A proposed MySQL Group Replication enhancement could help maintain database availability when one of two regions fails or becomes disconnected, provided the surviving region contains the primary. Explore how primary-aware partition handling could reduce downtime and support more resilient single-primary deployments, and share your feedback with the MySQL community: https://github.com/mysql/mysql-community/issues/159
Explore a proposed enhancement to MySQL Group Replication designed to improve database availability in deployments spanning two geographic regions or data centers. In this MySQL Summit session, the presentation examines how MySQL Group Replication handles region failures, network partitions, and quorum decisions, and introduces a potential approach for keeping a single-primary cluster operational when exactly half its members become unavailable.
MySQL Group Replication uses majority-based membership decisions to maintain consistency and prevent conflicting transactions. In a three-region deployment, the remaining majority can continue operating when one region fails. However, in a two-region deployment with an equal distribution of members, losing connectivity to one region can leave neither partition with a majority, causing the cluster to stop accepting transactions.
The proposed enhancement would modify this behavior for an evenly divided group. When exactly half the members remain connected, the partition containing the existing primary would be permitted to continue operating. The partition without the primary would stop participating, preserving the intended protection against split-brain scenarios.
The session discusses how this approach could help organizations maintain database availability during regional infrastructure failures, network disruptions, and planned maintenance. It also examines situations where administrators can move the primary to a preferred region before an anticipated outage, potentially allowing the surviving region to continue processing transactions.
The presentation explains the proposal's limitations, including its restriction to single-primary mode. If the primary becomes unavailable along with the failed region, the remaining members would still be unable to continue operating. The proposed behavior does not extend to multi-primary configurations.
Additional discussion covers recovery after network connectivity is restored, replacement of unavailable members, manual recovery using Group Replication force-members functionality, and alternative approaches involving arbitrator nodes or uneven member distributions.
The session concludes by inviting community feedback on whether primary-aware partition handling would improve uptime, support two-region deployment requirements, or help reduce the infrastructure costs associated with maintaining a third region.
This presentation describes a proposed MySQL Group Replication enhancement rather than an existing feature. The potential availability improvements depend on the deployment configuration and failure conditions discussed.
Explore a proposed enhancement to MySQL Group Replication designed to improve database availability in deployments spanning two geographic regions or data centers. In this MySQL Summit session, the presentation examines how MySQL Group Replication handles region failures, network partitions, and quorum decisions, and introduces a potential approach for keeping a single-primary cluster operational when exactly half its members become unavailable.
MySQL Group Replication uses majority-based membership decisions to maintain consistency and prevent conflicting transactions. In a three-region deployment, the remaining majority can continue operating when one region fails. However, in a two-region deployment with an equal distribution of members, losing connectivity to one region can leave neither partition with a majority, causing the cluster to stop accepting transactions.
The proposed enhancement would modify this behavior for an evenly divided group. When exactly half the members remain connected, the partition containing the existing primary would be permitted to continue operating. The partition without the primary would stop participating, preserving the intended protection against split-brain scenarios.
The session discusses how this approach could help organizations maintain database availability during regional infrastructure failures, network disruptions, and planned maintenance. It also examines situations where administrators can move the primary to a preferred region before an anticipated outage, potentially allowing the surviving region to continue processing transactions.
The presentation explains the proposal's limitations, including its restriction to single-primary mode. If the primary becomes unavailable along with the failed region, the remaining members would still be unable to continue operating. The proposed behavior does not extend to multi-primary configurations.
Additional discussion covers recovery after network connectivity is restored, replacement of unavailable members, manual recovery using Group Replication force-members functionality, and alternative approaches involving arbitrator nodes or uneven member distributions.
The session concludes by inviting community feedback on whether primary-aware partition handling would improve uptime, support two-region deployment requirements, or help reduce the infrastructure costs associated with maintaining a third region.
This presentation describes a proposed MySQL Group Replication enhancement rather than an existing feature. The potential availability improvements depend on the deployment configuration and failure conditions discussed.