On Mon, Jan 27, 2014 at 9:38 AM, Vadim Spivak <[email protected]> wrote: > On Sun, Jan 26, 2014 at 1:16 AM, Daniel Farina <[email protected]> wrote: >> >> On Sun, Jan 26, 2014 at 12:33 AM, <[email protected]> wrote: >> > On Saturday, November 2, 2013 5:31:48 PM UTC-7, Nick Auger wrote: >> >> Cascading replication was added to PG in 9.2 Is it possible to use WAL-E >> >> to log ship from a hot-standby? If so-- how would the configuration look? >> >> >> >> >> >> >> >> [master] -> [standby] -> [wal-e & s3] >> >> >> >> >> >> >> >> >> >> Thanks! >> >> >> >> >> >> >> >> -Nick >> > >> > I was wondering the same thing. Any guidance on WAL-E and cascading >> > replication? >> >> Cascaded systems can't take backups nor upload WAL, but since they are >> replicas of the primary one can transitively close the cascading and >> read directly from the primary's archives. >> >> Given your other question on the other thread, I'm guessing you are >> running into the thorny problem of double-promotions and remastering, >> is that right? > > We're just doing the initial planning at this point, but want to make > sure we do it right :)
Yeah. Read the primary's archives, regardless of the depth of cascading. The safest way to perform promotion is to have promotion of a standby *never* re-use a prefix of the old primary: have it archive to a different prefix. If you have cascaded standbys, you will want to point their restore-commands at the new prefix before the promotion. I don't think many people (and certainly not Heroku) have a lot of experience using WAL-E with such topologies, although some of us, self included, have thought about how to do so quite a bit. Should you invest in this, I would value your experiences here most highly, if you can share them. -- You received this message because you are subscribed to the Google Groups "wal-e" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
