clickhouse 查询group 分组最大值的一行数据。

qq_356408662024-03-31 6:05

按照 sql_finger_md5 分组取query_time_ms 最大的一行数据。

使用any函数可以去匹配到的第一行数据，所以可以先让数据按照query_time_ms 排序，然后再使用group by 和any结合取第一行数据，就是最大值的那一行数据。

bash 复制代码

select
  any (time) as time ,
  any (query_time_ms) as query_time_ms ,
  any (sqltext) as sqltext,
  any (inst_id) as inst_id,
  any (inst_name) as inst_name,
  any (dbname) as dbname,
  any (host_address) as host_address,
  any (lock_times) as lock_times,
  any (parse_row_counts) as parse_row_counts,
  any (return_row_counts) as return_row_counts,
  any (sql_finger) as sql_finger,
  sql_finger_md5,
  any (sqltext_md5) as sqltext_md5
FROM
  (
    SELECT
      inst_id,
      inst_name,
      dbname,
      execution_start_time as time,
      host_address,
      query_time_ms,
      sqltext,
      lock_times,
      parse_row_counts,
      return_row_counts,
      sql_finger,
      sql_finger_md5,
      sqltext_md5
    FROM
      cmdb.rds_all_slow_sql_record_distributed
    WHERE
      (
        execution_start_time >= toDateTime(1711079437) 
        AND execution_start_time <= toDateTime(1711684237) 
      )
      AND (dbname = 'leopard_admin')
      AND host_address  not like '%root%' AND host_address  not like '%bi_user%' AND sqltext  not like '%insert%'
    order by sql_finger_md5 desc,
      query_time_ms desc
  ) a
group by
  a.sql_finger_md5
order by query_time_ms desc
limit
  1000

窗体函数在数据量大的时候性能堪忧，在clickhouse中还有其他的处理方式。比如使用any()、anyLast()函数。

按官方文档的定义：any() "selects the first encountered value."，也就是返回遇到的首个值，看上去是很符合当前的情况。但文档又做了说明：因为查询可能是以任意顺序执行的，并且可能每次执行得顺序都不同（如同我们上面的select * from user_order返回的行顺序不同），所以这个函数的执行结果可能是不确定的。如果要获得确定的值，可以使用"min"或者"max"。或者，select的对象的是一个已经排序过的子查询。

参考资料：

http://www.tracefact.net/tech/112.html