How to convert a comma separated string into a SET to use in another query in MySQL?

Question

Let us assume that we have two MySQL tables.
master

id  |  master  |  meta
1   |  78564   |  2,3,4
2   |  89567   |  1,3
3   |  45782   |  2,1,4

and metadata

id  |  buyer_id  |  platform
1   |  25        |  mobile
2   |  35        |  web
3   |  25        |  web
4   |  17        |  api

I need to get an array of buyer_ids and platforms when I give the master name. The master column is a unique index and the id of metadata is also a unique index. I want something like:

SELECT buyer_id, platform FROM metadata WHERE id IN (SELECT meta FROM 
master WHERE master = '89567')

The above query doesn't work because the sub query returns a string. I need someway to convert that sub query result into a SET. I am currently using PHP to do it.

<?php $sql="SELECT meta FROM master WHERE master = '89567'";
$result = $connection->query($sql);
$result = $result->fetch_assoc();
$meta = $result['meta'];
$sql="SELECT buyer_id, platform FROM metadata WHERE id IN (".$meta.")";
$result = $connection->query($sql);?>

I am getting the required result but I want to optimize it. I want the whole thing to be done in single query using a sub query or a JOIN.

Background Story : I first used only one table to store all the data. It was serialized too. I did not use any indexes and the execution time was 0.290010929.
I then used non unique index for master column of master table. I was still using only one table and the execution time was 0.0186100006.
I then split the table into two with one unique index per each table as I explained above. I am using the above PHP code and the execution time is 0.0011548995.

I am sure that I can bring the execution time down to 0.0005 if you can answer my question. I am doing this for a very high traffic RESTful api. I want to optimize it to my best. Please help me.

Update : Thanks to @RiggsFolly.
I changed the tables' structure.
master

id  |  master(unique index)
1   |  78564
2   |  89567
3   |  45782

and metadata

id  |  master_id  |  buyer_id  |  platform
1   |     2       |  25        |  mobile
2   |     3       |  25        |  mobile
3   |     1       |  35        |  web
4   |     3       |  35        |  web
5   |     1       |  25        |  web
6   |     2       |  25        |  web
7   |     1       |  17        |  api
8   |     3       |  17        |  api

master_id is a non unique index.
Now, my query is

SELECT buyer_id, platform FROM metadata WHERE master_id = (SELECT id FROM master WHERE 
master = '89567')

and the execution time is 0.00079298019.
I do not think the execution time would be lesser even if I could convert a string into SET. But, I still want to know the answer to my question.

Answer 1

Keep a pre-computed table using a trigger.

create table master_metadata(
    masterid bigint(10),
    metaid   bigint(10)
);

Your query, then, should be:

SELECT buyer_id, platform
FROM metadata
WHERE id IN(
        SELECT metaid
        FROM master_metadata
        WHERE masterid = '89567'
    );

The code that would mantain this table follows.

drop procedure if exists sp_tr_master_insert_update;
drop trigger if exists tr_master_insert;
drop trigger if exists tr_master_update;

delimiter //

create
    definer=root@localhost
    procedure sp_tr_master_insert_update(
            in _masterid bigint(10)
        )
modifies sql data
reads sql data
begin
    declare metaids varchar(255);
    --
    declare commapos int;
    declare metaid   varchar(255);

    -- -------------------------------------------------------------------------

    delete from master_metadata
    where masterid = _masterid;

    -- -------------------------------------------------------------------------

    set metaids =(
            select meta
            from master
            where master = _masterid
        );

    if coalesce(metaids, '') <> '' then
        _split: LOOP
            set commapos = locate(',', metaids);

            if commapos = 0 then
                insert into master_metadata
                values(_masterid, metaids);

                leave _split;
            else
                set metaid = left(metaids, commapos - 1);
                set metaids = right(metaids, length(metaids) - commapos);

                insert into master_metadata
                values(_masterid, metaid);                
            end if;
        END LOOP _split;
    end if;
end //

CREATE DEFINER=root@localhost TRIGGER tr_master_insert
AFTER insert ON master
FOR EACH ROW
BEGIN
    CALL sp_tr_master_insert_update(NEW.master);
END//

CREATE DEFINER=root@localhost TRIGGER tr_master_update
AFTER update ON master
FOR EACH ROW
BEGIN
    CALL sp_tr_master_insert_update(NEW.master);
END//


delimiter ;

Index creation and other performance tweaks is up to you.