File size: 19,583 Bytes
38830c1 c427923 38830c1 c427923 38830c1 c427923 38830c1 c427923 38830c1 c427923 38830c1 c427923 38830c1 c427923 38830c1 c427923 38830c1 | 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 | from __future__ import annotations
from dataclasses import dataclass, field
from typing import Any
from backend.agents.planner import PlannerOutput
from backend.services.clickhouse_service import ClickHouseService, ColumnInfo
from backend.services.llm_service import LLMService
import re
@dataclass
class SQLAgentOutput:
sql: str
rationale: str
selected_tables: list[str] = field(default_factory=list)
selected_columns: list[dict[str, str]] = field(default_factory=list)
result: dict[str, Any] = field(default_factory=dict)
warnings: list[str] = field(default_factory=list)
error: str | None = None
def to_dict(self) -> dict[str, Any]:
return {
"sql": self.sql,
"rationale": self.rationale,
"selected_tables": self.selected_tables,
"selected_columns": self.selected_columns,
"result": self.result,
"warnings": self.warnings,
"error": self.error,
}
class SQLAgent:
def __init__(self, llm: LLMService, clickhouse: ClickHouseService) -> None:
self.llm = llm
self.clickhouse = clickhouse
def run(self, question: str, plan: PlannerOutput) -> SQLAgentOutput:
try:
schema_catalog = self.clickhouse.get_schema_catalog()
except Exception as exc:
return SQLAgentOutput(
sql="",
rationale="Schema lookup failed before SQL generation.",
warnings=["Unable to inspect ClickHouse schema."],
error=str(exc),
)
query_pattern = self._infer_query_pattern(question)
selected = self._select_relevant_columns(
question=question,
schema_catalog=schema_catalog,
query_pattern=query_pattern,
)
selected_tables = sorted({f"{column.database}.{column.table}" for column in selected})
fallback_sql = self._build_fallback_sql(
question=question,
plan=plan,
selected=selected,
query_pattern=query_pattern,
)
payload = self.llm.invoke_json(
system_prompt=(
"You are a senior ClickHouse SQL agent. "
"Return strict JSON with keys sql, rationale, selected_tables, selected_columns, warnings. "
"Use only read-only SELECT queries. Never reference columns outside the provided schema. "
"Choose the query pattern that matches the business intent, not a generic time trend."
"NEVER use raw aggregate functions (COUNT, AVG, SUM) inside HAVING. "
"ALWAYS use aliases defined in SELECT inside HAVING. "
"ALWAYS define aggregates explicitly in SELECT. "
"Never reuse alias names without defining them. "
"Example: AVG(star_rating) AS avg_rating must appear in SELECT before using avg_rating. "
"Example: COUNT(*) AS total_reviews → use total_reviews in HAVING. "
"Example: AVG(star_rating) AS avg_rating → use avg_rating in HAVING. "
),
user_prompt=self._build_generation_prompt(
question=question,
plan=plan,
selected=selected,
fallback_sql=fallback_sql,
query_pattern=query_pattern,
),
fallback={
"sql": fallback_sql,
"rationale": "Fallback SQL generated from semantic pattern detection over the grounded schema.",
"selected_tables": selected_tables,
"selected_columns": [column.to_dict() for column in selected],
"warnings": [],
},
)
sql = (payload.get("sql") or fallback_sql).strip().rstrip(";")
if not sql.lower().startswith("select"):
sql = fallback_sql
sql = self._enforce_readable_product_label(sql=sql, selected=selected, query_pattern=query_pattern)
sql = self._fix_having_clause_aliases(sql)
try:
result = self.clickhouse.query(sql)
error = None
except Exception as exc:
result = {"columns": [], "rows": [], "row_count": 0}
error = str(exc)
return SQLAgentOutput(
sql=sql,
rationale=payload.get("rationale") or "SQL generated for the detected business question.",
selected_tables=payload.get("selected_tables") or selected_tables,
selected_columns=payload.get("selected_columns") or [column.to_dict() for column in selected],
result=result,
warnings=payload.get("warnings") or [],
error=error,
)
def _enforce_readable_product_label(
self,
sql: str,
selected: list[ColumnInfo],
query_pattern: dict[str, Any],
) -> str:
if query_pattern["entity"] != "product":
return sql
sql_lower = sql.lower()
if "product_title" in sql_lower:
return sql
product_label = next(
(
column.name
for column in selected
if column.name.lower() in {"product_title", "product_name"}
),
None,
)
if not product_label:
return sql
product_key_aliases = ["product_parent", "product_id", " as product_id", " as product_key"]
if not any(alias in sql_lower for alias in product_key_aliases):
return sql
if " from " not in sql_lower:
return sql
select_prefix, remainder = sql.split("FROM", 1) if "FROM" in sql else sql.split("from", 1)
if "select" not in select_prefix.lower():
return sql
return f"{select_prefix.rstrip()}, any({product_label}) AS product_title FROM{remainder}"
def _infer_query_pattern(self, question: str) -> dict[str, Any]:
tokens = {token.strip(" ,.?").lower() for token in question.split()}
return {
"needs_time_grain": bool(
tokens & {"trend", "over", "monthly", "month", "daily", "day", "weekly", "quarterly", "quarter", "yearly", "year"}
),
"needs_popularity": bool(tokens & {"popular", "popularity", "top", "most"}),
"needs_low_rating": bool(tokens & {"poorly", "badly", "low", "worst", "negative"}),
"needs_high_rating": bool(tokens & {"highest", "best", "top-rated"}),
"entity": "product" if tokens & {"product", "products"} else "category" if "category" in tokens else "generic",
}
def _select_relevant_columns(
self,
question: str,
schema_catalog: list[ColumnInfo],
query_pattern: dict[str, Any],
) -> list[ColumnInfo]:
tokens = {token.strip(" ,.?").lower() for token in question.split()}
scored: list[tuple[int, ColumnInfo]] = []
for column in schema_catalog:
score = 0
table_name = column.table.lower()
column_name = column.name.lower()
column_type = column.type.lower()
for token in tokens:
if token and token in column_name:
score += 5
if token and token in table_name:
score += 3
if query_pattern["entity"] == "product" and column_name in {"product_parent", "product_title", "product_id"}:
score += 10
if query_pattern["entity"] == "category" and "category" in column_name:
score += 8
if query_pattern["needs_popularity"] and any(hint in column_name for hint in {"review", "count", "votes"}):
score += 8
if query_pattern["needs_low_rating"] and any(hint in column_name for hint in {"rating", "score", "star"}):
score += 8
if query_pattern["needs_time_grain"] and any(hint in column_name for hint in {"date", "time", "month", "year"}):
score += 6
if any(metric in column_name for metric in {"revenue", "sales", "amount", "price", "profit", "count", "rating", "votes", "review"}):
score += 2
if "int" in column_type or "float" in column_type or "decimal" in column_type:
score += 1
scored.append((score, column))
scored.sort(key=lambda item: (item[0], item[1].table, item[1].name), reverse=True)
top = [column for score, column in scored if score > 0][:10]
if top:
return top
return schema_catalog[:10]
def _build_generation_prompt(
self,
question: str,
plan: PlannerOutput,
selected: list[ColumnInfo],
fallback_sql: str,
query_pattern: dict[str, Any],
) -> str:
schema_lines = [
f"- table={column.database}.{column.table}, column={column.name}, type={column.type}"
for column in selected
]
return (
f"Question: {question}\n"
f"Intent: {plan.intent}\n"
f"Time range: {plan.time_range}\n"
f"Derived query pattern: {query_pattern}\n"
"Relevant schema:\n"
f"{chr(10).join(schema_lines)}\n\n"
"Requirements:\n"
"- Prefer one table unless a join is clearly necessary.\n"
"- Use aliases that are easy to read in a UI.\n"
"- Do not group by time unless the user clearly asked for a trend over time.\n"
"- For popularity questions, prefer COUNT(*) or review counts over unrelated sums unless votes were explicitly requested.\n"
"- For poorly rated questions, use AVG on the rating column and HAVING filters when the business ask implies thresholds.\n"
"- For product questions, group by a stable product identifier and include a readable product label when available.\n"
"- If both popularity and poor rating are requested, produce a ranking/filter query by product, not a time series.\n"
"- Use LIMIT 20 for ranked entity lists and LIMIT 200 for trends.\n"
f"- If unsure, use this safe fallback SQL:\n{fallback_sql}\n"
)
def _ensure_aggregates_exist(self, sql: str, selected: list[ColumnInfo]) -> str:
sql_lower = sql.lower()
if "group by" not in sql_lower or "from" not in sql_lower:
return sql
select_part, rest = sql.split("FROM", 1)
# Detect aliases used
needs_avg = "avg_" in sql_lower
needs_count = "total_" in sql_lower or "count_" in sql_lower
# Find candidate numeric columns dynamically
numeric_cols = [
col.name for col in selected
if any(t in col.type.lower() for t in ["int", "float", "decimal"])
]
# Heuristic mapping
rating_col = next((c for c in numeric_cols if "rating" in c.lower()), None)
count_col = next((c for c in numeric_cols if "id" in c.lower()), None)
# Inject AVG dynamically
if needs_avg and "avg(" not in select_part.lower() and rating_col:
select_part += f", AVG({rating_col}) AS avg_rating"
# Inject COUNT dynamically
if needs_count and "count(" not in select_part.lower():
if count_col:
select_part += f", COUNT({count_col}) AS total_count"
else:
select_part += ", COUNT(*) AS total_count"
return f"{select_part} FROM {rest}"
def _build_fallback_sql(
self,
question: str,
plan: PlannerOutput,
selected: list[ColumnInfo],
query_pattern: dict[str, Any],
) -> str:
if not selected:
return "SELECT 1 AS value LIMIT 1"
tokens = {token.strip(" ,.?").lower() for token in question.split()}
table = f"{selected[0].database}.{selected[0].table}"
time_column = next(
(column.name for column in selected if any(hint in column.name.lower() for hint in {"date", "time"})),
None,
)
dimension_column = next(
(column.name for column in selected if "string" in column.type.lower()),
None,
)
rating_column = next(
(column.name for column in selected if any(hint in column.name.lower() for hint in {"rating", "score", "star"})),
None,
)
votes_column = next(
(column.name for column in selected if any(hint in column.name.lower() for hint in {"votes", "helpful"})),
None,
)
product_key = next(
(column.name for column in selected if column.name.lower() in {"product_parent", "product_id"}),
None,
)
product_label = next(
(column.name for column in selected if column.name.lower() in {"product_title", "product_name"}),
None,
)
category_column = next(
(column.name for column in selected if "category" in column.name.lower()),
None,
)
if query_pattern["entity"] == "product" and product_key:
select_parts = [product_key]
if product_label:
select_parts.append(f"any({product_label}) AS product_title")
select_parts.append("COUNT(*) AS total_reviews")
if votes_column:
select_parts.append(f"SUM({votes_column}) AS total_votes")
if rating_column:
select_parts.append(f"AVG({rating_column}) AS avg_rating")
sql_parts = [
"SELECT " + ", ".join(select_parts),
f"FROM {table}",
f"GROUP BY {product_key}",
]
having_parts = []
if query_pattern["needs_popularity"]:
having_parts.append("total_reviews > 100")
if query_pattern["needs_low_rating"] and rating_column:
having_parts.append("avg_rating < 3")
if query_pattern["needs_high_rating"] and rating_column:
having_parts.append("avg_rating >= 4")
if having_parts:
sql_parts.append("HAVING " + " AND ".join(having_parts))
order_parts = []
if query_pattern["needs_popularity"]:
order_parts.append("total_reviews DESC")
if votes_column:
order_parts.append("total_votes DESC")
if query_pattern["needs_low_rating"] and rating_column and not query_pattern["needs_popularity"]:
order_parts.append("avg_rating ASC")
if query_pattern["needs_high_rating"] and rating_column and not query_pattern["needs_popularity"]:
order_parts.append("avg_rating DESC")
if order_parts:
sql_parts.append("ORDER BY " + ", ".join(order_parts))
sql_parts.append("LIMIT 20")
return " ".join(sql_parts)
if query_pattern["needs_time_grain"] and time_column:
period_expr = (
f"toStartOfMonth({time_column})" if {"month", "monthly"} & tokens
else f"toStartOfYear({time_column})" if {"year", "yearly", "annual"} & tokens
else f"toDate({time_column})"
)
grouping_dimension = category_column or dimension_column
metric_expr = "COUNT(*) AS total_reviews"
if rating_column and {"average", "avg", "mean"} & tokens:
metric_expr = f"AVG({rating_column}) AS avg_rating"
elif votes_column and "votes" in tokens:
metric_expr = f"SUM({votes_column}) AS total_votes"
dimension_sql = f", {grouping_dimension}" if grouping_dimension else ""
group_by = f"GROUP BY period{', ' + grouping_dimension if grouping_dimension else ''}"
order_by = f"ORDER BY period{', ' + grouping_dimension if grouping_dimension else ''}"
return (
f"SELECT {period_expr} AS period{dimension_sql}, {metric_expr} "
f"FROM {table} "
f"{group_by} "
f"{order_by} "
)
if category_column and rating_column and query_pattern["needs_low_rating"]:
return (
f"SELECT {category_column} AS category, COUNT(*) AS total_reviews, AVG({rating_column}) AS avg_rating "
f"FROM {table} "
"GROUP BY category "
"HAVING total_reviews > 20 AND avg_rating < 3 "
"ORDER BY total_reviews DESC "
"LIMIT 20"
)
if dimension_column:
metric_expr = "COUNT(*) AS total_count"
if votes_column and "votes" in tokens:
metric_expr = f"SUM({votes_column}) AS total_votes"
elif rating_column and {"average", "avg", "mean"} & tokens:
metric_expr = f"AVG({rating_column}) AS avg_rating"
return (
f"SELECT {dimension_column} AS category, {metric_expr} "
f"FROM {table} "
"GROUP BY category "
"ORDER BY 2 DESC "
"LIMIT 20"
)
return f"SELECT * FROM {table} LIMIT 50"
def _extract_aliases(self, sql: str) -> dict[str, str]:
"""
Extract mapping: aggregate_expression -> alias
Example:
AVG(star_rating) AS avg_rating → {"avg(star_rating)": "avg_rating"}
"""
import re
select_match = re.search(r"SELECT(.*?)FROM", sql, re.IGNORECASE | re.DOTALL)
if not select_match:
return {}
select_part = select_match.group(1)
alias_map = {}
# Match patterns like: AVG(col) AS alias
matches = re.findall(
r"(AVG|COUNT|SUM|MIN|MAX)\((.*?)\)\s+AS\s+(\w+)",
select_part,
re.IGNORECASE,
)
for func, col, alias in matches:
key = f"{func.lower()}({col.strip()})"
alias_map[key] = alias
return alias_map
def _fix_having_clause_aliases(self, sql: str) -> str:
if "having" not in sql.lower():
return sql
alias_map = self._extract_aliases(sql)
if not alias_map:
return sql # nothing to fix
parts = re.split(r"\bHAVING\b", sql, flags=re.IGNORECASE)
if len(parts) < 2:
return sql
before = parts[0]
having = parts[1]
# Replace aggregate expressions with correct aliases
for agg_expr, alias in alias_map.items():
pattern = re.escape(agg_expr)
having = re.sub(pattern, alias, having, flags=re.IGNORECASE)
return before + "HAVING " + having
def _validate_sql(self, sql: str) -> str:
sql_lower = sql.lower()
# Extract SELECT aliases
select_part = sql_lower.split("from")[0]
aliases = set(re.findall(r"as\s+(\w+)", select_part))
# Extract HAVING usage
if "having" in sql_lower:
having_part = sql_lower.split("having")[1]
used_aliases = set(re.findall(r"\b[a-z_]+\b", having_part))
# Find aliases used but not defined
undefined = used_aliases - aliases
# Ignore SQL keywords
keywords = {"and", "or", "not", "in", "between", "like"}
undefined = {u for u in undefined if u not in keywords}
if undefined:
raise ValueError(f"Undefined aliases in HAVING: {undefined}")
return sql |