如果查询是动态创建的，则Redshift-Concurrent Write-Insert不起作用

Question

我有一个存储过程：

CREATE OR REPLACE PROCEDURE public.lock_users(j_id "varchar",order_id "varchar",order_detail_id "varchar",insert_qry varchar(65535),rec_per_order "varchar")
LANGUAGE plpgsql
AS $$       
declare 
lc_stmt varchar(65535);
BEGIN 
lc_stmt = 'INSERT INTO test.USER_LOCK 
SELECT '''||$1||''', '''||$2||''', '''||$3||''', user_id,cast(TIMEOFDAY() as timestamp) FROM ('||insert_qry|| 'AND USER_ID NOT IN (SELECT USER_ID FROM test.USER_LOCK)) WHERE ORDER_CNT <='||rec_per_order||'))';
EXECUTE ''||lc_stmt||'';
END
$$
;

从上述过程生成的一个示例查询是：

INSERT INTO test.USER_LOCK
SELECT '657d7563-6de4-4dc9-ac74-3c23adf7a4e9', 'DSS-12345', 'DSS-74523-4-7569',
USER_ID,cast(TIMEOFDAY() as timestamp) 
FROM (
SELECT USER_ID FROM (
SELECT * FROM (
SELECT XA.USER_ID, XA.EMAIL_ID,YA.COMPANY_NAME
rank() OVER (PARTITION BY XA.account_id ORDER BY XA.account_id) ORDER_CNT 
FROM test.contacts_20 XA 
LEFT JOIN test.accounts_20 YA
ON XA.ACCOUNT_ID = YA.ACCOUNT_ID  
AND XA.COUNTRY = YA.COUNTRY 
WHERE XA.IS_CONTACT_SUPPRESSED = 0  
AND UPPER(XA.TELE_SUPPRESSION_LOB) != UPPER('DSS') 
AND XA.TELE_SUPPRESSION_LOB != 'BOTH' 
AND XA.IS_TELE_VERIFIED = 1 
AND XA.IS_TELE_SUPPRESSED = 0 
AND UPPER(PHONE_LINE) = 'DIRECT' 
AND XA.COUNTRY IN (
SELECT INCLUSION_VALUE FROM user_inc_list
WHERE JOB_ID = '657d7563-6de4-4dc9-ac74-3c23adf7a4e9' 
AND UPPER(INCLUSION_TYPE) = 'COUNTRY') 
AND XA.COUNTRY NOT IN (
SELECT EXCLUSION_VALUE FROM user_exc_list 
WHERE JOB_ID = '657d7563-6de4-4dc9-ac74-3c23adf7a4e9' 
AND UPPER(EXCLUSION_TYPE) = 'COUNTRY') 
AND XA.USER_ID NOT IN (
SELECT USER_ID FROM test.user_lead_20
WHERE (CURRENT_DATE - creation_date::date) <= 60 AND UPPER(LOB) != 'DSS' AND AGENCY_ID != '1456') 
AND XA.USER_ID NOT IN (
SELECT USER_ID FROM test.user_lead_20 
WHERE (CURRENT_DATE - creation_date::date) <= 60 AND UPPER(LOB) != 'DSS' AND SPONSOR_ID != '8659') 
AND USER_ID NOT IN (
select USER_ID 
from user_e_history 
where sf_campaign_id = 'DSS-12345' AND (CURRENT_DATE - creation_date::date) >= 7 AND channel = 'TELE') 
AND USER_ID NOT IN (
select USER_ID from user_e_history 
where creation_date::date = CURRENT_DATE AND channel = 'TELE' ) 
AND USER_ID NOT IN (
select USER_ID from test.user_lead_20
where sf_campaign_id = 'DSS-12345' GROUP BY USER_ID,"DOMAIN" HAVING COUNT(*) >= 3 ) 
AND USER_ID NOT IN (
select USER_ID from test.user_lead_20 
where AGENCY_ID = 1456 and (CURRENT_DATE - creation_date::date) <= 180 ) 
AND XA.E_domain NOT LIKE '%.gov'
AND USER_ID NOT IN (
SELECT USER_ID FROM test.USER_LOCK)) WHERE ORDER_CNT <=20));

当我并行执行这个存储过程时，会出现以下错误：

SQL Error [500310] [XX000]: [Amazon](500310) Invalid operation: 1023
Details: 
Serializable isolation violation on table - 132075, transactions forming the cycle are: 2040186, 2040187 (pid:14687);

当我更改我的存储过程，而不是传递参数和在查询中创建固定插入时，它就起作用了。

这是一个有效的存储过程：

CREATE OR REPLACE PROCEDURE public.new_procedure(type_value "varchar")
LANGUAGE plpgsql
AS $$   
declare 
lc_stmt varchar;
BEGIN 
lc_stmt = 'INSERT into temp_table 
select ct_id,email_id,first_name,last_name from users where active_type = '''||$1||''' ';
EXECUTE ''||lc_stmt||'';
END
$$
;

我无法理解造成这种情况的原因和解决办法。请帮忙。

Answer 1

动态过程中的Insert查询查找不在user_lock表中的user_id并插入结果。在两个动态版本之间的结果集中似乎有共同的user_id。

因此，假设在执行第一个版本时，它可能会向user_lock表中添加一个user_id，第二个版本可能也会在要插入user_lock表格的结果集中添加相同的user_id。

因此，根据首先运行的版本，与串行执行相比，两个版本的结果集将有所不同，即它们不是"可串行隔离的"。

在测试示例中不会出现错误，因为它是一种独立的插入(users和temp_table(。

尝试LOCK可能会解决此问题。