避免复制到临时表

Question

我通常在java中完成大部分工作，但我无法真正找到我在这里做的事情，导致它需要复制到表中并死亡。

SELECT company.tblusers.userid, 
db.operations.id AS operation_id, 
SUM(TIME_TO_SEC(db.batch_log.time_elapsed)) AS time_elapsed, 
SUM(db.tasks.estimated_nonrecurring + db.tasks.estimated_recurring) AS total_elapsed, 
COUNT(db2.ncr_ncr.id) AS number_of_ncrs 
FROM company.tblusers 
INNER JOIN db.operations 
INNER JOIN db.batch_log ON company.tblusers.userid = batch_log.userid 
INNER JOIN db.batches ON batch_log.batch_id = batches.id 
INNER JOIN db.tasks ON db.batches.id = db.tasks.batch_id 
INNER JOIN db2.ncr_ncr ON company.tblusers.sugar_name = db2.ncr_ncr.employee 
WHERE company.tblusers.departmentid = 8 
AND db.operations.id = db.batches.operation_id 
AND db.batches.id = db.tasks.batch_id 
AND db.batch_log.userid = company.tblusers.userid 
AND db2.ncr_ncr.employee = company.tblusers.sugar_name 
GROUP by company.tblusers.userid, db.batches.operation_id 

编辑：Explain输出

"id";"select_type";"table";"type";"possible_keys";"key";"key_len";"ref";"rows";"Extra" 
"1";"SIMPLE";"ncr_ncr";"ALL";NULL;NULL;NULL;NULL;"2700";"Using temporary; Using filesort" 
"1";"SIMPLE";"batch_log";"ALL";NULL;NULL;NULL;NULL;"78026";"" 
"1";"SIMPLE";"tblusers";"eq_ref";"PRIMARY";"PRIMARY";"52";"ramses.batch_log.userid";"1";"Using where" 
"1";"SIMPLE";"tasks";"ref";"Index 2";"Index 2";"38";"ramses.batch_log.batch_id";"2";"" 
"1";"SIMPLE";"batches";"eq_ref";"PRIMARY";"PRIMARY";"38";"ramses.tasks.batch_id";"1";"Using where" 
"1";"SIMPLE";"operations";"eq_ref";"PRIMARY";"PRIMARY";"4";"ramses.batches.operation_id";"1";"Using where; Using index" 

Answer 1

如果您汇总而分组对列的非索引组合的值，那么SQL引擎将创建临时表作为其正常功能的一部分为了保持部分汇总的值。您需要为company.tblusers.userid和db.batches.operation_id的每个组合计算两个SUM和COUNT。这些值必须存储，因为很明显，没有包含这两列的索引，因为它们在不同的表中。

但是，我不知道它为什么会死亡，除非它实际上将完全连接构建为临时表，因为COUNT聚合。

而不是加入到db2.ncr_ncr表中，您应该将其作为相关的子查询。这可能需要更少的存储：

SELECT 
     company.tblusers.userid, 
     db.operations.id AS operation_id, 
     SUM(TIME_TO_SEC(db.batch_log.time_elapsed)) AS time_elapsed, 
     SUM(db.tasks.estimated_nonrecurring + db.tasks.estimated_recurring) AS total_elapsed, 
     (SELECT COUNT(*) FROM db2.ncr_ncr WHERE db2.ncr_ncr.employee = company.tblusers.sugar_name) AS number_of_ncrs 
FROM 
     company.tblusers 
     INNER JOIN db.batch_log ON batch_log.userid = company.tblusers.userid 
     INNER JOIN db.batches ON db.batches.id = db.batch_log.batch_id 
     INNER JOIN db.operations ON db.operations.id = db.batches.operation_id 
     INNER JOIN db.tasks ON db.tasks.batch_id = db.batch_log.batch_id 
WHERE 
     company.tblusers.departmentid = 8 
GROUP BY 
     company.tblusers.userid, 
     db.batches.operation_id 

UPDATE

基础上解释计划的结果，你没有任何索引查询可以在ncr_ncr使用。这不好。它实际上创建了一个新的表格副本，它可以按employee排序，以便连接可以在合理的时间内发生。

即使该表格不是很大（只有2700行），您仍然需要在db2.ncr_ncr.employee上放置一个索引（最好是相关的，如果系统的其余部分允许），否则您的性能将受到严重影响。这张表格足够小，制作和分类临时副本不应该是一个问题，但它仍然会损害您的表现。

batch_log.userid也是如此。