Checkpoints

Disk checkpoint speed is adaptive, speeds up when required and slows down when not required. Checkpoints are partial which makes it possible to scale the in-memory size to very large sizes.

Concurrency Control

RonDB uses strict two-phase locking when using locks. READ COMMITTED is lockless and always reads the latest committed value. READ COMMITTED is default from MySQL.

Data Model

Fault Tolerance

Within one cluster, RonDB uses synchronous replication. For multi-region replication, RonDB supports both asynchronous active-passive and active-active (with conflict resolution) between clusters.

Foreign Keys

Indexes

Isolation Levels

By default, RonDB supports read committed. It can however also support the repeatable read mode by using SELECT ... LOCK IN SHARED MODE.

Joins

Can handle all joins handled by MySQL. Some queries can be pushed down to the RonDB data nodes which usually gives a 20x speedup of the query.

Query Interface

Storage Architecture

Durability is attained through a REDO log, no UNDO logging is required since transactional state is kept in memory and persistent state is only changed at COMMIT both for in-memory and disk columns. The REDO log is logical, so only records the operations and its new values.

Storage Format

Storage Model

System Architecture