Обсуждение: WAL-files restore and nextval('PK')

Andreas Gaab <A.Gaab@scanlab.de> writes:
> after an WAL-restore of our Postgres DB, we observe seemingly wrong values of our sequences.

> We have two postgres server (8.4) with pgpool in replication mode.

> Recently we tested our restore procedure and played our WAL-files into the second server after an old file-system
backupwas restored.
 
> Accidently, we aborted the starting server and had to restart it and therefore started WAL-replay again.

> Now we observe, that the newly restored server has higher values in his sequences as the other server.

It's normal for sequence counters to be a few counts higher after a
crash-and-restart than they would have been if no crash had occurred.
This is an intentional design tradeoff to minimize the WAL overhead
associated with assigning a sequence value.  If you find it intolerable
for what you're doing, I believe you can prevent it by adjusting the
sequence parameters to prevent any "caching" of values.
        regards, tom lane

Andreas Gaab wrote:
> Dear all,
> 
> after an WAL-restore of our Postgres DB, we observe seemingly wrong values of our sequences.

This is normal.  The reason is that sequences are WAL-logged in chunks,
not single values (otherwise they'd be too slow).  So after recovery,
the values jump to the logged values, which are necessarily higher than
the values they last delivered before the crash.

-- 
Alvaro Herrera                                http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.