Oh I see! I was conflating the data I see coming out of a replication slot with the internal organization of the WAL. I think the more specific question I am trying to answer is, as a consumer of a replication slot, how do I reason about what replication records will be made unavailable when I confirm an LSN? Here I am worried about situations where the replication connection is interrupted or the program processing the records crashes, and we need to replay records that may have been previously sent but were not fully processed.
For example, are the records sent by a replication slot always sent in the same order such that if I advance the confirmed_flush_lsn of a slot to the LSN of record "A", I will know that any records that had been streamed after record "A" will be replayable?
On Tue, Mar 25, 2025 at 12:14 PM Christophe Pettus <xof@thebuild.com> wrote:
Missed this question!
> On Mar 25, 2025, at 09:56, Phillip Diffley <phillip6402@gmail.com> wrote: > But when processing data from a replication slot, we confirm rows that have been processed and can be deleted from the WAL based on the LSN (eg. with pg_replication_slot_advance). How does postgres identify what parts of the WAL can be freed?
Basically, if no part of the system "needs" a particular LSN position, the segments that include that LSN position and earlier can be free.
The various things that can "need" a particular LSN point are:
1. Replication slots, if the other side has not confirmed that it has received it (under whatever synchronous commit rules that slot is operating under). 2. The wal_keep_size setting. 3. The max_wal_size setting. 4. The archive_command, if a WAL segment hasn't been successfully archived yet.
One thing to remember is that the WAL does *not* contain contiguous blocks of operations for a single transaction. The operations are written to the WAL by every session as they do operations, so the WAL is a jumble of different transactions. One of the jobs of the logical replication framework is to sort that out so it can present only the operations that belong to committed transactions to the output plugin. (This is why there's an internal structure called the "reorder buffer": it reorders WAL operations into transaction blocks.)
