Split Database Connection Pools

When the "database.enable-read-write-split" setting is enabled, the database connection pool is split into two: A read-only pool, and a read/write pool. The read-only pool is used for many of the administrator API calls that only fetch data. The read/write pool is used for all back-end message processing, as well as any operation that creates/modifies/deletes data.

When the connection pools are split, you can separately configure settings for the read-only pool, with the "database-readonly" options. By default none of the "database-readonly" options are set, meaning that the read-only pool defaults to the same configuration as the main read/write pool.

For example, if "database-readonly.max-connections" is not set, it defaults to the "database.max-connections" setting. So if you have your max connections set to 20 for the read/write pool, the read-only pool also has up to 20 connections, meaning the total number database connections is at most 40.

These settings also enable you to point the read-only connection pool to a completely different database instance. You may want to do this if you have a master DB and a horizontally scaling cluster of read-replica DBs. You can point the main read/write pool to the master DB, and point the read-only pool to the read-replica cluster instead. By doing this you can potentially reduce the traffic and strain on your master database.

When using read replicas however, the concept of "replica lag" should be taken into account. That refers to the amount of time a read replica DB is behind the master DB. If your replica lag is sufficiently large, all the selects done by the read-only pool may return out-of-date information, which can lead to unintended results in the Administrator.

The server keeps internal caches for channels, channel groups, code templates, and code template libraries. By default these caches use the read-only pool. But if replica lag is a concern, a separate option, "database.write-pool-cache", enables you to switch the caches over to using the read/write pool instead.