Durable slot

[DaemonSnake]

adds support for non-temporary replication slots

• adds the optional config toggle durable_slot
• when enabled it will:
  • check if slot already exists
  • try to create it if it doesn't
  • starts it
• the startup will fail if:
  • the slot already exists and is still active
  • multiple processes/nodes try to:
  • create the same durable slot around the same time (only one will succeed)
  • start the durable slot at the same time

When the slot is restarted, it will start back at the last wal_end+1 that walex returned in the keep-alive reply to postgres. Transactions processed between the last keep-alive and the interruption of walex will be replayed.

If event replay is problematic, the end-user must implement a way to mark LSN/transactions as processed to avoid re-processing.

Also event loss is possible currently in this situation:

• Walex receives transaction
• start processing asynchronously
• receives keep-alive request
• reply with the received wal_end+1 (which is greater than current transaction being processed)
• walex crashes / get interrupted before finishing processing the transaction
• restarts at position wal_end+1 (event was lost)

we made another PR to address this and will submit it shortly

[cpursley]

This is an awesome and very welcome improvement @DaemonSnake!

Let me know if you have other PRs planned - I need to cut a new release soon.

[cpursley]

@DaemonSnake did you have any more changes you wanted to get in before I create a new release?

[DaemonSnake]

@cpursley oh yes, will submit it today. Sorry for the late reply. Thanks you so much for the quick review and merge by the way 😁