Hi.
We have a setup of 3 Postgres 9.2 nodes (1x master, 2x slaves) running on
EC2 instances in AWS. We want to upgrade to 9.6 so we can move to Amazon
RDS (it requires at least a 9.3.5 version
<https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/CHAP_PostgreSQL.html>
).
As far as I know, we have 3 options here:
1. From 9.2 on EC2 to RDS
1. pg_dump on 9.2
2. Restore it in RDS (might not work; haven't tested it)
2. Get Slony replication working
1. Setup slony in our 9.2 nodes.
2. Get a fresh Postgres installation running Postgres 9.6
3. Copy the database over (pg_basebackup)
4. Start the replication (master 9.2 --> slave 9.2 --> slave 9.6)
5. failover to 9.6
6. re-create slaves from 9.6
3. From 9.2 to 9.6 on EC2; then to RDS:
1. upgrade the master 9.2 to 9.6
2. re-create all slaves with 9.6
3. after a while (months/weeks), use DMS or pg_dump/pg_restore and go
for RDS
Because of downtime, *Step 2* seems more with what we wanna go forward.
Questions are:
- At the moment I am using native Postgres streaming replication with
two slaves. Do I need to install Slony on my master, and get rid of the
native tool? and then re-create the slaves with slony?
- Slony or Bucardo are trigger-based replication solutions. That means
DDL changes won't be propagated. This is not good as new tables can be
added at any time. Isn't there a workaround for this?
Thanks!
