Tom Lane wrote:
One additional point: this means that one transaction in every 32K writing transactions *does* have to do extra work when it assigns itself an XID, namely create and zero out the next page of pg_clog. And that doesn't just slow down the transaction in question, but the next few guys that would like an XID but arrive on the scene while the zeroing-out is still in progress. This probably contributes to the behavior that Simon and Josh regularly complain about, that our transaction execution time is subject to unpredictable spikes. I'm not sure how to get rid of it though.
A thread maintaining a pool of assigned and cleared pg_clog pages, ahead of the immediate need? Possibly another job for an existing daemon thread. - Jeremy
