MySql 如何使用窗口函数加速工作查询

Question

EDIT: Actually, main problem is the last update setting A.last_52w_high_age_brutto_days (see below). Is it possible to opzimize this update or to integrate somehow in CURSOR update? Thank you.

---

I have a table with Stock Market daily prices.

Table Structure:

CREATE TABLE `cind_stocks_daily_rates` (
  `symbol` varchar(12) NOT NULL,
  `p_date` int unsigned NOT NULL,
  `open_price` decimal(9,4) NOT NULL,
  `high_price` decimal(9,4) NOT NULL,
  `low_price` decimal(9,4) NOT NULL,
  `close_price` decimal(9,4) NOT NULL,
  `price_52w_low` decimal(9,4) DEFAULT NULL,
  `price_52w_high` decimal(9,4) DEFAULT NULL,
  `last_52w_high_age_brutto_days` int unsigned DEFAULT NULL,
  PRIMARY KEY (`symbol`,`p_date `)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;

INSERT Statement:

INSERT INTO cind_stocks_daily_rates (symbol, datum, open_price, high_price, low_price, close_price, price_52w_low, price_52w_high, last_52w_high_age_brutto_days)
VALUES
('DELL', 20210616, 150, 152, 149, 151, 149, 152, 0),
('INTZ', 20210616, 250, 252, 249, 251, 249, 252, 0),
('MSFT', 20210616, 350, 352, 349, 351, 349, 352, 0),
('NTNX', 20210616, 452, 452, 449, 451, 449, 452, 0),
('DELL', 20210617, 148, 151, 147, 150, 147, 152, 1),
('INTZ', 20210617, 251, 254, 250, 252, 249, 254, 0),
('MSFT', 20210617, 346, 349, 345, 347, 345, 352, 1),
('NTNX', 20210617, 450, 454, 450, 453, 449, 454, 0),
('DELL', 20210618, 146, 147, 144, 145, NULL, NULL, NULL),
('INTZ', 20210618, 254, 256, 253, 255, NULL, NULL, NULL),
('MSFT', 20210618, 349, 351, 349, 350, NULL, NULL, NULL),
('NTNX', 20210618, 453, 456, 452, 454, NULL, NULL, NULL);

Desired result:

symbol  datum   open_price  high_price  low_price   close_price price_52w_low   price_52w_high  last_52w_high_age_brutto_days
DELL    20210616    150 152 149 151 149 152 0
INTZ    20210616    250 252 249 251 249 252 0
MSFT    20210616    350 352 349 351 349 352 0
NTNX    20210616    452 452 449 451 449 452 0
DELL    20210617    148 151 147 150 147 152 1
INTZ    20210617    251 254 250 252 249 254 0
MSFT    20210617    346 349 345 347 345 352 1
NTNX    20210617    450 454 450 453 449 454 0
DELL    20210618    146 147 144 145 144 152 2
INTZ    20210618    254 256 253 255 249 256 0
MSFT    20210618    349 351 349 350 345 352 2
NTNX    20210618    453 456 452 454 449 456 0

After I have filled the table with prices/rates for a given day I want to calculate the high & low rates for a last 52 Weeks– as well as duration in days between day with last 52 Weeks high price and a value of date column.

I create a cursor:

SELECT distinct symbol FROM cind_stocks_daily_rates
WHERE  price_52w_low IS NULL OR price_52w_high IS NULL;

And loop the symbols through the Cursor (curr_symbol):

UPDATE cind_stocks_daily_rates AS A
CROSS JOIN
(SELECT AA.p_date, AA.symbol,
Min(AA.low_price) OVER (ORDER BY AA.p_date ROWS BETWEEN 260 PRECEDING AND CURRENT ROW) AS low_price_52w,
Max(AA.high_price) OVER (ORDER BY AA.p_date ROWS BETWEEN 260 PRECEDING AND 
CURRENT ROW) AS high_price_52w
FROM cind_stocks_daily_rates AA
WHERE AA.symbol = curr_symbol
ORDER BY AA.p_date) as B ON B.symbol = A.symbol AND B.p_date = A.p_date
SET A.price_52w_low = B.low_price_52w,
A.price_52w_high = B.high_price_52w,
WHERE  A.price_52w_low IS NULL OR A.price_52w_high IS NULL;

After cursor loop I have another update to determine how old is current 52 Weeks High price from today:

EDIT: Actually, following update is my main problem (and only problem). Duration approx.+/- 10 minutes. END_OF_EDIT

UPDATE cind_stocks_daily_rates AS A
CROSS JOIN
 (SELECT AA.p_date, AA.symbol,
DATEDIFF(STR_TO_DATE(AA.p_date, "%Y %m %d"),STR_TO_DATE(
    (SELECT CCC.p_date
    FROM cind_stocks_daily_rates CCC
    WHERE CCC.symbol = AA.symbol AND CCC.p_date <= AA.p_date AND CCC.high_price = AA.price_52w_high ORDER BY CCC.p_date DESC LIMIT 1)
    , "%Y %m %d")) AS last_high_Date
    FROM cind_stocks_daily_rates AA) as B ON B.symbol = A.symbol AND B.p_date = A.p_date
        SET A.last_52w_high_age_brutto_days = B.last_high_Date
WHERE A.last_52w_high_age_brutto_days IS NULL;

It works everything as wanted without errors, just it takes too much time. Is it possible to speed it up? Is it possible to set the field “last_52w_high_age_brutto_days” together within cursor (without second update after cursor loop)? Please any ideas to make the queries faster?

Answer 1

Let's try to avoid doing everything every day.

A big task is finding the 52-week highest price. Let's see if there is a shortcut.

Maintain a separate table, one row per ticker, of the 52-week high and date of such.

If the date (for a given ticker) of the 52-week high is less than 52 weeks ago, then the price and date do not to be recomputed.

Then it is easy and fast to discover which, if any, tickers need their highs recomputed. Update this table. Then use this table to populate the table you are asking about; this UPDATE...JOIN... will be efficient.

Some potential problems --

• Non-trading days: I don't think, this messes up the queries at all; ignore it.
• Occasionally, most of the tickers will have their high exactly 52 weeks ago: This will be rare, and, alas, slow.