What's the best practice when using `pool` in a package?

For instance, if I need to implement a connection pool within a package, how should I proceed?

Previously, I created pool::dbPool() in the .onLoad() function and closed it with reg.finalizer(), which worked well. However, adding more data sources would clutter the .onLoad() function. Also, how can I establish a connection to the database only when it's needed for the first time?

Furthermore, if I invoke the following function multiple times, will it create multiple connections in a single pool, or will it create multiple pools?

con <- function() {
  pool::dbPool(odbc::odbc(), driver = "Oracle Instant Client", dbq = "xxx", uid = "xxx", pwd = "xxx")
}

What would be the best approach in this scenario, or is there an example package that addresses this?

Review the package's existing pool::dbPool(), .onLoad(), and reg.finalizer() usage, then investigate how repeated calls affect pools and connections. A useful outcome would be a decided, documented package pattern for lazy initialization and cleanup; no files or tests are named.