src/services/database.py

import time
from datetime import date, datetime, timezone

import pandas as pd
from pandas.errors import EmptyDataError
from loguru import logger
from sqlalchemy import create_engine, text

from src.config import DB_URL, GarchParams

engine = None

if DB_URL:
    try:
        engine = create_engine(DB_URL)
    except Exception:
        logger.exception(f"Invalid DB_URL format. Length of DB_URL: {len(DB_URL)}")
        engine = None
else:
    logger.warning("DB_URL is not set. Database features will be disabled.")


def create_preds_table() -> None:
    if engine is None:
        logger.warning("Database not configured, skipping table creation.")
        return

    sql_create = text("""
        CREATE TABLE IF NOT EXISTS garch_preds (  
            id INTEGER GENERATED BY DEFAULT AS IDENTITY PRIMARY KEY,
            ticker VARCHAR(10) NOT NULL,
            execution_time TIMESTAMP DEFAULT NOW(),
            target_date DATE NOT NULL,
            model_config VARCHAR(20) NOT NULL,
            prediction DOUBLE PRECISION NOT NULL,
            CONSTRAINT unique_pred UNIQUE (ticker, target_date, model_config)
        );
    """)
    with engine.begin() as conn:
        conn.execute(sql_create)
        logger.info("Succesfully created table 'garch_preds' or table exists")


def store_preds(
    ticker: str, pred: float, target_date: date, params: GarchParams
) -> None:
    if engine is None:
        logger.info(f"Skipping DB save for {ticker} (DB not configured)")
        return

    execution_time = datetime.now(timezone.utc)
    pred = float(pred)
    model_config = "_".join(str(atr) for atr in vars(params).values())

    sql_insert = text("""
        INSERT INTO garch_preds (ticker, target_date, prediction, execution_time, model_config)
        VALUES (:ticker, :target_date, :prediction, :execution_time, :model_config)
        ON CONFLICT (ticker, target_date, model_config) 
        DO UPDATE SET 
            prediction = EXCLUDED.prediction,
            execution_time = EXCLUDED.execution_time;
    """)
    with engine.begin() as conn:
        conn.execute(
            sql_insert,
            {
                "ticker": ticker,
                "target_date": target_date,
                "execution_time": execution_time,
                "model_config": model_config,
                "prediction": pred,
            },
        )
        logger.info(f"Stored prediction for {ticker} (Target: {target_date})")


def get_error_data() -> pd.DataFrame:
    error_df = None

    if engine is None:
        raise Exception("Could not connect to DB")

    attempts = 10
    for i in range(attempts):
        try:
            sql_extract = text("""
                SELECT p.ticker, p.target_date, p.model_config, gp.error_abs, gp.error_rel, gp.error_sq, gp.error_raw
                FROM garch_performance gp 
                    JOIN garch_preds p
                    ON gp.prediction_id = p.id
                WHERE p.target_date < CURRENT_DATE
                    AND p.target_date >= CURRENT_DATE - INTERVAL '10 days'
            """)

            with engine.begin() as conn:
                error_df = pd.read_sql(sql_extract, conn)
                logger.info("Got performance data for last week from DB")
                logger.debug(f"Error DF rows: {error_df.count()}")

            if error_df is not None and not error_df.empty:
                break

            logger.debug(
                f"Attempt {i + 1}/{attempts}: Could not get data from 'garch_performance' DB. Retrying..."
            )

        except Exception as e:
            logger.debug(
                f"Attempt {i + 1}/{attempts}: Exception fetching data: {e}. Retrying..."
            )

        if i < attempts - 1:
            time.sleep(5)

    if error_df is None or error_df.empty:
        logger.error(
            "Could not get data from 'garch_performance' DB\nMax attempt reached\nReturning error page"
        )
        raise EmptyDataError

    return error_df