TDSQL supports real-time online scaling. There are two scaling methods: adding new shards and scaling up existing shards. The entire scaling process is completely imperceptible to the business with no business interruption caused. Only the involved shards will become read-only (or interrupted) for just seconds during scaling, while the entire cluster will not be affected.
TDSQL ensures the stability of automatic scaling with its proprietary automatic rebalancing technology.
Adding a new shard
- After you click scale in the console, the system will calculate the bottleneck of node A (multiple nodes may actually be affected) based on the load and capacity.
- According to the configuration of new node G, some data of node A will be migrated (from the slave) to node G.
- After the data is completely synced, nodes A and G will check the database (read-only status for one to tens of seconds), but the entire service will not be suspended.
- The scheduling system notifies the proxy to switch route.
Scaling up an existing shard
Scaling up an existing shard is actually replacing it with a larger physical shard.
This does not add any new shards and will not change the logic rules of sharding or number of shards.
- Assign a new physical shard (the "new shard") based on the required configuration.
- Sync the data and configuration of the physical shard to be upgraded (the "old shard") to the new shard.
- After the data sync is completed, switch the route in the Tencent Cloud gateway to the new shard for continued use.