Sql 统计select语句中匹配的项
我们有一个针对一些相当大的表执行的存储过程,当连接到一个较大的表时,它也会记录有多少记录与相应的批处理id相匹配。我试图弄清楚的是,我是否可以使用计数函数或其他方法来改进这一点?正在尝试摆脱嵌套的SELECT COUNT(*)语句。CCTransactions表有140万行,BatchItems表有660万行Sql 统计select语句中匹配的项,sql,select,count,large-data,Sql,Select,Count,Large Data,我们有一个针对一些相当大的表执行的存储过程,当连接到一个较大的表时,它也会记录有多少记录与相应的批处理id相匹配。我试图弄清楚的是,我是否可以使用计数函数或其他方法来改进这一点?正在尝试摆脱嵌套的SELECT COUNT(*)语句。CCTransactions表有140万行,BatchItems表有660万行 SELECT a.ItemAuthID, a.FeeAuthID, a.Batch_ID, a.ItemAuthCode, a.FeeAuthCode, b.Amoun
SELECT a.ItemAuthID, a.FeeAuthID, a.Batch_ID, a.ItemAuthCode,
a.FeeAuthCode, b.Amount, b.Fee,
(SELECT COUNT(*) FROM BatchItems WHERE Batch_ID = a.Batch_ID) AS BatchCount,
ItemBillDate, FeeBillDate, b.AccountNumber,
b.Itemcode, ItemAuthToken, FeeAuthToken,
cc.ItemMerchant, cc.FeeMerchant
FROM CCTransactions a WITH(NOLOCK)
INNER JOIN BatchItems b WITH(NOLOCK)
ON a.Batch_ID = b.Batch_ID
INNER JOIN CCConfig cc WITH(NOLOCK)
ON a.ClientCode = cc.ClientCode
WHERE ((ItemAuthCode > '' AND ItemBillDate IS NULL)
OR (FeeAuthCode > '' AND FeeBillDate IS NULL))
AND TransactionDate BETWEEN DATEADD(d,-7,GETDATE())
AND convert(char(20),getdate(),101) + ' ' + @Cutoff
ORDER BY TransactionDate
当DBMS支持窗口聚合函数时,可以将其重写为
COUNT(*) OVER (PARTITION BY Batch_ID)
当然,这只返回SELECT返回的每个批次ID的行数。如果内部联接导致行数减少,则该行数不正确
然后将标量子查询重写为联接可能更有效(取决于您的DBMS):
SELECT a.ItemAuthID, a.FeeAuthID, a.Batch_ID, a.ItemAuthCode,
a.FeeAuthCode, b.Amount, b.Fee,
dt.BatchCount,
ItemBillDate, FeeBillDate, b.AccountNumber,
b.Itemcode, ItemAuthToken, FeeAuthToken,
cc.ItemMerchant, cc.FeeMerchant
FROM CCTransactions a WITH(NOLOCK)
INNER JOIN BatchItems b WITH(NOLOCK)
ON a.Batch_ID = b.Batch_ID
INNER JOIN CCConfig cc WITH(NOLOCK)
ON a.ClientCode = cc.ClientCode
INNER JOIN
(
SELECT BatchCount, COUNT(*) AS BatchCount
FROM BatchItems
GROUP BY Batch_ID
) AS dt ON a.Batch_ID = dt.Batch_ID
WHERE ((ItemAuthCode > '' AND ItemBillDate IS NULL)
OR (FeeAuthCode > '' AND FeeBillDate IS NULL))
AND TransactionDate BETWEEN DATEADD(d,-7,GETDATE())
AND convert(CHAR(20),getdate(),101) + ' ' + @Cutoff
ORDER BY TransactionDate
我对计数(*)进行了多次测试,虽然它确实减少了集合统计IO上的扫描计数,但增加了计划的子树成本。现在让我测试子查询的内部联接,看看它是如何执行的。