MySQL managed database is configured to use inodbfileper_table=ON which means that for every table its data and indexes are stored together in a separate .idb file.
Over time, when a table receives a lot of inserts and deletes, the amount of space it occupies on disk can grow significantly larger than the current data in the table. A classic example of this would be a table containing jobs for a work queue in which rows are repeatedly added to the end of the table and removed from the beginning.
InnoDB does not release this allocated space back to the operating system automatically (in case the table grows larger in the future) but this can cause problems. Since every other table exists in its own .idb file, the allocated but unused space is unavailable for them to grow into.
Identify problem tables
First run the following query to identify tables with significant allocated but unused space.
(TABLES.DATALENGTH + TABLES.INDEXLENGTH) / 1024 / 1024 AS “MB used (estimate)”,
TABLES.DATAFREE / 1024 / 1024 AS “MB allocated but unused (estimate)”,
INNODBTABLESPACES.FILESIZE / 1024 / 1024 AS “MB on disk”
JOIN informationschema.INNODBTABLESPACES ON (INNODBTABLESPACES.NAME = TABLES.TABLESCHEMA || ’/’ || TABLES.TABLENAME)
WHERE INNODBTABLESPACES.FILESIZE > 10 * 1024 * 1024
ORDER BY TABLES.DATAFREE DESC;
You can see from the query results that if a table has significantly more allocated but unused space than used space then that increases the size of the table on disk. Note that, by default, statistics in informationschema.TABLES are only updated every 24 hours or whenever the ANALYZETABLE command is run.
Reclaim disk space
For tables with a high ratio of allocated but unused space to used space, you can ask InnoDB to release disk space back to the operating system by running the OPTIMIZE TABLE command.
Beware that under certain conditions(e.g. including the presence of a FULLTEXT index) this command will copy the data to a new table containing just the current data and then drop and rename the new table to match the old one. During this process data modification will be blocked and it will require enough free space to store two copies of the current data at once.