Known issues with MariaDB
-
As of MariaDB 10.2 the
DATETIMEdata type supports sub-second fractions. The default of MariaDB internally is to use no fractions. The number of fractions can be explicitly set via scale:@Column(scale=6)will lead to aDATETIME(6)column and able to store microseconds. A value of@Column(scale=-1)will explicitly turn off all fractions. -
As of MariaDB 10.7 the
UUIDdata type is supported, but the JDBC driver does not writes the object directly, so UUID field support is made through UUID to String conversion. -
MariaDB’s InnoDB performs
ALTER TABLE … ADD COLUMNandALTER TABLE … DROP COLUMNinstantly and keeps the dropped columns in the clustered index record format, where they continue to count towards the 8126 byte maximum row size. Re-synchronising a mapping with a drop/add schema action against the same table - eitheropenjpa.jdbc.SynchronizeMappingsset tobuildSchema(SchemaAction='drop,add'), ormappingtool -action refreshin a deployment script - therefore accumulates that metadata on every cycle and eventually fails with
Row size too large. The maximum row size for the used table type, not counting BLOBs, is 8126. [code=1118, state=42000]
even though the table’s visible columns fit well inside the limit; information_schema gives no indication of the accumulated metadata.
Rebuild the table to discard it (ALTER TABLE <name> FORCE or OPTIMIZE TABLE <name>), or do not use a drop/add schema action against a long-lived MariaDB schema.