Shared database connection pools

[slashdotdash]

By default each event store will start its own Postgrex database connection pool. The size of the pool is configured with the pool_size config option.

When you have multiple event stores running you will also end up with multiple connection pools. If they are all connecting to the same physical Postgres database then it can be useful to share a single pool amongst all event stores. This pull request adds the shared_connection_pool config option to specify a name for the shared connection pool. Then you can configure the event stores you'd like to share the pool by using the same name.

This can be done in config:

# config/config.exs
config :my_app, MyApp.EventStore1, shared_connection_pool: :shared_pool
config :my_app, MyApp.EventStore2, shared_connection_pool: :shared_pool
config :my_app, MyApp.EventStore3, shared_connection_pool: :shared_pool

Or when starting the event stores, such as via a Supervisor:

Supervisor.start_link(
  [
    {MyApp.EventStore, name: :eventstore1, shared_connection_pool: :shared_pool},
    {MyApp.EventStore, name: :eventstore2, shared_connection_pool: :shared_pool},
    {MyApp.EventStore, name: :eventstore3, shared_connection_pool: :shared_pool}
  ],
  opts
)

Not specifying the shared_connection_pool config option indicates the event store will start its own connection pool. This is the existing behaviour. Not using this new shared pool feature should have no affect.

Closes #198.

[slashdotdash]

Database connections used by an event store instance are tied to whatever schema has been configured for the event store. This is because the Postgres search_path is set for the session once connected to the database by using the :after_connect config option passed to Postgrex. The purpose of doing this was to not need to prefix the database queries with the schema name as it would require looking up the schema from the event store config each time a database request is made via the EventStore public API.

This limitation means that sharing a connection pool won't work as expected if the event stores have been configured to use different schemas which is highly likely. Thus sharing the connection pool requires further work to include the event store schema when making all database requests.

[apolkk]

Database connections used by an event store instance are tied to whatever schema has been configured for the event store. This is because the Postgres search_path is set for the session once connected to the database by using the :after_connect config option passed to Postgrex. The purpose of doing this was to not need to prefix the database queries with the schema name as it would require looking up the schema from the event store config each time a database request is made via the EventStore public API.

This limitation means that sharing a connection pool won't work as expected if the event stores have been configured to use different schemas which is highly likely. Thus sharing the connection pool requires further work to include the event store schema when making all database requests.

Thank you for making this work on #217 🎖️