unique_ids, raw_results):
if isinstance(result, Exception):
logger.error(f"Batch get_memory failed for {nid}: {result}")
result_by_id[nid] = None
else:
result_by_id[nid] = result
return [result_by_id.get(nid) for nid in node_ids]
async def anticipate(self, node_ids: List[str]) -> None:
"""
Phase 13.2: Anticipatory Memory
Pre-loads specific nodes into the HOT active tier (working memory).
This forces the nodes out of WARM/COLD and into RAM for near-zero latency retrieval.
"""
for nid in set(node_ids):
# Check if already in HOT
in_hot = False
async with self.lock:
if nid in self.hot:
in_hot = True
if not in_hot:
# Load from WARM
node = await self._load_from_warm(nid)
if node:
# Force promote to HOT
await self._promote_to_hot(node)
async def delete_memory(self, node_id: str):
"""Robust delete from all tiers."""
async with self.lock:
if node_id in self.hot:
_node = self.hot[node_id]
# Maintain inverted chain index before removal
if _node.previous_id:
self._next_chain.pop(_node.previous_id, None)
del self.hot[node_id]
self._remove_from_faiss(node_id)
logger.debug(f"Deleted {node_id} from HOT")
await self._delete_from_warm(node_id)
def _remove_from_faiss(self, node_id: str):
"""Remove node from the ANN index (HNSW preferred, legacy flat as fallback)."""
# Phase 4.0: delegate to HNSWIndexManager
if self._hnsw is not None:
self._hnsw.remove(node_id)
return
# Legacy FAISS flat path
if not self.faiss_index:
return
fid = self.node_id_to_faiss_id.get(node_id)
if fid is not None:
try:
ids_to_remove = np.array([fid], dtype='int64')
self.faiss_index.remove_ids(ids_to_remove)
del self.faiss_id_map[fid]
del self.node_id_to_faiss_id[node_id]
except Exception as e:
logger.error(f"Failed to remove node {node_id} from FAISS: {e}")
async def _delete_from_warm(self, node_id: str) -> bool:
"""
Internal helper to delete from warm and return if found.
Returns:
True if deleted, False otherwise.
Note:
Errors are logged but don't propagate to allow graceful degradation.
"""
deleted = False
if self.use_qdrant:
try:
await self.qdrant.delete(self.config.qdrant.collection_warm, [node_id])
deleted = True
except CircuitOpenError as e:
logger.warning(f"Cannot delete {node_id}: {e}")
except StorageError as e:
logger.warning(f"Storage error deleting {node_id}: {e}")
except Exception as e:
logger.warning(f"Qdrant delete failed for {node_id}: {e}")
# Filesystem fallback
if hasattr(self, 'warm_path') and self.warm_path:
def _fs_delete():
d = False
npy = self.warm_path / f"{node_id}.npy"
jsn = self.warm_path / f"{node_id}.json"
if npy.exists() or jsn.exists():
try:
if npy.exists():
npy.unlink()
if jsn.exists():
jsn.unlink()
d = True
except OSError:
pass
return d
if await self._run_in_thread(_fs_delete):
deleted = True
logger.debug(f"Deleted {node_id} from WARM (FS)")
return deleted
def _prepare_eviction_from_hot(self, exclude_node_id: Optional[str] = None) -> Optional[MemoryNode]:
"""
Prepare eviction by finding and removing the victim from HOT.
Returns the victim node to be saved to WARM (caller must do I/O outside lock).
Returns None if HOT tier is empty or no valid victim found.
Args:
exclude_node_id: Optional ID to protect from eviction (e.g., the node just added).
"""
if not self.hot:
return None
candidates = self.hot.values()
if exclude_node_id:
candidates = [n for n in candidates if n.id != exclude_node_id]
if not candidates:
return None
victim = min(candidates, key=lambda n: n.ltp_strength)
logger.info(f"Evicting {victim.id} from HOT to WARM (LTP: {victim.ltp_strength:.4f})")
# Remove from HOT structure (maintain inverted chain index)
if victim.previous_id:
self._next_chain.pop(victim.previous_id, None)
del self.hot[victim.id]
self._remove_from_faiss(victim.id)
# Mark as warm for state consistency
victim.tier = "warm"
return victim
async def _save_to_warm(self, node: MemoryNode):
"""
Save memory node to WARM tier (Qdrant or fallback).
Raises:
StorageError: If save fails (to allow caller to handle appropriately).
Note:
Falls back to filesystem if Qdrant save fails.
"""
if self.use_qdrant:
try:
from qdrant_client import models
# Unpack binary vector for Qdrant storage (Bipolar Phase 4.5)
bits = np.unpackbits(node.hdv.data)
vector = (bits.astype(float) * 2.0 - 1.0).tolist()
point = models.PointStruct(
id=node.id,
vector=vector,
payload={
"content": node.content,
"metadata": node.metadata,
"created_at": node.created_at.isoformat(),
"last_accessed": node.last_accessed.isoformat(),
"ltp_strength": node.ltp_strength,
"access_count": node.access_count,
"epistemic_value": node.epistemic_value,
"pragmatic_value": node.pragmatic_value,
"dimension": node.hdv.dimension,
"hdv_type": "binary",
# Phase 4.3: Temporal metadata for time-based indexing
"unix_timestamp": node.unix_timestamp,
"iso_date": node.iso_date,
"previous_id": node.previous_id,
}
)
await self.qdrant.upsert(
collection=self.config.qdrant.collection_warm,
points=[point]
)
return
except CircuitOpenError as e:
logger.warning(f"Cannot save {node.id} to Qdrant (circuit open), falling back to FS: {e}")
# Fall through to filesystem fallback
except StorageError as e:
logger.error(f"Storage error saving {node.id} to Qdrant, falling back to FS: {e}")
# Fall through to filesystem fallback
except Exception as e:
logger.error(f"Failed to save {node.id} to Qdrant, falling back to FS: {e}")
# Fall through to filesystem fallback
# Fallback (File System)
if not hasattr(self, 'warm_path') or not self.warm_path:
self.warm_path = Path(self.config.paths.warm_mmap_dir)
self.warm_path.mkdir(parents=True, exist_ok=True)
def _fs_save():
hdv_path = self.warm_path / f"{node.id}.npy"
np.save(hdv_path, node.hdv.data)
meta_path = self.warm_path / f"{node.id}.json"
data = {
"id": node.id,
"content": node.content,
"metadata": node.metadata,
"created_at": node.created_at.isoformat(),
"last_accessed": node.last_accessed.isoformat(),
"ltp_strength": node.ltp_strength,
"access_count": node.access_count,
"tier": "warm",
"epistemic_value": node.epistemic_value,
"pragmatic_value": node.pragmatic_value,
"hdv_type": "binary",
"dimension": node.hdv.dimension,
# Phase 4.3: Temporal metadata
"unix_timestamp": node.unix_timestamp,
"iso_date": node.iso_date,
"previous_id": node.previous_id,
}
with open(meta_path, "w") as f:
json.dump(data, f)
await self._run_in_thread(_fs_save)
async def _load_from_warm(self, node_id: str) -> Optional[MemoryNode]:
"""
Load memory node from WARM tier.
Returns:
MemoryNode if found, None if not found.
Note:
Returns None for both "not found" and storage errors to maintain
backward compatibility. Storage errors are logged but don't propagate
to avoid disrupting higher-level operations.
"""
if self.use_qdrant:
try:
record = await self.qdrant.get_point(
self.config.qdrant.collection_warm, node_id
)
if record:
payload = record.payload
vec_data = record.vector
try:
# Reconstruct BinaryHDV
arr = np.array(vec_data) > 0.5
packed = np.packbits(arr.astype(np.uint8))
hdv = BinaryHDV(data=packed, dimension=payload["dimension"])
except (ValueError, KeyError, TypeError) as e:
logger.error(f"Data corruption for {node_id} in Qdrant: {e}")
return None
return MemoryNode(
id=payload.get("id", node_id),
hdv=hdv,
content=payload["content"],
metadata=payload["metadata"],
created_at=datetime.fromisoformat(payload["created_at"]),
last_accessed=datetime.fromisoformat(payload["last_accessed"]),
tier="warm",
access_count=payload.get("access_count", 0),
ltp_strength=payload.get("ltp_strength", 0.0),
epistemic_value=payload.get("epistemic_value", 0.0),
pragmatic_value=payload.get("pragmatic_value", 0.0),
previous_id=payload.get("previous_id"), # Phase 4.3
)
return None # Not found
except CircuitOpenError as e:
logger.warning(f"Cannot load {node_id}: {e}")
return None
except StorageError as e:
logger.error(f"Storage error loading {node_id}: {e}")
return None
except Exception as e:
logger.error(f"Unexpected error loading {node_id} from Qdrant: {e}")
return None
# Fallback (File System)
if hasattr(self, 'warm_path') and self.warm_path:
def _fs_load():
hdv_path = self.warm_path / f"{node_id}.npy"
meta_path = self.warm_path / f"{node_id}.json"
if not hdv_path.exists() or not meta_path.exists():
return None # Not found
try:
with open(meta_path, "r") as f:
data = json.load(f)
hdv_data = np.load(hdv_path)
hdv = BinaryHDV(data=hdv_data, dimension=data["dimension"])
return MemoryNode(
id=data["id"],
hdv=hdv,
content=data["content"],
metadata=data["metadata"],
created_at=datetime.fromisoformat(data["created_at"]),
last_accessed=datetime.fromisoformat(data["last_accessed"]),
tier="warm",
access_count=data.get("access_count", 0),
ltp_strength=data.get("ltp_strength", 0.0),
epistemic_value=data.get("epistemic_value", 0.0),
pragmatic_value=data.get("pragmatic_value", 0.0),
previous_id=data.get("previous_id"), # Phase 4.3
)
except (json.JSONDecodeError, ValueError, KeyError) as e:
logger.error(f"Data corruption in filesystem for {node_id}: {e}")
return None
except Exception as e:
logger.error(f"Error loading {node_id} from filesystem: {e}")
return None
return await self._run_in_thread(_fs_load)
return None
def _should_promote(self, node: MemoryNode) -> bool:
"""Pure check: return True if node qualifies for promotion (no mutation)."""
threshold = self.config.tiers_hot.ltp_threshold_min
delta = self.config.hysteresis.promote_delta
return node.ltp_strength > (threshold + delta)
def _should_demote(self, node: MemoryNode) -> Optional[MemoryNode]:
"""
Pure check: return the node if it qualifies for demotion (after updating its tier).
Returns None if no demotion needed. No I/O performed.
"""
threshold = self.config.tiers_hot.ltp_threshold_min
delta = self.config.hysteresis.demote_delta
if node.ltp_strength < (threshold - delta):
return node
return None
async def _promote_to_hot(self, node: MemoryNode):
"""Promote node from WARM to HOT (I/O first, then atomic state update).
Order is critical:
1. Delete from WARM (I/O) - no lock held
2. Insert into HOT (in-memory) - under lock
This prevents double-promotion from concurrent callers.
"""
# Step 1: I/O outside lock (may fail gracefully)
deleted = await self._delete_from_warm(node.id)
if not deleted:
logger.debug(f"Skipping promotion of {node.id}: not found in WARM (already promoted?)")
return
# Step 2: Atomic state transition under lock
victim_to_save = None
async with self.lock:
# Double-check: another caller may have already promoted this node
if node.id in self.hot:
logger.debug(f"{node.id} already in HOT, skipping duplicate promotion")
return
logger.info(f"Promoting {node.id} to HOT (LTP: {node.ltp_strength:.4f})")
node.tier = "hot"
self.hot[node.id] = node
self._add_to_faiss(node)
# Maintain inverted chain index
if node.previous_id:
self._next_chain[node.previous_id] = node.id
# Check if we need to evict - prepare under lock, execute outside
if len(self.hot) > self.config.tiers_hot.max_memories:
victim_to_save = self._prepare_eviction_from_hot()
# Step 3: Perform eviction I/O outside lock
if victim_to_save:
await self._save_to_warm(victim_to_save)
async def get_stats(self) -> Dict:
"""Get statistics about memory distribution across tiers."""
stats = {
"hot_count": len(self.hot),
"warm_count": 0,
"cold_count": 0,
"using_qdrant": self.use_qdrant,
# Phase 4.0: HNSW index stats
"ann_index": self._hnsw.stats() if self._hnsw is not None else {"index_type": "none"},
}
if self.use_qdrant:
info = await self.qdrant.get_collection_info(self.config.qdrant.collection_warm)
if info:
stats["warm_count"] = info.points_count
else:
stats["warm_count"] = -1
else:
if hasattr(self, 'warm_path') and self.warm_path:
def _count():
return len(list(self.warm_path.glob("*.json")))
stats["warm_count"] = await self._run_in_thread(_count)
return stats
async def list_warm(self, max_results: int = 500) -> List[MemoryNode]:
"""
List nodes from the WARM tier (Phase 4.0 — used by SemanticConsolidationWorker).
Returns up to max_results MemoryNode objects from the WARM tier.
Falls back gracefully if Qdrant or filesystem is unavailable.
"""
nodes: List[MemoryNode] = []
if self.use_qdrant:
try:
points_result = await self.qdrant.scroll(
self.config.qdrant.collection_warm,
limit=max_results,
offset=None,
with_vectors=True,
)
points = points_result[0] if points_result else []
for pt in points:
payload = pt.payload
try:
arr = np.array(pt.vector) > 0.5
packed = np.packbits(arr.astype(np.uint8))
hdv = BinaryHDV(data=packed, dimension=payload["dimension"])
node = MemoryNode(
id=payload.get("id", pt.id),
hdv=hdv,
content=payload["content"],
metadata=payload.get("metadata", {}),
created_at=datetime.fromisoformat(payload["created_at"]),
last_accessed=datetime.fromisoformat(payload["last_accessed"]),
tier="warm",
access_count=payload.get("access_count", 0),
ltp_strength=payload.get("ltp_strength", 0.0),
previous_id=payload.get("previous_id"), # Phase 4.3: episodic chain
)
nodes.append(node)
except Exception as exc:
logger.debug(f"list_warm: could not deserialize point {pt.id}: {exc}")
except Exception as exc:
logger.warning(f"list_warm Qdrant failed: {exc}")
elif hasattr(self, "warm_path") and self.warm_path:
def _list_fs() -> List[MemoryNode]:
result = []
for meta_file in list(self.warm_path.glob("*.json"))[:max_results]:
try:
import json as _json
with open(meta_file, "r") as f:
data = _json.load(f)
hdv_path = self.warm_path / f"{data['id']}.npy"
if not hdv_path.exists():
continue
hdv_data = np.load(hdv_path)
hdv = BinaryHDV(data=hdv_data, dimension=data["dimension"])
result.append(
MemoryNode(
id=data["id"],
hdv=hdv,
content=data["content"],
metadata=data.get("metadata", {}),
created_at=datetime.fromisoformat(data["created_at"]),
last_accessed=datetime.fromisoformat(data["last_accessed"]),
tier="warm",
ltp_strength=data.get("ltp_strength", 0.0),
previous_id=data.get("previous_id"), # Phase 4.3: episodic chain
)
)
except Exception as exc:
logger.debug(f"list_warm FS: skip {meta_file.name}: {exc}")
return result
nodes = await self._run_in_thread(_list_fs)
return nodes
async def get_next_in_chain(self, node_id: str) -> Optional[MemoryNode]:
"""
Return the MemoryNode that directly follows node_id in the episodic chain.
This is a typed wrapper around QdrantStore.get_by_previous_id() that
returns a fully-deserialized MemoryNode instead of a raw models.Record,
making the episodic-chain API consistent with the rest of TierManager.
Returns:
The next MemoryNode in the chain, or None if not found / Qdrant
unavailable.
"""
# 1. Check HOT tier via O(1) inverted index (Fix 7).
async with self.lock:
next_id = self._next_chain.get(node_id)
if next_id and next_id in self.hot:
return self.hot[next_id]
# 2. Check WARM tier (Qdrant)
if not self.use_qdrant or not self.qdrant:
return None
record = await self.qdrant.get_by_previous_id(
self.qdrant.collection_warm, node_id
)
if record is None:
return None
# Resolve to a full MemoryNode via the standard warm-load path
return await self._load_from_warm(str(record.id))
async def consolidate_warm_to_cold(self):
"""
Batch move from WARM to COLD based on archive criteria.
This is an expensive operation, typically run by a background worker.
"""
min_ltp = self.config.tiers_warm.ltp_threshold_min
if self.use_qdrant:
offset = None
while True:
points_result = await self.qdrant.scroll(
self.config.qdrant.collection_warm,
limit=100,
offset=offset,
with_vectors=True
)
points = points_result[0]
next_offset = points_result[1]
if not points:
break
ids_to_delete = []
for pt in points:
payload = pt.payload
ltp = payload.get("ltp_strength", 0.0)
if ltp < min_ltp:
vec_data = pt.vector
if vec_data:
arr = np.array(vec_data) > 0.5
packed = np.packbits(arr.astype(np.uint8))
payload["hdv_vector"] = packed.tolist()
await self._write_to_cold(payload)
ids_to_delete.append(pt.id)
if ids_to_delete:
await self.qdrant.delete(
self.config.qdrant.collection_warm, ids_to_delete
)
offset = next_offset
if offset is None:
break
else:
# Filesystem fallback
if hasattr(self, 'warm_path') and self.warm_path:
def _process_fs():
to_delete = []
for meta_file in self.warm_path.glob("*.json"):
try:
with open(meta_file, "r") as f:
meta = json.load(f)
if meta.get("ltp_strength", 0.0) < min_ltp:
to_delete.append((meta["id"], meta))
except Exception:
pass
return to_delete
candidates = await self._run_in_thread(_process_fs)
for nid, meta in candidates:
await self._archive_to_cold(nid, meta)
async def search(
self,
query_vec: BinaryHDV,
top_k: int = 5,
time_range: Optional[Tuple[datetime, datetime]] = None,
metadata_filter: Optional[Dict[str, Any]] = None,
include_cold: bool = False,
) -> List[Tuple[str, float]]:
"""
Global search across all tiers.
Combines HNSW/FAISS (HOT), Qdrant/FS (WARM), and optionally COLD.
Phase 4.3: time_range filters results to memories within the given datetime range.
Fix 1: include_cold=True enables a bounded linear scan of the COLD archive.
"""
# 1. Search HOT via FAISS (time filtering done post-hoc for in-memory)
hot_results = self.search_hot(query_vec, top_k)
# Apply time filter and metadata filter to HOT results if needed
if time_range or metadata_filter:
filtered_hot = []
async with self.lock:
for nid, score in hot_results:
node = self.hot.get(nid)
if not node:
continue
if time_range:
start_ts = time_range[0].timestamp()
end_ts = time_range[1].timestamp()
if not (start_ts <= node.created_at.timestamp() <= end_ts):
continue
if metadata_filter:
match = True
node_meta = node.metadata or {}
for k, v in metadata_filter.items():
if node_meta.get(k) != v:
match = False
break
if not match:
continue
filtered_hot.append((nid, score))
hot_results = filtered_hot
# 2. Search WARM via Qdrant
warm_results = []
if self.use_qdrant:
try:
q_vec = np.unpackbits(query_vec.data).astype(float).tolist()
hits = await self.qdrant.search(
collection=self.config.qdrant.collection_warm,
query_vector=q_vec,
limit=top_k,
time_range=time_range, # Phase 4.3: Pass time filter to Qdrant
metadata_filter=metadata_filter, # BUG-09: Agent Isolation
)
warm_results = [(hit.id, hit.score) for hit in hits]
except Exception as e:
logger.error(f"WARM tier search failed: {e}")
# 3. Optionally search COLD tier (Fix 1: bounded linear scan)
cold_results: List[Tuple[str, float]] = []
if include_cold:
cold_results = await self.search_cold(query_vec, top_k)
# 4. Combine and Sort (HOT scores take precedence over WARM/COLD for same ID)
combined = {}
for nid, score in hot_results:
combined[nid] = score
for nid, score in warm_results:
combined[nid] = max(combined.get(nid, 0), score)
for nid, score in cold_results:
combined[nid] = max(combined.get(nid, 0), score)
sorted_results = sorted(combined.items(), key=lambda x: x[1], reverse=True)
return sorted_results[:top_k]
def search_hot(self, query_vec: BinaryHDV, top_k: int = 5) -> List[Tuple[str, float]]:
"""Search HOT tier using HNSW or FAISS binary index (Phase 4.0)."""
# Phase 4.0: use HNSWIndexManager (auto-selects flat vs HNSW)
if self._hnsw is not None and self._hnsw.size > 0:
try:
return self._hnsw.search(query_vec.data, top_k)
except Exception as e:
logger.error(f"HNSWIndexManager search failed, falling back: {e}")
# Legacy FAISS flat path
if not self.faiss_index or not self.hot:
return self._linear_search_hot(query_vec, top_k)
try:
q = np.expand_dims(query_vec.data, axis=0)
distances, ids = self.faiss_index.search(q, top_k)
results = []
for d, fid in zip(distances[0], ids[0]):
if fid == -1:
continue
node_id = self.faiss_id_map.get(int(fid))
if node_id:
sim = 1.0 - (float(d) / self.config.dimensionality)
results.append((node_id, sim))
return results
except Exception as e:
logger.error(f"FAISS search failed, falling back: {e}")
return self._linear_search_hot(query_vec, top_k)
def _linear_search_hot(self, query_vec: BinaryHDV, top_k: int = 5) -> List[Tuple[str, float]]:
"""Fallback linear scan for HOT tier."""
scores = []
for node in self.hot.values():
sim = query_vec.similarity(node.hdv)
scores.append((node.id, sim))
scores.sort(key=lambda x: x[1], reverse=True)
return scores[:top_k]
async def _load_from_cold(self, node_id: str) -> Optional[MemoryNode]:
"""
Scan COLD archive files for a specific node (Fix 1: COLD read path).
Archives are gzip JSONL, sorted newest-first for early-exit on recent data.
Returns None if not found or on error.
"""
def _scan():
for archive_file in sorted(
self.cold_path.glob("archive_*.jsonl.gz"), reverse=True
):
try:
with gzip.open(archive_file, "rt", encoding="utf-8") as f:
for line in f:
line = line.strip()
if not line:
continue
try:
rec = json.loads(line)
if rec.get("id") == node_id:
return rec
except json.JSONDecodeError:
continue
except Exception:
continue
return None
rec = await self._run_in_thread(_scan)
if rec is None:
return None
try:
raw_vec = rec.get("hdv_vector")
dim = rec.get("dimension", self.config.dimensionality)
if raw_vec:
hdv_data = np.array(raw_vec, dtype=np.uint8)
hdv = BinaryHDV(data=hdv_data, dimension=dim)
else:
hdv = BinaryHDV.zeros(dim)
node = MemoryNode(
id=rec["id"],
hdv=hdv,
content=rec.get("content", ""),
metadata=rec.get("metadata", {}),
tier="cold",
ltp_strength=rec.get("ltp_strength", 0.0),
previous_id=rec.get("previous_id"),
)
if "created_at" in rec:
node.created_at = datetime.fromisoformat(rec["created_at"])
return node
except Exception as e:
logger.error(f"Failed to reconstruct node {node_id} from COLD: {e}")
return None
async def search_cold(
self,
query_vec: BinaryHDV,
top_k: int = 5,
max_scan: int = 1000,
) -> List[Tuple[str, float]]:
"""
Linear similarity scan over COLD archive (Fix 1: COLD search path).
Bounded by max_scan records to keep latency predictable.
Returns results sorted by descending similarity.
"""
config_dim = self.config.dimensionality
def _scan():
candidates: List[Tuple[str, float]] = []
scanned = 0
for archive_file in sorted(
self.cold_path.glob("archive_*.jsonl.gz"), reverse=True
):
if scanned >= max_scan:
break
try:
with gzip.open(archive_file, "rt", encoding="utf-8") as f:
for line in f:
if scanned >= max_scan:
break
line = line.strip()
if not line:
continue
try:
rec = json.loads(line)
raw_vec = rec.get("hdv_vector")
if not raw_vec:
continue
dim = rec.get("dimension", config_dim)
hdv = BinaryHDV(
data=np.array(raw_vec, dtype=np.uint8),
dimension=dim,
)
sim = query_vec.similarity(hdv)
candidates.append((rec["id"], sim))
scanned += 1
except Exception:
continue
except Exception:
continue
return candidates
candidates = await self._run_in_thread(_scan)
candidates.sort(key=lambda x: x[1], reverse=True)
return candidates[:top_k]
async def _write_to_cold(self, record: dict):
"""Write a record to the cold archive."""
record["tier"] = "cold"
record["archived_at"] = datetime.now(timezone.utc).isoformat()
today = datetime.now(timezone.utc).strftime("%Y-%m-%d")
archive_file = self.cold_path / f"archive_{today}.jsonl.gz"
def _write():
with gzip.open(archive_file, "at", encoding="utf-8") as f:
f.write(json.dumps(record) + "\n")
await self._run_in_thread(_write)
async def _archive_to_cold(self, node_id: str, meta: dict):
"""Move memory to COLD storage (File System Fallback)."""
if not self.warm_path:
return
def _read_vec():
hdv_path = self.warm_path / f"{node_id}.npy"
if not hdv_path.exists():
return None
return np.load(hdv_path)
hdv_data = await self._run_in_thread(_read_vec)
if hdv_data is None:
return
record = meta.copy()
record["hdv_vector"] = hdv_data.tolist()
await self._write_to_cold(record)
await self._delete_from_warm(node_id)
|