Il giorno mar 16 feb 2021 alle ore 10:51 Jehan-Guillaume de Rorthais <jgdr@xxxxxxxxxx> ha scritto:
On Mon, 15 Feb 2021 18:55:14 +0100
Paolo Saudin <paolosaudin@xxxxxxxxx> wrote:
> Hi all,
> I have two servers, a primary and a secondary one with a streaming replica
> setup.
> Today I noticed that some sequences are not lined-up, the replica ones are
> well ahead, while the records number is the same. How is it possible?
This is because sequences are not WAL logged on every nextval() call, but by
batch of sequence cache+32 values. As standbys are fed with WAL stream from
primary, their sequences appears jump, then stall, then jump, etc.
See:
https://git.postgresql.org/cgit/postgresql.git/tree/src/backend/commands/sequence.c?id=f672df5fdd22dac14c98d0a0bf5bbaa6ab17f8a5#n52
https://git.postgresql.org/cgit/postgresql.git/tree/src/backend/commands/sequence.c?id=f672df5fdd22dac14c98d0a0bf5bbaa6ab17f8a5#n661
I'm not sure if this is documented somewhere... ?
Thank you very much!
So in case the primary server crashes, and the backup one gets live, it will eventually have different sequences. I suppose there is no way to avoid this, isnt'it?
Paolo
