I realize that synchronous replication is in the works for 9.1, but
for 9.0, could you hack your way to synchronous replication by polling
the standby's pg_last_xlog_receive_location() and waiting until it's
>= pg_current_xlog_location() from your last transaction? Any major
pitfalls here (besides the need to balance polling frequency and
latency)? Any recommended polling periods?
Aside: would 9.1's synchronous replication be adding anything that
lowers the replication delay compared to asynchronous replication? Or
could the changes be summarized as "commits now wait for the standby"?
--
Yang Zhang
http://yz.mit.edu/