Table converted from myISAM to INNODB is slowing down a query

Question

I have a table that was converted from myISAM to INNODB that is slowing down a query. This is a big table with lot of indexes. MyIsam (on mysql5.6) returns result instantly, INNODB (on mysql5.7) takes 2 to 3 seconds. fnota is float. serieid and epnumber are int. Any idea why this is taking more time when doing index_merge?

Explain on MYISAM TABLE:

explain SELECT count( fnota ) , avg( fnota ) FROM myrates
          WHERE serieid =4376 AND epnumber ='149'\G
*************************** 1. row ***************************
           id: 1
  select_type: SIMPLE
        table: myrates
         type: ref
possible_keys: serieid,epnumber,serieid_2
          key: serieid_2
      key_len: 8
          ref: const,const
         rows: 8207
        Extra: Using index condition

Explain on INNODB Table:

explain SELECT count( fnota ) , avg( fnota ) FROM myrates
          WHERE serieid =4376 AND epnumber ='149'\G
*************************** 1. row ***************************
          id: 1
 select_type: SIMPLE
       table: myrates
  partitions: NULL
        type: index_merge
possible_keys: serieid,epnumber,serieid_2
         key: serieid,epnumber,serieid_2
     key_len: 4,4,8
         ref: NULL
        rows: 2
    filtered: 100.00
       Extra: Using intersect(serieid,epnumber,serieid_2); Using where

Indexes:

SHOW INDEX FROM myrates;
+---------+------------+-----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| Table   | Non_unique | Key_name  | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment |
+---------+------------+-----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
| myrates |          0 | fbid      |            1 | userid      | A         |     1405506 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          0 | fbid      |            2 | serieid     | A         |     8617224 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          0 | fbid      |            3 | epnumber    | A         |   139638192 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          1 | serieid   |            1 | serieid     | A         |      257656 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          1 | epnumber  |            1 | epnumber    | A         |       93431 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          1 | serieid_2 |            1 | serieid     | A         |      186213 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          1 | serieid_2 |            2 | epnumber    | A         |     3309332 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          1 | userid    |            1 | userid      | A         |      866339 |     NULL | NULL   |      | BTREE      |         |               |
| myrates |          1 | userid    |            2 | serieid     | A         |     4656575 |     NULL | NULL   |      | BTREE      |         |               |
+---------+------------+-----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+

SHOW CREATE TABLE MYRATES; 
Table: myrates
Create Table: CREATE TABLE `myrates` (
  `userid` bigint(10) NOT NULL,
  `serieid` int(6) NOT NULL,
  `epnumber` float NOT NULL,
  `nota` int(6) NOT NULL DEFAULT '0',
  `fnota` float NOT NULL,
  `timestamp` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
  UNIQUE KEY `fbid` (`userid`,`serieid`,`epnumber`),
  KEY `serieid` (`serieid`),
  KEY `epnumber` (`epnumber`),
  KEY `serieid_2` (`serieid`,`epnumber`),
  KEY `userid` (`userid`,`serieid`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1
1 row in set (0.00 sec)

Answer 1

DESCRIBE does not precisely spell out the indexes you have. Please provide SHOW CREATE TABLE . It sounds like it is getting confused over your long list of overlapping indexes.

When you have a composite index such as (serieid, epnumber) you don't also need (serieid) . Drop the latter index to 'fix' the problem.

It seems that key "userid" is redundant also, given that "fbid" starts with both of its two columns in the same order.