缓慢的sql脚本优化

Question

I have a query like:

select row_number() over(order by m_time desc) as row, m.m_id, m_time, m_log_id
    (case when exists(select 1 from t_approved_phonenumber where phone_number = m.c_phone_number) then 1 else 0 end) as approved, 
    (case when exists(select 1 from t_log log where log.c_id < m.m_log_id and log.phone_number = m.phone_number) then 'N' else 'Y' end) as is_first_time
from t_message m

all the tables t_approved_phonenumber, t_message and t_log have a lot of records. and all the IDs are primary key, phone_number column has index. sometimes my query is very slow. Is there any way to speed it up?

thanks,

Answer 1

Maybe make an index on t_log with first column phone_number and second column c_id so that the second exists can just use this index and -dependent on the DBMS - may not even have to access the table t_log itself. Declaring the index unique promotes its usage for the optimizer - but of course you only can do this if it is really unique.

And of course, make an index on m_time if there is none so far, this would probably help the row_number() .

Answer 2

Try to replace correlated subqueries with plain "left join" subqueries as they usually perform better. Regular subquery is run just once, opposite to correlated one which is run against each row. It also is more flexible - you can add some more data to subqueries if needed. Your query should look like this (errors possible - don't have Management Studio right now):

select 
    row_number() over(order by m_time desc) as row
    ,m.m_id
    ,m_time
    ,m_log_id
    ,coalesce(approved, 0) as approved
    ,coalesce(is_first_time, 'Y') as is_first_time
from 
    t_message m
    left join (select phone_number, 1 as [approved] from t_approved_phonenumber group by phone_number) as a
    on a.phone_number = m.c_phone_number
    left join (select phone_number, min(c_id) as c_id, 'N' as is_first_time from t_log group by phone_number) log
    on log.c_id < m.m_log_id and log.phone_number = m.phone_number

Answer 3

In my experience joins are much better than select in the predicate. try this query.

I assume that for each t_message there is at least one call for each Phone_Number in t_message so I'm using INNER JOIN instead of left join in the second JOIN .

select row_number() over(order by m_time desc) as row, 
    m.m_id, 
    m_time, 
    m_log_id,
    (case WHEN approved.phone_number IS NULL then 0 else 1 end) as approved, 
    (case when First_c_id < m.m_log_id then 'N' else 'Y' end) as is_first_time
from t_message  m
LEFT JOIN t_approved_phonenumber approved ON m.c_phone_number = approved.phone_number 
INNER JOIN (SELECT phone_number, MIN(c_id)  AS First_c_id 
           FROM t_log 
           GROUP BY phone_number) FirstLog ON FirstLog.PhoneNumber = m.phone_number

anyway, the indexes suggested by @FrankPl can help to gain much more performance,