Spaces:
Sleeping
Sleeping
File size: 12,662 Bytes
399944f | 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 | from __future__ import annotations
import os
from dataclasses import dataclass
from datetime import datetime, timezone
from typing import Any, List, Optional, Sequence, cast
from typing_extensions import LiteralString
from dotenv import load_dotenv
from rich.progress import track
from neo4j import GraphDatabase, Driver, Query
from neo4j.exceptions import Neo4jError
from kbdebugger.types import GraphRelation, EdgeProperties
from .utils import rows_to_graph_relations
from .types import BatchUpsertSummary
from .aura_api import ensure_aura_running_from_env
import rich
from rich.console import Console
from rich.panel import Panel
# Load env vars once here
load_dotenv(override=True)
@dataclass
class GraphStore:
"""
Central access point to the knowledge graph.
- Handles connecting to Neo4j
- Exposes:
- `query(...)` for arbitrary Cypher
- `upsert_relation(...)` for writing extracted relations
"""
# inner: Neo4jGraph
driver: Driver
# ---------- construction / connection ----------
@classmethod
def connect(
cls,
*,
uri: Optional[str] = None,
username: Optional[str] = None,
password: Optional[str] = None,
auto_env: bool = True,
verbose: bool = True,
) -> "GraphStore":
"""
Build a GraphStore from environment variables (or explicit args).
Env vars:
- NEO4J_URI
- NEO4J_USERNAME (default "neo4j")
- NEO4J_PASSWORD (default "")
"""
if auto_env:
load_dotenv(override=True)
# ✅ Preflight: make sure Aura is running before Neo4j driver even tries DNS
ensure_aura_running_from_env(verbose=verbose)
neo4j_uri = uri or os.getenv("NEO4J_URI")
neo4j_user = username or os.getenv("NEO4J_USERNAME", "neo4j")
neo4j_pass = password or os.getenv("NEO4J_PASSWORD", "")
if not neo4j_uri:
raise RuntimeError("NEO4J_URI is not set (pass uri=... or set env var).")
# inner = Neo4jGraph(url=neo4j_uri, username=neo4j_user, password=neo4j_pass)
driver = GraphDatabase.driver(neo4j_uri, auth=(neo4j_user, neo4j_pass))
if verbose:
rich.print(
f"[kbdebugger] Connected to Neo4j at {neo4j_uri!r} "
f"as user {neo4j_user!r}"
)
# return cls(inner=inner)
return cls(driver=driver)
def close(self) -> None:
self.driver.close()
# ---------- basic query API ----------
def query(
self,
cypher: str,
params: dict[str, Any] | None = None,
) -> list[dict[str, Any]]:
"""
Run a Cypher query with consistent error handling.
This is the *only* low-level escape hatch other code should use.
"""
# print("GraphStore backend:", type(self.driver), type(self))
try:
with self.driver.session() as session:
cypher_as_literal_str = cast(LiteralString, cypher)
# result = session.run(Query(cypher), params or {})
result = session.run(cypher_as_literal_str, params or {})
return [record.data() for record in result]
except Neo4jError as e:
message = "Neo4j query failed!\n" \
f"Error: {e.__class__.__name__}: {e}\n" \
f"Query:\n{cypher}\n" \
f"Params:\n{params}"
rich.print(
f"[bold red]{message}[/bold red]\n"
)
raise RuntimeError(message) from e
except Exception as e:
raise RuntimeError(
"Unexpected error during Neo4j query:\n"
f"{e}\nQuery:\n{cypher}\nParams:\n{params}"
) from e
def query_relations(
self,
cypher: str,
params: dict[str, Any] | None = None,
*,
source_key: str = "source",
target_key: str = "target",
predicate_key: str = "predicate",
props_key: str = "props",
) -> list[GraphRelation]:
"""
Run a Cypher query that returns (source, target, predicate, props) columns
and coerce it to List[GraphRelation].
"""
rows = self.query(cypher, params=params or {})
return rows_to_graph_relations(
rows,
source_key=source_key,
target_key=target_key,
predicate_key=predicate_key,
props_key=props_key,
)
# ---------- high-level write API ----------
def upsert_relation(self, relation: GraphRelation) -> list[dict[str, Any]]:
"""
Insert or update a single GraphRelation into Neo4j.
- Nodes are always `(:Node {label: ...})`
- Relationships are always `[:REL {label: ..., ...}]`
- Dedupe is based on (`source_label`, `target_label`, `rel.label`, `edge.properties['source']`)
- i.e., we assume that the same relation from the same source text is the same fact.
Example relation:
```
relation = {
'source': {
'label': 'Monitoring'
},
'target': {
'label': 'KI system'
},
'edge': {
'label': 'is performed during operation of',
'properties': {
'sentence': 'Monitoring is performed during operation of KI system',
'source': '20241015_MISSION_KI_Glossar_v1.0 en.pdf',
'page_number': 1,
'start_index': 0
}
},
}
```
"""
src_label = relation["source"]["label"]
tgt_label = relation["target"]["label"]
rel_label = relation["edge"]["label"]
props_all = relation["edge"]["properties"]
# Key used to *match* existing relationships (dedupe policy)
rel_key_props: EdgeProperties = {
"label": rel_label,
"source": props_all.get("source", ""),
# Add more fields here if we want stricter deduplication
}
# ⚠️ remember: the rel_key_props are the props of the relationship used to find existing rels!
# so choose them carefully to avoid over- or under-merging.
# Why this is safer
# - We won't create a new rel just because page_number or some metadata changed.
# - We still persist all the rich properties on first create, and can update on matches.
now_iso = datetime.now(timezone.utc).isoformat()
# Properties set only when a relationship is first created
on_create_props: EdgeProperties = {
**props_all,
"created_at": now_iso,
}
# Properties updated whenever we re-encounter the same relationship
on_match_props = {
"last_updated_at": now_iso,
}
"""
apoc.merge.relationship is a Neo4j APOC procedure used to dynamically merge or create a relationship between two nodes, which either exists or is created based on the provided parameters.
It handles dynamic relationship types and properties for creation or matching, and its signature is apoc.merge.relationship(
startNode,
relType, # e.g., "REL"
relKeyProps, # properties used to identify existing relationships
onCREATEProps, # properties set only when a new relationship is created
endNode,
onMatchProps # properties updated when a relationship already exists
).
This is useful for building Cypher queries where the relationship details are provided as parameters, for example, when unwinding a list of rows.
"""
# About {} vs {{}}
# - In a plain triple-quoted Python string (""" ... """), {} is just literal braces — fine.
# - In an f-string (f""" ... """), {} is interpolation. To emit literal braces, we must escape as {{}}.
# So:
# - If we keep it as a plain string: """ ... {} ... """ is OK.
# - If we switch to f""" ... """: change to {{}}.
cypher = """
MERGE (s:Node {label: $source_label})
ON CREATE SET s.created_at = datetime()
SET s.last_updated_at = datetime(),
s.created_at = coalesce(s.created_at, datetime())
MERGE (t:Node {label: $target_label})
ON CREATE SET t.created_at = datetime()
SET t.last_updated_at = datetime(),
t.created_at = coalesce(t.created_at, datetime())
WITH s, t
CALL apoc.merge.relationship(
s,
$rel_type,
$rel_key_props,
$on_create,
t,
$on_match
) YIELD rel
// ensure temporal props exist on the rel too
SET rel.created_at = coalesce(rel.created_at, datetime()),
rel.last_updated_at = datetime()
RETURN s, t, rel
"""
return self.query(
cypher,
params={
"source_label": src_label,
"target_label": tgt_label,
"rel_type": "REL", # always 'REL' as the relationship type; actual label is in properties
"rel_key_props": rel_key_props,
"on_create": on_create_props, # set only on create of the relationship
"on_match": on_match_props, # set only on match (update) of the relationship
},
)
def upsert_relations(
self,
relations: Sequence[GraphRelation],
*,
pretty_print: bool = True,
) -> BatchUpsertSummary:
"""
Upsert multiple GraphRelation objects into Neo4j.
This is a convenience wrapper around `upsert_relation()` that:
- preserves the dedupe semantics of the single-upsert operation
- continues on individual failures (best-effort write)
- returns a typed summary for logging and monitoring
Parameters
----------
relations:
Relations to be inserted/merged into the KG.
Returns
-------
BatchUpsertSummary
Counts and error strings describing any failures.
Notes
-----
- This method performs no client-side deduplication.
Deduplication is handled inside `upsert_relation()` via the APOC merge policy.
- If you later want "fail-fast" semantics, add a flag like `stop_on_error`.
"""
if not relations:
# Nothing to upsert is not an error here; extraction may legitimately produce nothing.
return BatchUpsertSummary(
attempted=0,
succeeded=0,
failed=0,
errors=[],
)
attempted = len(relations)
succeeded = 0
errors: List[str] = []
for i, rel in track(
enumerate(relations, start=1),
description="➕🛸 Upserting triplets (relations) into Knowledge Graph",
total=len(relations)):
try:
self.upsert_relation(rel)
succeeded += 1
except Exception as e: # pylint: disable=broad-exception-caught
src = rel.get("source", {}).get("label", "?")
tgt = rel.get("target", {}).get("label", "?")
pred = rel.get("edge", {}).get("label", "?")
errors.append(f"[{i}/{attempted}] {src} - {pred} -> {tgt}: {e}")
failed = attempted - succeeded
summary = BatchUpsertSummary(
attempted=attempted,
succeeded=succeeded,
failed=failed,
errors=errors,
)
if pretty_print:
console = Console()
body_lines = [
f"[bold]Attempted:[/bold] {summary.attempted}",
f"[bold green]Succeeded:[/bold green] {summary.succeeded}",
f"[bold red]Failed:[/bold red] {summary.failed}",
]
if summary.failed > 0:
body_lines.append("\n[bold red]Errors:[/bold red]")
for err in summary.errors:
body_lines.append(f" • {err}")
console.print(
Panel(
"\n".join(body_lines),
title="[bold cyan]🧠📊 Knowledge Graph Upsert Summary[/bold cyan]",
border_style="cyan",
padding=(1, 2),
)
)
return summary
|