v2026.1 STS Stable release with standard-term support v2025.2 LTS Stable release with long-term support v2025.1 STS Stable release with standard-term support v2024.2 LTS Stable release with long-term support Unsupported versions

Slot behavior by version EARLY ACCESS

Slot scope, retention, restart time, and expiry across releases

The following table summarizes how replication slot behavior has changed across releases and how it compares with PostgreSQL 15, which YugabyteDB logical replication is based on. For the behavior itself, see Key concepts.

Behavior YugabyteDB PostgreSQL 15
Tables retained for a slot All versions: All tables in the database, from slot creation. Tables created later are added automatically. WAL, shared by all tables and databases.
Streaming without a publication All versions: All tables, for test_decoding and wal2json. pgoutput and yboutput require a publication. Same.
Publication changes while streaming v2024.2 to v2025.2: Applied at the next periodic refresh, every 15 minutes by default.
v2026.1 and later: Applied at commit time with implicit publication.		 Applied at commit time.
Tables not of interest All versions: A table not polled within 4 hours of slot creation becomes unqualified. Its changes are no longer retained for the slot, even if it is added to the publication later. The cdc_skip_unqualified_tables_for_polling flag lets the slot skip such tables instead of failing.
v2024.2 to v2025.2: Flag available from v2025.2.7.0.
v2026.1 and later: Flag available from v2026.1.2.0.		 No equivalent. A slot retains WAL regardless of which tables are published.
When the restart time advances All versions: On acknowledgment, to the latest fully acknowledged commit. Every 60 seconds while idle, if everything sent has been acknowledged. On acknowledgment, restart_lsn moves to the oldest WAL still needed to decode unacknowledged transactions.
Retention while the client is disconnected All versions: WAL, intents, and, for tables with replica identity FULL or DEFAULT, row history, from the restart time, for up to 8 hours since the last poll (cdc_intent_retention_ms, cdc_wal_retention_time_secs). WAL from restart_lsn, without a time limit. max_slot_wal_keep_size (off by default) caps it by size.
Resuming after reconnection All versions: From the restart time. Transactions after it are re-sent in full with the same LSNs. From restart_lsn. Transactions after confirmed_flush_lsn are re-sent in full.
When a slot becomes unusable All versions: After 8 hours without a poll (expired), or when the publication contains an unqualified table (unless cdc_skip_unqualified_tables_for_polling is set). The slot must be dropped and recreated. When WAL beyond max_slot_wal_keep_size is removed. The slot must be dropped and recreated.
Slot state in pg_replication_slots All versions: active is true if the slot was consumed in the last 60 seconds. Includes yb_stream_id and yb_restart_commit_ht. From v2025.1, also yb_lsn_type and yb_restart_time, and wal_status is lost when no tablet of the slot was polled for 8 hours. active is true while a process holds the slot. wal_status is lost for an invalidated slot.