max_slot_wal_keep_size

• 19
• 18
• 17
• 16
• 15
• 14
• 13

如果 max_slot_wal_keep_size 设置为正值，并且复制槽变为不活跃，pg_replication_slots 中的 safe_wal_size 列将包含在槽无效（即下游连接无法再使用该槽检索所需的 WAL）之前可以写入 WAL 的字节数。

postgres=# SELECT slot_name, slot_type, active, restart_lsn, wal_status, safe_wal_size
             FROM pg_replication_slots;
 slot_name | slot_type | active | restart_lsn | wal_status | safe_wal_size 
-----------+-----------+--------+-------------+------------+---------------
 node2     | physical  | f      | 0/301B830   | reserved   |      15057816
(1 row)

如果槽的 restart_lsn 与当前 LSN 之间的差值超过了 max_slot_wal_keep_size 的设置值，复制槽将被失效：wal_status 将被设置为 unreserved，并且（在 PostgreSQL 14 中）safe_wal_size 将显示一个负值。

postgres=# SELECT slot_name, slot_type, active, restart_lsn, wal_status, safe_wal_size
             FROM pg_replication_slots;
 slot_name | slot_type | active | restart_lsn | wal_status | safe_wal_size 
-----------+-----------+--------+-------------+------------+---------------
 node2     | physical  | f      | 0/301B830   | unreserved |     -80436096
(1 row)

在下一次 检查点时，会发出类似如下的日志行：

[2021-08-15 23:41:21 UTC]    LOG:  00000: checkpoint starting: immediate force wait
[2021-08-15 23:41:21 UTC]    LOG:  00000: invalidating slot "node2" because its restart_lsn 0/301B830 exceeds max_slot_wal_keep_size

之后 wal_status 将被设置为 lost，safe_wal_size 将被设置为 NULL。

postgres=# SELECT slot_name, slot_type, active, restart_lsn, wal_status, safe_wal_size
             FROM pg_replication_slots;
 slot_name | slot_type | active | restart_lsn | wal_status | safe_wal_size 
-----------+-----------+--------+-------------+------------+---------------
 node2     | physical  | f      |             | lost       |              
(1 row)

• PostgreSQL 文档： max_slot_wal_keep_size

• PostgreSQL 13：不要让槽拖垮你的主库 - 2ndQuadrant 2020 年 7 月博客文章