>note postgres' WAL archive is by block, not by transaction.
My understanding is that only the first time a block is updated after a
checkpoint,
is the entire block is written to the WAL logs. And for that
full_page_writes has to be set to ON.
The only other time PG writes entire block to the WAL is during the time of
backup regardless of full_page_writes setting.
AFAIK rest of the time, WAL takes only row changes. Otherwise PG will be
generating large number
of WAL logs.
I hope I am right :-)
> also note that postgres effectively does copy-on-write, since update's are
> treated
>as insert+delete, so the same blocks aren't written over and over nearly
>as much as they might be in the oracle storage model.
Good point.
--
View this message in context: http://www.postgresql-archive.org/Incremental-Level-1-backup-in-PG-tp5951072p5951343.html
Sent from the PostgreSQL - general mailing list archive at Nabble.com.