Recent analysis has shown that turning off full_page_writes is not safe
under *any* circumstances, even if you have infrastructure such as a
battery-backed disk cache that should prevent partial page writes.
We have seen two recent reports of full_page_writes = off causing
failure to restart after a database crash, even though no OS- or
hardware-level problem occurred.
Our plan is to disable the variable in 8.1.4 and future 8.1.* releases
(forcing the system to act as though it's always ON), but not remove
the variable as that would cause unexpected breakage to postgresql.conf
files. We will look into whether it's practical to re-enable it for
8.2; at present it appears that doing so will require a chunk of
additional code that would be imprudent to back-patch.
If you are running on 8.1.0 - 8.1.3, however, you need to make sure that
full_page_writes is ON.
For technical details see this thread:
regards, tom lane
pgsql-announce by date
|Next:||From: David Fetter||Date: 2006-04-10 06:01:26|
|Subject: == PostgreSQL Weekly News - April 09 2006 ==|
|Previous:||From: Greg Sabino Mullane||Date: 2006-04-06 03:05:02|
|Subject: DBD::Pg version 1.48 released|