Use PG17+ failover slots when creating replication slots

[altmannmarcelo]

Summary

PostgreSQL 17 introduced failover slots (CREATE_REPLICATION_SLOT ... (FAILOVER)), which allow logical replication slots to be synchronized to physical standbys. When a standby is promoted, the slot survives with its confirmed_flush_lsn intact, enabling subscribers to reconnect without a full resnapshot.

Readyset should use this option when creating replication slots on PG17+ servers. This is a small change that gives free failover resilience.

Current behavior

Readyset creates replication slots with:

CREATE_REPLICATION_SLOT <name> LOGICAL pgoutput EXPORT_SNAPSHOT

When the primary fails over to a standby, the slot is lost, and Readyset must do a full resnapshot against the new primary.

Proposed change

When server_version_num >= 170000, use the PG17 parenthesized option syntax with the FAILOVER flag:

CREATE_REPLICATION_SLOT <name> LOGICAL pgoutput (FAILOVER, SNAPSHOT 'export')

For older versions, keep the existing syntax unchanged.

Files to change

• replicators/src/postgres_connector/connector.rs — create_replication_slot(): accept version parameter, conditionally add FAILOVER
• replicators/src/noria_adapter.rs — pass version_num through to create_replication_slot()

Impact

• PG17+: Slot survives failover to a promoted standby → no resnapshot needed
• PG < 17: No change in behavior (existing full-resnapshot fallback still works)