Aurora MySQL 3.14 Is Generally Available: Hybrid Post-Quantum TLS Key Exchange Needs Client Support Too, DELETE Gains RETURNING and Makes It a Reserved Word, and Row-Size Behavior Stays Older Than Community 8.0.46

Published · AI Daily — AI-assisted deep research, methodology & disclosure

On November 2, 2026, AWS announced Aurora MySQL 3.14 (compatible with MySQL 8.0.46) as generally available, headlined by post-quantum TLS (PQ-TLS) key exchange. The release notes add the hybrid groups X25519MLKEM768 and SecP256r1MLKEM768 for TLS 1.3, negotiated only if the client supports them and otherwise falling back to classical groups, checkable with the Aurora_ssl_named_group status variable. Also listed are faster minor upgrades from 3.08, RETURNING for single-table DELETE (which makes RETURNING reserved, so identifiers need backticks), a FTSCacheMemoryUsed metric, and retained pre-8.0.46 row-size behavior that differs from community MySQL. Audit logs now record the invoking client, not the definer. The notes are dated October 26 and give no upgrade-speed or TLS-overhead figures.

On November 2, 2026, AWS announced that Amazon Aurora MySQL-Compatible Edition 3, the MySQL 8.0 line, now supports MySQL 8.0.46 as Aurora MySQL 3.14. The announcement says the release carries community MySQL fixes and Aurora-specific improvements, and that its headline addition is support for post-quantum TLS (PQ-TLS) key exchange. It is available in all AWS Regions where Aurora MySQL is available. One date needs care: the Aurora release notes list version 3.14.0 under the date 2026-10-26, so the engine was documented about a week before the "starting today" announcement.

The announcement is short, and the release notes explain what the post-quantum feature does. Aurora MySQL adds two hybrid key exchange groups, X25519MLKEM768 and SecP256r1MLKEM768, alongside the classical elliptic-curve groups for TLS 1.3 connections. Clients that support post-quantum key exchange automatically negotiate a quantum-resistant shared secret, and clients that do not keep using classical groups. So turning it on is not a setting you flip on the server. It happens only when the client library also supports these groups and uses TLS 1.3. To check what a session negotiated, run SHOW STATUS LIKE 'Aurora_ssl_named_group'.

The release notes list other changes that may matter more to operators. Minor version upgrades to 3.14 from version 3.08 or higher now complete faster, so the writer instance returns to service sooner, with no configuration change or opt-in. Single-table DELETE statements gain a RETURNING clause, which returns values from the deleted rows without a separate SELECT. Because RETURNING is now a reserved keyword, any identifier named returning must be quoted with backticks, and an unquoted one will fail to parse after the upgrade. A new CloudWatch metric, FTSCacheMemoryUsed, shows memory used by the InnoDB full-text search cache across all tables on an instance.

One compatibility choice is worth reading closely. Community MySQL 8.0.46 changed how it handles tables with many columns of certain kinds, so creating such a table can fail with ERROR 1118, "Row size too large," even though the same statement worked on earlier 8.0 releases. The release notes say Aurora MySQL 3.14.0 retains the older behavior, so those statements keep working as in 3.13.0 and earlier. That means the behavior of Aurora 3.14 differs from community 8.0.46 here, which matters if you test or migrate between the two.

The notes also list security fixes. The Advanced Audit log previously recorded the definer's user and host, not the calling client's, for statements run inside a SQL SECURITY DEFINER routine, and now records the invoking client. Prepared statements no longer create duplicate audit entries. A fix covers SET ROLE NONE not clearing a previous role's privileges in sessions that use write forwarding, which could allow operations that should have been denied. The release also lists fixes for two high-severity CVEs and a longer list of medium-severity ones. Teams that parse audit logs for compliance should expect the user and host values to change after the upgrade.

On rollout, AWS says you can upgrade during maintenance windows with automatic minor version upgrades, and that the AWS Organizations Upgrade Rollout Policy can phase upgrades across clusters. Upgrades can be in place or through snapshot restore. For a team, the practical checklist is short: check that client libraries negotiate the hybrid groups if PQ-TLS is the goal, search code for unquoted returning identifiers, review audit-log consumers, and test in a lower environment first. The announcement and notes give no figures for the speed of the faster upgrade or for any TLS overhead.

Sources

FAQ

Does upgrading to Aurora MySQL 3.14 make my connections quantum-resistant automatically?

Not necessarily. The release notes add the hybrid groups X25519MLKEM768 and SecP256r1MLKEM768 for TLS 1.3, but they are negotiated only when the client also supports post-quantum key exchange; otherwise the session uses classical groups. Check the Aurora_ssl_named_group status variable to see what a session negotiated.

What compatibility problem is easiest to hit before upgrading?

RETURNING is now a reserved keyword because single-table DELETE gained a RETURNING clause. Any identifier named returning must be quoted with backticks or it will fail to parse after the upgrade. Also, the audit log now records the invoking client in definer routines, so parsers that expect the old user or host values need review.

Does Aurora 3.14 handle row size like community MySQL 8.0.46?

No. Community 8.0.46 can fail some many-column table statements with ERROR 1118, Row size too large, but the release notes say Aurora MySQL 3.14.0 retains the pre-8.0.46 behavior, so those statements still succeed as in 3.13.0 and earlier. Watch for this difference when testing or migrating between the two.