Bug Lock contention on thread-read tracking under very high load

There is a bug in this version
X

Xon

Guest
I've had a client report that under very high number of readers, the insert on duplicate key update query that is part of markThreadReadByUser causes snowballing slow queries due to lock contention.

MySQL was set to report deadlocks, but did not report anything during this catastrophically slow queries. innodb_flush_log_at_trx_commit had the default value of 1, but changing it to 2 did not appear to help.

A sample of the slow query log for this query...

Read more

Continue reading...