What’s New in MySQL 8.0 InnoDB? Key Features and Performance Boosts

This article is based on a talk by Lai Zheng at the 728 Database Salon of the Efficient Operations Community, focusing on MySQL 8.0 InnoDB new features.

In the presentation, Lai Zheng divided the content into three parts: functional improvements, performance improvements, and future outlook for InnoDB in MySQL 8.0.

Functional Improvements

The most significant change is the redesign of the data dictionary. Previously MySQL used two dictionaries: one in the MySQL layer stored in .frm files and another in InnoDB. This caused non‑atomic DDL, inconsistencies, deadlocks, and potential corruption. MySQL 8.0 consolidates the dictionary into a single InnoDB‑based store, enabling atomic DDL, transactional DDL, and eliminating .frm files.

Benefits include a single source of truth for schema, support for atomic operations, ACID‑compliant DDL, and unified lock management (MDL) for the dictionary.

Another notable feature is the "fast add column" capability. Previously adding a column required rebuilding the table and copying data. In MySQL 8.0, the column can be added instantly without table rebuild, thanks to changes in page structures.

Encryption enhancements extend beyond table‑level encryption introduced in 5.7. InnoDB now encrypts redo‑log and undo‑log data, as well as shared tablespaces, improving data‑at‑rest security.

The Memcached plugin has been enhanced to support multiple‑get operations and range queries, allowing direct API access to InnoDB data without going through the SQL layer.

Auto‑increment handling has been improved by storing the current value in the data dictionary, eliminating the need to scan tables on restart and preventing duplicate values.

Additional changes include new information_schema tables for better visibility into schema state.

Performance Improvements

The biggest performance gain comes from parallel redo‑log writes. The previous single‑mutex log buffer became a bottleneck under high concurrency. MySQL 8.0 replaces it with a lock‑free algorithm that pre‑allocates log buffers based on LSN, achieving more than 50% throughput improvement in high‑concurrency workloads.

Other performance enhancements include cost‑based optimizer statistics for in‑memory pages, buffer‑pool partitioning, and table‑id based purge grouping, which reduces contention among purge threads.

Deletion now marks secondary index entries for later removal by purge threads, speeding up delete operations.

Deadlock detection can now be dynamically tuned, and various internal optimizations further improve scalability.

Future Outlook

Planned future work includes more advanced tablespace management (versioned tablespaces, self‑describing tablespaces using SDI with JSON metadata), better JSON and BLOB storage handling, further MVCC and transaction‑lifecycle optimizations, and additional file‑system level improvements.

For detailed performance test results, refer to the dedicated testing site linked by the presenter.