Larry Prikockis wrote:
> ok... so maybe a better question would be: other than cases of power
> failure or something else that caused the server to shut down
> unexpectedly and uncleanly, what is the actual risk of data loss? In
> my case, the possibility of losing a recent transaction or two in
> the hopefully uncommon case of some catastrophic failure might very
> well be acceptable in exchange for increased write performance.
>
> random/unpredictable risk of data loss or corruption for other
> reasons is probably NOT acceptable to me though, regardless of the
> performance gain.
[Barring software bugs,j asynchronous commit does not cause data
corruption or unpredictable data loss. You can lose the last batch of
transactions on a crash, but even if you do, the data written by
previous transactions is valid.
Worst case of # of transactions lost is determined by 3x
wal_writer_delay.
--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services