Smooth scale-out
This topic explains the basic principles of smooth scale-out in PolarDB-X 1.0.
If the underlying data storage for a logical database reaches a physical bottleneck, you can scale the storage horizontally. For example, if the remaining disk space is close to 30%, you can use the smooth scale-out feature in the console to resolve this.
Smooth scale-out is an online method for horizontal scaling. It smoothly migrates existing sharded databases to newly added RDS or PolarDB for MySQL instances. This increases the number of instances, expands the total data storage capacity, and reduces the processing load on each instance.
- Create a scale-out plan
Select new RDS or PolarDB for MySQL instances and the sharded databases to migrate to them. After you submit the task, the system automatically creates databases and accounts on the destination instances and starts a data migration and synchronization task.
- Full migration
The system selects a point in time and performs a full data replication and migration of all data that exists before that point.
- Incremental data synchronization
After the full migration is complete, incremental synchronization starts by applying change logs from a point in time before the full migration began. This keeps the data in the source and target sharded databases synchronized in real time.
- Data validation
When incremental synchronization becomes near real-time, the system automatically performs a full data validation. It also corrects any data inconsistencies caused by synchronization latency.
- Stop application writes and switch routing
After validation is complete and while incremental synchronization remains near real-time, you can perform the switch at a time that you choose. To ensure strict data consistency, stop the application. (You can also keep the application online, but this may cause issues where concurrent updates overwrite the same data.) The engine layer then switches the routing based on the database sharding rules to direct subsequent traffic to the new databases. The switch process completes in seconds.
The entire scale-out process has almost no impact on the normal operation of your applications. (However, some impact may occur if the RDS or PolarDB for MySQL instance types are too small or the instances are under heavy load.) If you keep the application online during the switch, perform the operation during off-peak database access hours to reduce the risk of concurrent updates overwriting the same data.