Tweak mysql server for a single large-table query

Question

I have a very special use case. I have a copy of my production database (MySQL 5.5.29, InnoDB) on a separate VM (which I just threw 8GB of ram onto) and need a result on just one query, which is taking a very long time.

This is the query:

select *
from bew 
inner join bewa on bewa.id = bew.id
inner join sa on bew.sa_id = sa.id
inner join revinfo ri on bewa.rev = ri.id
where sa.vp_id = 6
  and bewa.prio is not null
  and bewa.rev > 0
  and bew.person_id > 0
order by bew.person_id, bewa.rev, bewa.prio

These are the column types:

bew.id           bigint(20)       [PRIMARY KEY, NOT NULL]
bew.person_id    bigint(20)       [(FOREIGN) KEY, NOT NULL - index: fk_bew_person26]
bew.sa_id        bigint(20)       [(FOREIGN) KEY, NOT NULL - index: fk_bew_sa27]
bewa.id          bigint(20)       [NOT NULL (bewa.id, bewa.rev) = PRIMARY KEY]
bewa.rev         bigint(20)       [NOT NULL (bewa.id, bewa.rev) = PRIMARY KEY, (FOREIGN) KEY - index: fk_bew_aud_revinfo28]
bewa.prio        int(11)          [index: tmp_idx_bewa_prio]
ri.id            bigint(20)       [PRIMARY KEY, NOT NULL]
sa.id            bigint(20)       [PRIMARY KEY, NOT NULL]
sa.vp_id         bigint(20)       [(FOREIGN) KEY, NOT NULL - index: fk_sa_vp141]

All columns above are indexed either as PRIMARY KEY or as single-column indexes. There are no compound indexes so far (I have tried a few but the plan did not change so I dropped them again).

Also here's some numbers:

select count(*) from sa where sa.vp_id = 6;
> 179

select count(*) from bew;
> 240808

select count(*) from bew 
inner join sa on bew.sa_id = sa.id 
where sa.vp_id = 6 and bew.person_id > 0;
> 199955

select count(*) from bewa
> 5126493

select count(*)
from bew 
inner join bewa on bewa.id = bew.id
inner join sa on bew.sa_id = sa.id
inner join revinfo ri on bewa.rev = ri.id
where sa.vp_id = 6
  and bewa.prio is not null
  and bewa.rev > 0
  and bew.person_id > 0;
> (still counting after 10 min ... :( )

This is the explain output:

*************************** 1. row ***************************
           id: 1
  select_type: SIMPLE
        table: bew
         type: ALL
possible_keys: PRIMARY,fk_bew_person26,fk_bew_sa27
          key: NULL
      key_len: NULL
          ref: NULL
         rows: 241594
        Extra: Using where; Using temporary; Using filesort
*************************** 2. row ***************************
           id: 1
  select_type: SIMPLE
        table: sa
         type: eq_ref
possible_keys: PRIMARY,fk_sa_vp141
          key: PRIMARY
      key_len: 8
          ref: bew.sa_id
         rows: 1
        Extra: Using where
*************************** 3. row ***************************
           id: 1
  select_type: SIMPLE
        table: bewa
         type: ref
possible_keys: PRIMARY,fk_bew_aud_revinfo28,tmp_idx_bewa_prio
          key: PRIMARY
      key_len: 8
          ref: bew.id
         rows: 6
        Extra: Using where
*************************** 4. row ***************************
           id: 1
  select_type: SIMPLE
        table: ri
         type: eq_ref
possible_keys: PRIMARY
          key: PRIMARY
      key_len: 8
          ref: bewa.REV
         rows: 1
        Extra: 

There's not much I can do to actually optimize the SQL (it's quite an amount of data and I need it sorted...) and since I only need to run it once that does not really matter, anyway.

So just take this query as an example for my actual, broader question:

How can I tweak the settings in my my.cnf file for such (reporting) queries to run faster than they are now? Actually, what I want is to give the single session as much memory as possible to perform a single query. I don't need things like query caches or good multi-user performance. Given about 6GB of free memory, how can I configure the mysql server to use as much of it as possible to compute one single query?

Answer 1

Ok, I have googled and experimented a bit. I set the following values in my.cnf:

innodb_buffer_pool_size = 768M
sort_buffer_size        = 64M 
key_buffer_size         = 64M
read_buffer_size        = 64M

Which brings my query down to 20 minutes which is ok for me.

Some references:

• http://www.mysqlperformanceblog.com/2007/11/03/choosing-innodb_buffer_pool_size/
• http://www.mysqlperformanceblog.com/2010/10/25/impact-of-the-sort-buffer-size-in-mysql/
• There is also a tool: http://mysqltuner.com/ which gives hints about configuration (I had to interpret the results to be applicable my particular case, though)