我有一张约有380万行的表。当我查询整个表格时,我得到了
错误:值溢出数字格式
引用用户定义函数返回的值。 但如果我将表格大致分成两半(见下文),一切正常。
SELECT day,item,price,
CAST(my_func(price) OVER (PARTITION BY item ORDER BY day) AS numeric(8,2)),
FROM my_table
--WHERE day < '3/1/2013';
--WHERE day >= '3/1/2013';
具有WHERE
子句的语句在没有错误的情况下执行。
价格为numeric(8,2)
,价格列中没有任何数字大于numeric(8,2)
。无论如何,将格式更改为numeric(20,2)
没有任何区别。
以下是表格定义:
CREATE TABLE my_table
(
item character(5) NOT NULL,
day date NOT NULL,
price numeric(8,2),
CONSTRAINT my_table_pk PRIMARY KEY (item, day)
);
...和功能:
CREATE OR REPLACE FUNCTION my_func2 (avg numeric, IN price numeric)
RETURNS numeric AS $$
DECLARE
alpha numeric;
BEGIN
alpha := 2.0/51;
RETURN
CASE
WHEN avg IS NULL THEN price -- avg is NULL for the first row, so price is returned
ELSE round((alpha * price + (1-alpha) * avg),2)
END;
END;
$$ LANGUAGE PLPgSQL;
...用于聚合:
CREATE AGGREGATE my_func(numeric) (SFUNC = my_func2, STYPE = numeric);
答案 0 :(得分:1)
错误发生在您的施法操作中。格式numeric(8,2)
非常严格,可能my_func()
会返回不满足格式定义的值。要证明这一点,请查看以下查询:
select 12.34::numeric(8,2);
numeric
---------
12.34
select 12.345678::numeric(8,2);
numeric
---------
12.35
select 12.3456789::numeric(8,2);
numeric
---------
12.35
select 123456.123456789::numeric(8,2);
numeric
-----------
123456.12
select 1234567.123456789::numeric(8,2);
ERROR: numeric field overflow
DETAIL: A field with precision 8, scale 2 must round to an absolute value less than 10^6.
select 1234567.8::numeric(8,2);
ERROR: numeric field overflow
DETAIL: A field with precision 8, scale 2 must round to an absolute value less than 10^6.
如果您注意到,返回的号码的总数字不会超过8
个数字,并且始终具有2
个十进制数字。最后两个查询会出错,因为它们应返回超过8
个数字。例如,您希望将数字1234567.123456789
舍入为1234567.12
,但1234567.12
由9
数字组成,而不是8
。数字1234567.8
也是如此,即使您有8
个数字也是如此。这是因为在返回的数值中你需要2
十进制数字,所以postgres sohuld输出1234567.80
但是再次,这里你有9
个数字而不是{{ 1}}。
换句话说,您有不同的方法来解决此问题:
8
my_func()
个数字来增加numeric(16,2)
所需的总小数位数(选择您想要的数字)。16
或numeric
。例如:real
(my_func(price) OVER (PARTITION BY item ORDER BY day))::real
。否则,请修改round(my_func(price) OVER (PARTITION BY item ORDER BY day), 2)
以返回my_func()
。为了帮助您了解和/或找到导致错误的原因,请考虑这一点。对于评估round(returned_value, 2)
的至少一个或一行,您会在左侧获得一个数字超过my_func()
的数字。要查找生成错误的行,您只需执行以下查询:
6
此查询返回的行会生成强制转换错误。显然,如果您没有在WITH not_casted AS (
SELECT day,item,price,
my_func(price) OVER (PARTITION BY item ORDER BY day) AS fprice
FROM my_table
)
SELECT * FROM not_casted
WHERE fprice > 999999.99
内部numeric(8,2)
进行类型转换,则此方法有效,否则会对您进行类型转换的值生成错误。在不知道功能代码的情况下,不可能做出其他假设。
我提出了一个基于模拟的例子。代码执行以下操作:
- 创建具有不同类型转换和舍入方法的不同my_func()
- 在充当您的数据的模拟随机样本上执行每个AGGREGATE
(希望如此)。它每天生成AGGREGATE
个价格,每个价格都有10
个item
项,超过10
天。这对于证明精度损失并不重要,所以如果没有正确模拟数据,请不要责备我:)
以下是创建四个函数和聚合的代码:
31
现在,代码模拟数据并应用三个函数:
-- typecast price and arithmetics to numeric(8,2)
CREATE OR REPLACE FUNCTION my_func_numeric_8_2_a (avg numeric(8,2), IN price numeric(8,2))
RETURNS numeric(8,2) AS $$
DECLARE
alpha numeric;
BEGIN
alpha := 2.0/51;
RETURN
CASE
WHEN avg IS NULL THEN price
ELSE (alpha * price + (1-alpha) * avg)::numeric(8,2)
END;
END;
$$ LANGUAGE PLPgSQL;
CREATE AGGREGATE my_func_numeric_8_2(numeric(8,2)) (SFUNC = my_func_numeric_8_2_a, STYPE = numeric(8,2));
-- typecast price and arithmetics to numeric and round(arithmetics, 2)
CREATE OR REPLACE FUNCTION my_func_numeric_round_a(avg numeric, IN price numeric)
RETURNS numeric AS $$
DECLARE
alpha numeric;
BEGIN
alpha := 2.0/51;
RETURN
CASE
WHEN avg IS NULL THEN price
ELSE round((alpha * price + (1-alpha) * avg), 2)
END;
END;
$$ LANGUAGE PLPgSQL;
CREATE AGGREGATE my_func_numeric_round(numeric) (SFUNC = my_func_numeric_round_a, STYPE = numeric);
-- no typecast (double precision type)
CREATE OR REPLACE FUNCTION my_func_dp_a(avg double precision, IN price double precision)
RETURNS double precision AS $$
DECLARE
alpha double precision;
BEGIN
alpha := 2.0/51;
RETURN
CASE
WHEN avg IS NULL THEN price
ELSE (alpha * price + (1-alpha) * avg)
END;
END;
$$ LANGUAGE PLPgSQL;
CREATE AGGREGATE my_func_dp(double precision) (SFUNC = my_func_dp_a, STYPE = double precision);
-- typecast price and arithmetics to numeric
CREATE OR REPLACE FUNCTION my_func_numeric_a(avg numeric, IN price numeric)
RETURNS numeric AS $$
DECLARE
alpha numeric;
BEGIN
alpha := 2.0/51;
RETURN
CASE
WHEN avg IS NULL THEN price
ELSE (alpha * price + (1-alpha) * avg)
END;
END;
$$ LANGUAGE PLPgSQL;
CREATE AGGREGATE my_func_numeric(numeric) (SFUNC = my_func_numeric_a, STYPE = numeric);
由于WITH sample AS
(
SELECT "day", (random())*10 AS price, generate_series(1,10)::text AS item
FROM (SELECT generate_series('2000-01-01'::timestamp, '2000-01-31'::timestamp, '1 day'::interval)::date AS "day") AS calendar
)
SELECT "day", item, price,
-- typecast price and arithmetics to numeric(8,2)
my_func_numeric_8_2(price::numeric(8,2)) OVER (PARTITION BY item ORDER BY "day") AS numeric_8_2,
-- typecast price and arithmetics to numeric and round(arithmetics, 2)
my_func_numeric_round(price::numeric) OVER (PARTITION BY item ORDER BY "day") AS numeric_round,
-- typecast price and arithmetics to numeric and round the final result
round(my_func_numeric(price::numeric) OVER (PARTITION BY item ORDER BY "day"), 2) AS round_numeric,
-- no typecast (double precision type)
my_func_dp(price) OVER (PARTITION BY item ORDER BY "day") AS no_typecast,
-- typecast price and arithmetics to numeric
my_func_numeric(price::numeric) OVER (PARTITION BY item ORDER BY "day") AS numeric
FROM sample
ORDER BY item, "day"
的使用,每次查询执行都会生成不同的结果。向下滚动结果,即使random()
与计算所有四个值相同,您也会看到许多行具有不同的值。此外,通过降低精度损失(或提高精度)对列进行排序:price
是四者之间最精确的,而my_func_dp(price)
则不太精确,但最精确&#34;精确&#34 ;
如果您从命令行运行上一个查询,您会注意到my_func_numeric_8_2(price::numeric(8,2))
会返回长度增加的数字,因为my_func_numeric(price::numeric)
尽可能精确,所以他的固定长度可以变化。如果你从pgAdmin执行它,你将获得一个全长数字的四舍五入数字。