Craig,
I was starting the first node then letting all of the other nodes join as quick as they could which clearly won't work. It also explains why it worked when I did it manually, I can only do it sequentially myself ;) I had suspected a race condition and it seems I was in the right area :)
Thank you for this, I will alter what I am doing to start sequentially.
On Tue, May 12, 2015 at 1:19 AM, Craig Ringer <craig@xxxxxxxxxxxxxxx> wrote:
On 12 May 2015 at 14:33, Wayne E. Seguin <wayneeseguin@xxxxxxxxx> wrote:7. on nodes 1-4 I am doing (adjusted for the nodes IP):SELECT bdr.bdr_group_join(local_node_name := 'pgbdr1',node_external_dsn := 'host=10.244.2.6 port=5432 user=postgres dbname=pgbdr',join_using_dsn := 'host=10.244.2.2 port=5432 user=postgres dbname=pgbdr');At a guess you're probably not waiting between joins to ensure that each new node has finished joining before starting another node join.
BDR really needs to be enhanced to either support parallel join of multiple nodes or identify and reject it.--
