Hi all,
i am not really sure that this is the right forum, but anyway. We are running a 200 GB Postgres 9.1 64bit database. We do lots of writing and even more reading. The main database in office A is replicated on-site via streaming as a pure failover and also offsite via streaming to another office B over an asynchronous 100Mbit-Down/10-Mbit Up-Internetconnection. The other office is doing all reads from the read-only-streamed-to server and writes to the main-production-server in office A. During regular operations the delay is usually sub-second, which is perfectly fine for us.
But we have been bitten multiple times by the 32bit XID wraparound and the necessary vacuums that create a high amount of data-turnover that needs to go over the thin upload. We need to keep a very high count of wal-archives ( 8000+ ) to make sure all files are transferred before reuse and the transfers can take multiple days until both machines are in sync again.
The streaming replication is a very very cool feature and we are very enthusiastic postgres-users. I believe this problem was not really that severe before streaming replication was introduced, but is now becoming quite a problem.
My question is: do other users experience the same problem and what is their preferred way of dealing with it ? Also, i am aware of a discussion a few years ago that dealt with the usage of 64bit XIDs, but led nowhere. Are there any more recent plans or discussions concerning 64bit XIDs ? ( I am not a postgres-developer and have no idea about the complexity or performance/space side-effects of such a change, i am purely argumenting from a users point of view).
Thanks,
Frank