A connection pool reuses server connections across client requests. Session pooling keeps a server connection associated with a client session. Transaction pooling can return a connection to the pool at transaction end, so a later transaction may use a different server session.
Do not assume session settings, temporary tables or session locks follow a client across transaction-pooled requests. Put tenant context in the transaction with SET LOCAL and derive it from trusted identity. Prepared-statement support depends on the pooler's configuration and protocol handling; verify the deployed behavior instead of repeating a blanket compatibility claim.
A larger pool isn't automatically faster. Too many active database workers can increase contention while idle transactions hold locks and old snapshots. Inspect wait events and transaction age before raising limits. The supplied service_sessions table is a fictional diagnostic extract, not the browser engine's own activity view.
An idle transaction can retain locks and snapshots; a plain idle session is a different state.