I am trying to select from two much big table using join:
EXPLAIN SELECT SQL_NO_CACHE
e.*
FROM `table_A` e
JOIN
(SELECT id FROM `table_B` /* FORCE index (primary, index_A) */
WHERE id > 338107 AND `index_field_A` = 900000000 AND `index_field_B` = 1
ORDER BY id) AS c
ON `c`.`id` = `e`.`fk_id`
WHERE e.`some_field` IS NULL;
LIMIT 2000;
/* BEST EXPLAIN RESULT
USING intersect(index_A,index_B); USING WHERE; USING INDEX
*/
I store current id at the application side (338107) to be able to get total result by batching (from start id to max id)
There are no problem if I get rid off ORDER BY id But I am not sure that MySQL order PK by default.
There are no problem TOO if I use separate SELECT without JOIN :
SELECT id FROM `table_B` WHERE id > 338107
AND `index_field_A` = 900000000 AND `index_field_B` = 1 ORDER BY id
but it is useless
A little better EXPLAIN I can get if I forcing index :
FORCE index (primary, index_A)
But it too far from good.
Can I get rid of ORDER BY id without negative aftermath?
Addition : ordering PK field is autoincrement, InnoDB tables
What about something like this?
SELECT *
FROM `table_A`
WHERE `some_field` IS NULL
AND `fk_id` IN (
SELECT `id`
FROM `table_B`
WHERE id > 338107 AND `index_field_A` = 900000000 AND `index_field_B` = 1
)
ORDER BY `fk_id`
;
or
SELECT e.*
FROM `table_B` AS c
LEFT JOIN `table_A` AS e
ON c.`id` = e.`fk_id`
AND e.`someField` IS NULL
WHERE c.id > 338107
AND c.`index_field_A` = 900000000
AND c.`index_field_B` = 1
HAVING e.fk_id IS NOT NULL
ORDER BY e.`fk_id`
;
Related
I have 2 tables.
How do i search for all rows in the first table that has no reference in the second table.
The connection field is: res_srvs.id = inv_supp2srv.srvID
So, I want to get all table "res_srvs" rows that has no srvID in table "inv_supp2srv".
TABLE: res_srvs
Collation Attributes
id int(11)
clientID int(6)
resNum int(9)
net decimal(7,2)
tax decimal(7,2)
from_date(date)
TABLE: inv_supp2srv
Collation Attributes
clientID int(6)
invNum int(10)
srvID int(11)
amount decimal(7,2)
valid tinyint(1)
This is what i tried:
SELECT srv.net , srv.tax , srv.net+srv.tax AS amount, srv.id AS srv_id
FROM res_srvs AS srv , inv_supp2srv AS i2s
WHERE srv.clientID = 1
AND srv.from_date >= '2020-03-01'
AND i2s.clientID = 1
AND i2s.srvID = srv.id
AND (NOT EXISTS
(
SELECT *
FROM inv_supp2srv AS i2s
WHERE i2s.srvID = srv.id
)
)
What you want is a left outer join with exclusion :
SELECT r.*
FROM res_srvs r
LEFT JOIN inv_supp2srv i
ON r.id = i.srvID
WHERE i.srvID IS NULL
AND (
-- Your others where clauses go there
);
You can use LEFT JOIN for second table and filter by NULL joined value like:
SELECT srv.net , srv.tax , srv.net+srv.tax AS amount, srv.id AS srv_id
FROM res_srvs AS srv
LEFT JOIN inv_supp2srv AS i2s ON i2s.srvID = srv.id
WHERE
srv.clientID = 1
AND srv.from_date >= '2020-03-01'
-- AND i2s.clientID = 1 not relevant condition
AND i2s.srvID IS NULL;
Another approach is using NOT EXISTS condition:
SELECT srv.net , srv.tax , srv.net+srv.tax AS amount, srv.id AS srv_id
FROM res_srvs AS srv
WHERE
srv.clientID = 1
AND srv.from_date >= '2020-03-01'
AND NOT EXISTS (
SELECT srvID FROM inv_supp2srv AS i2s WHERE i2s.srvID = srv.id
);
I want to get all table res_srvs rows that have no srvID in table inv_supp2srv.
It looks like you are overcomplicating this. I don't see the point for the join between the tables in the outer query - it attempts to match the tables, which contradicts the not exists condition.
I think you just want:
select r.*
from res_srvs r
where
r.from_date >= '2020-03-01'
and r.clientID = 1
and not exists (
select 1
from inv_supp2srv i
where i.srvID = r.id and i.clientID = r.clientID
)
I am unsure whether you want clientID in the correlation clause or not - your query makes it look like it is the case, so I added it.
I need to do a join of 3 tables.
They all have an ID value, the issue is:
Table 1
Is the main one, name : cnpj_cnae and the ID is CNPJ_CNAE
Table 2
I need to get four columns from this one (Up to here I already got it).
The ID here is CNPJ_CNAE
Table 3
Here's the issue. For each ID value on table 1 I can have more than 1 even 10 on this table. I need to get 4 columns from ONLY THE FIRST TWO of this table.
The ID here is CNPJ_SOCIO
The finishing table would need to look like
ID INFOTABLE1 INFOTABLE2 INFOTABLE3ROW1 INFOTABLE3ROW2
I have tried some joins, but being a relative newbie with mysql I am suffering >.<
I have tried to do this:
CREATE TABLE cnpj_cnae_emp_test3 AS (
SELECT
`cnpj_cnae_test2`.*,
`cnpj_soci`.NOME_SOCIO,`cnpj_soci`.CNPJ_CPF_SOCI,`cnpj_soci`.ID_QUALIFICACAO_REP,`cnpj_soci`.DESC_QUALIFICACAO_REP,
`cnpj_emp_02`.MUNICIPIO,`cnpj_emp_02`.BAIRRO,`cnpj_emp_02`.TIPO_LOGRADOURO,`cnpj_emp_02`.LOGRADOURO,`cnpj_emp_02`.NUMERO,`cnpj_emp_02`.COMPLEMENTO
FROM `cnpj_cnae_test2`, `cnpj_soci`, `cnpj_emp_02`
WHERE `cnpj_cnae_test2`.CNPJ_CNAE = `cnpj_soci`.CNPJ_SOCIO AND `cnpj_cnae_test2`.CNPJ_CNAE = `cnpj_emp_02`.CNPJ
);
The issue is this will generate extra rows for every SOCI where instead of 1 soci on each one I need 2 socis on each line.
Table 1:
Layout of CNAE table
Table 2:
Layout of EMP table
Table 3:
Layout of Socio table
The resulting table would look like this:
More detailed end result Final layout with descriptions
Layout of result
In the end this by Owl was exactly what I needed:
WITH soci_partition AS (
SELECT CNPJ_SOCIO
,NOME_SOCIO
,CNPJ_CPF_SOCIO
,ID_QUALIFICACAO
,DESC_QUALIFICACAO_SOCIO
,row_number() OVER (PARTITION BY CNPJ_SOCIO ORDER BY IDX ASC) AS rownum
FROM cnpj_soci
)
SELECT * -- add actual columns
FROM cnpj_cnae_test2 AS cnae
INNER JOIN cnpj_emp_02 AS emp -- left join instead? depends on table structure
ON cnae.ID_CNAE = emp.IDX
LEFT JOIN soci_partition AS soci1
ON soci1.CNPJ_SOCIO = cnae.CNPJ_CNAE and soci1.rownum = 1
LEFT JOIN soci_partition AS soci2
ON soci2.CNPJ_SOCIO = cnae.CNPJ_CNAE and soci2.rownum = 2
Not sure if I got the joins/partition right, the foreign keys weren't clear.
Even if it's not perfect, hopefully this explains the general theory. Please edit in your correct answer if you end up debugging.
Also don't use old-style joins. You'll immediately fail interviews on that alone; they were deprecated 25 years ago.
WITH soci_partition AS (
SELECT CNPJ_SOCIO
,NOME_SOCIO
,CNPJ_CPF_SOCIO
,ID_QUALIFICACAO
,DESC_QUALIFICACAO_SOCIO
,row_number() OVER (PARTITION BY CNPJ_SOCIO ORDER BY IDX ASC) AS rownum
FROM cnpj_soci
)
SELECT * -- add actual columns
FROM cnpj_cnae_test2 AS cnae
INNER JOIN cnpj_emp_02 AS emp -- left join instead? depends on table structure
ON cnae.ID_CNAE = emp.IDX
LEFT JOIN soci_partition AS soci1
ON soci1.CNPJ_SOCIO = cnae.CNPJ_CNAE and soci1.rownum = 1
LEFT JOIN soci_partition AS soci2
ON soci2.CNPJ_SOCIO = cnae.CNPJ_CNAE and soci2.rownum = 2
Give a try to below query
select * from cnae t1
left join emp t2 on t1.cnpj_cnae = t2.cnpj
left join socio t3 on t1.cnpj_cnae = t2.cnpj_socio
group by idx_cnae ;
I have the following code and I'm trying to group the messages
Here is a picture of database table and how the groups should be
and here is the SQL statement
SELECT a.* FROM `user_messages` `a`
JOIN (
SELECT `sender`, MAX(`id`) `last_id` FROM `user_messages` WHERE `receiver` = '1' GROUP BY `sender`
) `b`
ON `a`.`sender` = `b`.`sender` AND `a`.`id` = `b`.`last_id`
WHERE `a`.`receiver` = '1'
ORDER BY `id` DESC
OUTPUT:
I want to get somehow the last record where "receiver" is not my id, but "sender" is and name receiver column as "id" or something.
...so what i want is following result:
id | msg
13852 123
48 Hello!
17 321
Here is a fiddle: http://sqlfiddle.com/#!9/e06d57/3/0
To map my generic answer to your particular use case (using example 1):
SELECT receiver AS id, msg
FROM user_messages outerTable
WHERE NOT EXISTS
( SELECT *
FROM user_messages innerTable
WHERE innerTable.sender = outerTable.sender
AND innerTable.receiver = outerTable.receiver
AND innerTable.added > outerTable.added
)
AND sender = 1
This is a very common use case. There are several ways to write this code. Depending on the SQL engine used, they will be of different speeds.
I will use fairly generic column names. Tweak as needed.
SELECT common_id, msg
FROM myTable outerTable
WHERE NOT EXISTS
( SELECT *
FROM myTable innerTable
WHERE innerTable.common_id = outerTable.common_id
AND innerTable.time > outerTable.time
)
Please note that if there are two rows with identical common_id and time columns, then both will show up in the output. You can replace the > with >= to hide both of those rows.
The other common approach is kind of difficult to make sense of, but here goes. Notice the similarities to the NOT EXISTS approach.
SELECT outerTable.common_id, outerTable.msg
FROM myTable outerTable
LEFT JOIN myTable innerTable
ON innerTable.common_id = outerTable.common_id
AND innerTable.time > outerTable.time
WHERE innerTable.common_id IS NULL
According to your description, you seem to want something like this:
select um.receiver as id, um.msg
from user_messages um
where um.sender = 1 and
um.id = (select max(um2.id)
from user_messages um2
where um2.msg = um.msg and um2.receiver <> 1 and um.sender = 1
);
It doesn't produce the desired output, but that is because the output is inconsistent with the text description.
I have this query:
SELECT `assemblies`.`id`,
`assemblies`.`type`,
`assemblies`.`champion`,
`assemblies`.`name`,
`assemblies`.`author`,
`assemblies`.`githublastmod`,
( assemblies.forum IS NOT NULL ) AS forumExists,
Count(votes.id) AS votesCount,
Count(install_clicks.id) AS installCount,
Count(github_clicks.id) AS githubCount,
Count(forum_clicks.id) AS forumCount
FROM `assemblies`
INNER JOIN `votes`
ON `votes`.`assembly` = `assemblies`.`id`
INNER JOIN `install_clicks`
ON `install_clicks`.`assembly` = `assemblies`.`id`
INNER JOIN `github_clicks`
ON `github_clicks`.`assembly` = `assemblies`.`id`
INNER JOIN `forum_clicks`
ON `forum_clicks`.`assembly` = `assemblies`.`id`
WHERE `assemblies`.`type` = 'utility'
AND Unix_timestamp(Date(assemblies.githublastmod)) > '1419536536'
GROUP BY `assemblies`.`id`
ORDER BY `votescount` DESC,
`githublastmod` DESC
For some reason this query is very slow, I'm using the database engine MyISAM. I hope someone can help me out here :)
Explain command:
I believe this is a case where making the subqueries for the counts will make it run a lot faster (and the values will be correct).
The problem with the original query is the explosion of the number of intermediate rows: For each 'assembly', there were n1 votes, n2 installs, etc. That led to n1*n2*... rows per assembly.
SELECT `assemblies`.`id`, `assemblies`.`type`, `assemblies`.`champion`,
`assemblies`.`name`, `assemblies`.`author`, `assemblies`.`githublastmod`,
( assemblies.forum IS NOT NULL ) AS forumExists,
( SELECT Count(*)
FROM votes
WHERE `assembly` = `assemblies`.`id`
) AS votesCount,
( SELECT Count(*)
FROM install_clicks
WHERE `assembly` = `assemblies`.`id`
) AS installCount,
( SELECT Count(*)
FROM github_clicks
WHERE `assembly` = `assemblies`.`id`
) AS githubCount,
( SELECT Count(*)
FROM forum_clicks.id
WHERE `assembly` = `assemblies`.`id`
) AS forumCount
FROM `assemblies`
WHERE `assemblies`.`type` = 'utility'
AND Unix_timestamp(Date(assemblies.githublastmod)) > '1419536536'
ORDER BY `votescount` DESC, `githublastmod` DESC
Each secondary table needs an INDEX starting with assembly.
Your problem should be fixed using the right indices:
CREATE INDEX index_name_1 ON `votes`(`assembly`);
CREATE INDEX index_name_2 ON `install_clicks`(`assembly`);
CREATE INDEX index_name_3 ON `github_clicks`(`assembly`);
CREATE INDEX index_name_4 ON `forum_clicks`(`assembly`);
Try your query again after creating these indices and it should be quite faster.
I have 4 tables:
Table talks
table talks_fan
table talks_follow
table talks_comments
What I'm trying to achieve is counting all comments, fans, followers for every single talk.
I came up with this so far.
All tables have talk_id and only in talks table is a primary key
SELECT
g. *,
COUNT( m.talk_id ) AS num_of_comments,
COUNT( f.talk_id ) AS num_of_followers
FROM
talks AS g
LEFT JOIN talks_comments AS m
USING ( talk_id )
LEFT JOIN talks_follow AS f
USING ( talk_id )
WHERE g.privacy = 'public'
GROUP BY g.talk_id
ORDER BY g.created_date DESC
LIMIT 30;
I also tried using this method
SELECT
t.*,
COUNT(b.talk_id) AS comments,
COUNT(bt.talk_id) AS followers
FROM
talks t
LEFT JOIN talks_follow bt
ON bt.talk_id = t.talk_id
LEFT JOIN talks_comments b
ON b.talk_id = t.talk_id
GROUP BY t.talk_id;
Both give me the same results ....?!
Update: Create Statements
CREATE TABLE IF NOT EXISTS `talks` (
`talk_id` bigint(20) NOT NULL AUTO_INCREMENT,
`user_id` mediumint(9) NOT NULL,
`title` varchar(255) NOT NULL,
`content` text NOT NULL,
`created_date` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`privacy` enum('public','private') NOT NULL DEFAULT 'private',
PRIMARY KEY (`talk_id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 AUTO_INCREMENT=7 ;
CREATE TABLE IF NOT EXISTS `talks_comments` (
`comment_id` bigint(20) NOT NULL AUTO_INCREMENT,
`talk_id` bigint(20) NOT NULL,
`user_id` mediumint(9) NOT NULL,
`comment` text NOT NULL,
`date_created` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`status` tinyint(1) NOT NULL DEFAULT '0',
PRIMARY KEY (`comment_id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 AUTO_INCREMENT=8 ;
CREATE TABLE IF NOT EXISTS `talks_fan` (
`fan_id` bigint(20) NOT NULL AUTO_INCREMENT,
`talk_id` bigint(20) NOT NULL,
`user_id` bigint(20) NOT NULL,
`created_date` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
`status` tinyint(1) NOT NULL DEFAULT '1',
PRIMARY KEY (`fan_id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 AUTO_INCREMENT=4 ;
CREATE TABLE IF NOT EXISTS `talks_follow` (
`follow_id` bigint(20) NOT NULL AUTO_INCREMENT,
`talk_id` bigint(20) NOT NULL,
`user_id` mediumint(9) NOT NULL,
`date_created` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
PRIMARY KEY (`follow_id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8 AUTO_INCREMENT=5 ;
The final query that works
SELECT t.* , COUNT( DISTINCT b.comment_id ) AS comments,
COUNT( DISTINCT bt.follow_id ) AS followers,
COUNT( DISTINCT c.fan_id ) AS fans
FROM talks t
LEFT JOIN talks_follow bt ON bt.talk_id = t.talk_id
LEFT JOIN talks_comments b ON b.talk_id = t.talk_id
LEFT JOIN talks_fan c ON c.talk_id = t.talk_id
WHERE t.privacy = 'public'
GROUP BY t.talk_id
ORDER BY t.created_date DESC
LIMIT 30
EDIT: Final answer to the whole issue...
I have modified the Query and created some code in PHP (Codeigniter) to solve my issue apone the reccomendation of #Bill Karwin
$sql="
SELECT t.*,
COUNT( DISTINCT b.comment_id ) AS comments,
COUNT( DISTINCT bt.follow_id ) AS followers,
COUNT( DISTINCT c.fan_id ) AS fans,
GROUP_CONCAT( DISTINCT c.user_id ) AS list_of_fans
FROM talks t
LEFT JOIN talks_follow bt ON bt.talk_id = t.talk_id
LEFT JOIN talks_comments b ON b.talk_id = t.talk_id
LEFT JOIN talks_fan c ON c.talk_id = t.talk_id
WHERE t.privacy = 'public'
GROUP BY t.talk_id
ORDER BY t.created_date DESC
LIMIT 30
";
$query = $this->db->query($sql);
if($query->num_rows() > 0)
{
$results = array();
foreach($query->result_array() AS $talk){
$fan_user_id = explode(",", $talk['list_of_fans']);
foreach($fan_user_id AS $user){
if($user == 1 /* this supposed to be user id or session*/){
$talk['list_of_fans'] = 'yes';
}
}
$follower_user_id = explode(",", $talk['list_of_follower']);
foreach($follower_user_id AS $user){
if($user == 1 /* this supposed to be user id or session*/){
$talk['list_of_follower'] = 'yes';
}
}
$results[] = array(
'talk_id' => $talk['talk_id'],
'user_id' => $talk['user_id'],
'title' => $talk['title'],
'created_date' => $talk['created_date'],
'comments' => $talk['comments'],
'followers' => $talk['followers'],
'fans' => $talk['fans'],
'list_of_fans' => $talk['list_of_fans'],
'list_of_follower' => $talk['list_of_follower']
);
}
}
I STILL BELIEVE IT COULD BE OPTIMIZED IN THE DB AND JUST USE THE RESULT...
Im thinking if there are 1000 follower and 2000 fans of every single TALK then the result will take much longer to load.. HOW IF YOUT MULTIPLY THE NO WITH 10. Or im mistaking hear...
EDIT: adding benchmark for the query test...
I have used codeigniter profiler to know how long it take for the query to finish excuting.
that been said i also start adding data in the tables gratually
the result as follows.
Testing the DB after answerting data into it
Query Results time
table Talks
---------------
table data 50 rows.
Time: 0.0173 seconds
Table Rows: 644 rows
Time: 0.0535 seconds
Table Rows: 1250 rows
Time: 0.0856 seconds
Adding data to other tables
--------------------------
Talks = 1250 rows
talks_follow = 4115
talks_fan = 10 rows
Time: 2.656 seconds
Adding data to other tables
--------------------------
Talks = 1250 rows
talks_follow = 4115
talks_fan = 10 rows
talks_comments = 3650 rows
Time: 10.156 seconds
After replacing LEFT JOIN with STRAIGHT_JOIN
Time: 6.675 seconds
It seems that its extremely heavy on the DB.....
NOW Im Going to another dilemma on how to enhance its performance
Edited: using #leonardo_assumpcao suggestion
After rebuilding the DB using #leonardo_assumpcao suggestion
for indexing few fields..........
Adding data to other tables
--------------------------
Talks = 6000 Rows
talks_follow = 10000 Rows
talks_fan = 10000 Rows
talks_comments = 10000 Rows
Time: 17.940 second
Is this normal for heavy data DB......?
I can say this is (at least) one of the coolest select statements I improved today.
SELECT STRAIGHT_JOIN
t.* ,
COUNT( DISTINCT b.comment_id ) AS comments,
COUNT( DISTINCT bt.follow_id ) AS followers,
COUNT( DISTINCT c.fan_id ) AS fans
FROM
(
SELECT * FROM talks
WHERE privacy = 'public'
ORDER BY created_date DESC
LIMIT 0, 30
) AS t
LEFT JOIN talks_follow bt ON (bt.talk_id = t.talk_id)
LEFT JOIN talks_comments b ON (b.talk_id = t.talk_id)
LEFT JOIN talks_fan c ON (c.talk_id = t.talk_id)
GROUP BY t.talk_id ;
But it seems to me that your problem resides on your tables; A first step to obtain efficient queries is to index every field involved on your desired joins.
I've made some modifications on the tables you shown above; You can see its code here (updated).
Quite interesting, isn't it? Since we're here, take also your ERR model:
First try it using MySQL test database. Hopefully it will solve your performance troubles.
(Forgive my english, it's my second language)
You can force this into one query like so:
SELECT COUNT(*) num, 'talks' item FROM talks
UNION
SELECT COUNT(*) num, 'talks_fan' item FROM talks_fan
UNION
SELECT COUNT(*) num, 'talks_follow' item FROM talks_follow
UNION
SELECT COUNT(*) num, 'talks_comment' item FROM talks_comment
This will give you a five row resultset with one row per table. Each row is the count in a particular table.
If you must get it all into a single row you can do a pivot like so.
SELECT
SUM( CASE item WHEN 'talks' THEN num ELSE 0 END ) AS 'talks',
SUM( CASE item WHEN 'talks_fan' THEN num ELSE 0 END ) AS 'talks_fan',
SUM( CASE item WHEN 'talks_follow' THEN num ELSE 0 END ) AS 'talks_follow',
SUM( CASE item WHEN 'talks_comment' THEN num ELSE 0 END ) AS 'talks_comment'
FROM
( SELECT COUNT(*) num, 'talks' item FROM talks
UNION
SELECT COUNT(*) num, 'talks_fan' item FROM talks_fan
UNION
SELECT COUNT(*) num, 'talks_follow' item FROM talks_follow
UNION
SELECT COUNT(*) num, 'talks_comment' item FROM talks_comment
) counts
(This doesn't take into account your WHERE g.privacy = clause because I don't understand that. But you could add a WHERE clause to one one of the four queries in the UNION item to handle that.)
Notice that this truly is four queries on four separate tables coerced into a single query.
And, by the way, there is no difference in value between COUNT(*) and COUNT(id) when id is the primary key of the table. COUNT(id) doesn't count the rows for which the id is NULL, but if id is the primary key, then it is NOT NULL. But COUNT(*) is faster, so use it.
Edit if you need the number of fan, follow, and comment rows for each distinct talk, do this. It's the same idea of doing a union and a pivot, but with an extra parameter.
SELECT
talk_id,
SUM( CASE item WHEN 'talks_fan' THEN num ELSE 0 END ) AS 'talks_fan',
SUM( CASE item WHEN 'talks_follow' THEN num ELSE 0 END ) AS 'talks_follow',
SUM( CASE item WHEN 'talks_comment' THEN num ELSE 0 END ) AS 'talks_comment'
FROM
(
SELECT talk_id, COUNT(*) num, 'talks_fan' item
FROM talks_fan
GROUP BY talk_id
UNION
SELECT talk_id, COUNT(*) num, 'talks_follow' item
FROM talks_follow
GROUP BY talk_id
UNION
SELECT talk_id, COUNT(*) num, 'talks_comment' item
FROM talks_comment
GROUP BY talk_id
) counts
GROUP BY talk_id
After doing this for (too) many years, I've discovered that the best way to describe a query you need is to say to yourself "I need a result set with one row for each xxx, with columns for yyy, zzz, and qqq."
The reason the counts are the same is that it's counting rows after the joins have combined the tables. By joining to multiple tables, you're creating a Cartesian product.
Basically, you're counting not only how many comments per talk, but how many comments * followers per talk. Then you count the followers as how many followers * comments per talk. Thus the counts are the same, and they're all way too high.
Here's a simpler way to write a query to count each distinct comment, follower, etc. only once:
SELECT t.*,
COUNT(DISTINCT b.comment_id) AS comments,
COUNT(DISTINCT bt.follow_id) AS followers
FROM talks t
LEFT JOIN talks_follow bt ON bt.talk_id = t.talk_id
LEFT JOIN talks_comments b ON b.talk_id = t.talk_id
GROUP BY t.talk_id;
Re your comment: I wouldn't fetch all the followers in the same query. You could do it this way:
SELECT t.*,
COUNT(DISTINCT b.comment_id) AS comments,
COUNT(DISTINCT bt.follow_id) AS followers,
GROUP_CONCAT(DISTINCT bt.follower_name) AS list_of_followers
FROM talks t
LEFT JOIN talks_follow bt ON bt.talk_id = t.talk_id
LEFT JOIN talks_comments b ON b.talk_id = t.talk_id
GROUP BY t.talk_id;
But what you'd get back is a single string with the follower names separated by commas. Now you have to write application code to split the string on commas, you have to worry if some follower names actually contain commas already, and so on.
I'd do a second query, fetching the followers for a given talk. It's likely you want to display the followers only for a specific talk anyway.
SELECT follower_name
FROM talks_follow
WHERE talk_id = ?