What can cause TRUNCATE TABLE to take a really long time?

Question

I'm running MySQL5.5 with Master/Slave replication (1 master, 2 slaves).

I have a process that runs once a week and truncate a specific table. The table is not large and only has a few thousands records.

For some reason, the TRUNCATE TABLE command takes really long time to execute (both on the master and on the slave). It takes about 400K ms to execute!! When it runs on the slave, it causes it to lag from the Master. After the TRUNCATE TABLE finishes, everything is back to normal.

I know that one of the slaves didn't receive any reads while performing the TRUNCATE TABLE since its a dedicated slave and the process that reads from that slave was down. Also, on this slave, it took the same amount of time to execute.

Here is the table structure: http://pastebin.com/qEQB4juR

Any thoughts on how I can speed up the TRUNCATE TABLE ?

Answer 1

Using TRUNCATE TABLE on an InnoDB table requires a full table lock because TRUNCATE TABLE is DDL (Data Definition Language) not DML (Data Manipulation).

Doing DELETE FROM user_engagements; will not help because MVCC info is written to the undo logs in ibdata1, and that can hold up the table from being emptied. If any uncommitted transactions are holding onto user_engagements, that could potentially hold up a TRUNCATE TABLE as well.

You could rename the table so that it is immediately available

SET FOREIGN_KEY_CHECKS = 0;
CREATE TABLE user_engagements_new LIKE user_engagements;
ALTER TABLE user_engagements RENAME user_engagements_zap;
ALTER TABLE user_engagements_new RENAME user_engagements;
DROP TABLE user_engagements_zap;
SET FOREIGN_KEY_CHECKS = 1;

This should replicate quickly except for the last statement.

Give it a Try !!!

If you have MySQL 5.1.16+, TRUNCATE TABLE requires DROP privilege. My answer performs what TRUNCATE TABLE now does.

If you have MySQL 5.1.15 and back, you need DELETE privilege, which my answer covers.

Answer 2

First of all you should use latest version of MySQL or MariaDB as in these version this bug has been fixed.

This bug can come up when you DROP or TRUNCATE a table and you have a large buffer pool, there can be server lock-up or stall of multiple seconds which makes your table unresponsive.

Depending on the server’s CPU and memory bandwidth, the stall can be as much as 1 second per 32GB of RAM. If your buffer pool is 1TB that can mean your entire database locks up for as long as 30 seconds. There is different workaround for this bug depending on your circumstances as you have not shared complete detail so, I am suggesting some. You can also find more details MySQL troubleshooting article.

One possible solution can be delete all rows before dropping a table, this will not cause a memory stall because there will be no data in memory.

Another solution can be use of non-InnoDB storage engine for temporary tables. You can specify ENGINE=MyISAM in your CREATE TABLE statement. If you have a large code base and don’t explicitly specify the storage engine, you can change the default storage engine to MyISAM by adjusting the following configuration option:

default_tmp_storage_engine = MyISAM

The good news is that the fix itself has been implemented MariaDB 10.2.19, MySQL 8.0.23 and later.