I have a setup kind of like yours, one primary and two stanby, but with streaming replication.
Here is what i use: -> on application (Django webapp) servers: |---> a local HAProxy as a frontend for database servers, doing the fencing through a check on a custom daemon running on database servers -> on PostgreSQL servers: |---> repmgr for the automatic failover |---> the custom daemon which expose the state of the node (by trying to retreive its state according to each node's repmgr status table and doing a kind of vote, for exemple "if according to me i'm the primary but not according to the two other nodes, i'm fenced") Regards, Nicolas KAROLAK | SysAdmin 6-8, rue André Voguet - 94200 Ivry-sur-Seine +33 1 77 56 77 96 | www.ubicast.eu | www.ubicast.tv Le mer. 5 sept. 2018 à 18:39, Dmitri Maziuk <dmaz...@bmrb.wisc.edu> a écrit : > > On Wed, 5 Sep 2018 17:45:05 +0200 > Thomas Poty <thomas.p...@gmail.com> wrote: > > > We have one master and 2 slaves so 3 nodes > > So what is the "HA" that you're trying to "increase"? Are you adding a 3rd > slave? A 2nd master? A hot standby? All of the above? > > -- > Dmitri Maziuk <dmaz...@bmrb.wisc.edu> >
