Select statement can trigger dead lock on table in mysql? - mysql

Question

The SQL below is inside a MySQL stored procedure.
The procedure run by a cron job every day once at midnight to populate report table with result.
this procedure take around 2 min to run.
please note that table1 has millions of records.
i put this to run at midnight because there are INSERT/UPDATE transactions during the day but unfortunately there are some few transaction at night also.
when this procedure runs and if there are other transactions running then a deadlock error on table1 occurs.
my question is
why SELECT statement cause deadlock on table1?
how can I avoid deadlock in this kind of situation?
DROP report;
CREATE TABLE IF NOT EXISTS report AS (
SELECT
DISTINCT
companies.id company_id,
(
SELECT
SUM(`message_count`) single_phone
FROM
`table1`
WHERE
`table1`.`company_id` = companies.id
AND
`status` != 'error'
) AS single_phone,
(
SELECT
SUM(`message_count`)
FROM
`table1`
WHERE
`table1`.`company_id` = companies.id
AND
`status` != 'not error'
) AS log,
(
SELECT
SUM(`message_count`)
FROM
`table1`
WHERE
`table1`.`company_id` = companies.id
AND
`status` != 'error'
) AS log_monthly,
(
SELECT
SUM(`number_of_sms`) AS aggregate
FROM
`messages`
WHERE
`messages`.`company_id` = companies.id
) AS p_monthly
FROM
companies
INNER JOIN company_users ON companies.id = company_users.company_id
WHERE
company_users.confirmed = 1
AND
company_users.deleted_at IS NULL
);

Answer 1

thanks you very much for help but i have found the problem. yes this procedure cause the deadlock on table but the actual cause of the issue is that i have put ->everyMinute() in my laravel Kernal for schedule run. and there is also a cron job configured by another developer for the same that run every minute. these will run schedule every minute and that is the real cause of the deadlock problem. i have change my Kernal schedule to ->dailyAt('02:00'); now the problem is solved.

Answer 2

Your field-level queries should be done ONCE in the from clause to get pre-aggregates done ONCE per company ID and left-joined in case a given company may NOT have qualified records in a given category. Additionally, your query to get Single_Phone is the same as your 'log_monthly', but have no criteria showing a
break or filter on the dates of activity to filter out a single month vs overall total of everything. So, I added a where clause for filtering, but only GUESSING if such some date exists.
This query might substantially improve your performance. By moving the COLUMN-based queries for every company ID into its own subquery via left-join, those will be summed() and grouped by company ONCE, then the JOIN for the final result. COALESCE() is used so if no such counts exists, the value returned will be 0 instead of null
DROP report;
CREATE TABLE IF NOT EXISTS report AS (
SELECT
c.id company_id,
coalesce( PhoneSum.Msgs, 0 ) as Single_Phone,
coalesce( PhoneLog.Msgs, 0 ) as Log,
coalesce( MonthLog.Msgs, 0 ) as Log_Monthly,
coalesce( SMSSummary.Aggregate, 0 ) as p_monthly
from
-- this will declare an in-line variable if you do need to filter by a month as a couple of your
-- column result names infer, but have no other indicator of filtering by a given month.
( select #yesterday := date_sub( date(curdate()), interval -1 day ),
#beginOfThatMonth := date_sub( #yesterday, interval dayOfMonth( #yesterday ) -1 day ) sqlvars,
companies c
INNER JOIN company_users cu
ON m.company.id = cu.company_id
AND cu.confirmed = 1
AND cu.deleted_at IS NULL
LEFT JOIN
( SELECT
t.company_id,
SUM( t.message_count ) Msgs
FROM
table1 t
INNER JOIN company_users cu
ON t.company.id = cu.company_id
AND cu.confirmed = 1
AND cu.deleted_at IS NULL
where
t.status != 'error'
GROUP BY
t.company_id ) AS PhoneSum,
on c.id = PhoneSum.company_id
LEFT JOIN
( SELECT
t.company_id,
SUM( t.message_count ) Msgs
FROM
table1 t
INNER JOIN company_users cu
ON t.company.id = cu.company_id
AND cu.confirmed = 1
AND cu.deleted_at IS NULL
where
t.status != 'not error'
GROUP BY
t.company_id ) AS PhoneLog,
on c.id = PhoneLog.company_id
LEFT JOIN
( SELECT
t.company_id,
SUM( t.message_count ) Msgs
FROM
table1 t
INNER JOIN company_users cu
ON t.company.id = cu.company_id
AND cu.confirmed = 1
AND cu.deleted_at IS NULL
where
t.status != 'error'
-- this would only get counts of activity for current month currently active
-- but since you are running at night, you need the day before current
AND t.SomeDateFieldOnTable1 >= #beginOfThatMonth
GROUP BY
t.company_id ) AS MonthLogMsgs,
on c.id = MonthLogMsgs.company_id
LEFT JOIN
( SELECT
m.company_id,
SUM( m.number_of_sms ) aggregate
FROM
messages m
INNER JOIN company_users cu
ON m.company.id = cu.company_id
AND cu.confirmed = 1
AND cu.deleted_at IS NULL
where
m.SomeDateFieldOnMessagesTable >= #beginOfThatMonth
GROUP BY
company_id ) AS SMSSummary,
on c.id = SMSSummary.company_id