The slow query log costs less than you think to leave on

The slow query log is usually described as a debugging tool to switch on when there is a problem, which means it is off at exactly the moment the problem happens and nobody can say which query was responsible.

SET GLOBAL slow_query_log      = 1;
SET GLOBAL long_query_time      = 1;
SET GLOBAL log_queries_not_using_indexes = 0;

-- then, over the accumulated log
-- mysqldumpslow -s t -t 20 /var/log/mysql/slow.log

At a one-second threshold the write volume on a healthy server is a handful of lines a day, and the overhead is a timestamp comparison per query. Leave log_queries_not_using_indexes off unless you are actively auditing — on a schema with small lookup tables it fills the log with scans that are entirely correct. mysqldumpslow groups by normalised query, which is what makes the log readable at all.