I've installed postgres master and slave nodes by this manual:

At the master (

sudo -u postgres initdb -D /data/postgres/main

Change postgresql.conf:

listen_addresses = '*'
wal_level = hot_standby
max_wal_senders = 2
logging_collector = on
max_wal_senders = 5

Add record for replication to pg_hba.conf:

host replication replica md5

Then start postgres

service postgresql start

Create replication user:


And add replication slot:

SELECT pg_create_physical_replication_slot('standby_slot');

At the slave (

pg_basebackup -h -U replica -D /db/postgres/main -X s -P

Change state to standby in postgresql.conf:

hot_standby = on

Add recovery.conf:

standby_mode = 'on'
primary_conninfo = 'host= port=5432 user=replica password=123456'
primary_slot_name = 'standby_slot'
trigger_file = 'trigger'

Start standby:

service postgresql start

So, replication is working fine. If I create database, it will appear both at the master and the slave.

If slave is going down and then starting, all changes come there. It all fine too.

But I have problem if master is down. When it stops, I create trigger file at the slave. Then I create another database at the slave, which now is primary. After that I start master, trigger file remove automatically, recovery.conf rename to recovery.done automatically too. Ok. But how to synchronize changes from slave (new primary) to master (old primary)?

I found that I've got to do these four steps:

At the master:

Stop postgres:

service postgresql stop

At the slave:

Switch backup mode on:

SELECT pg_backup_start('backup', true);

Then rsync new data from slave to master.

And then switch backup mode off:

SELECT pg_backup_stop();

After last step I get message

NOTICE: WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup

And then if I start master it doesn't start with log message:

FATAL: could not locate required checkpoint record

I don't understand, how to get new data at the master and make it primary again.

So at the future I plan to install pgpool2 for balancing and failover, but test installing did not work because I don't know, how this mechanism should work.

1 Answer 1


You cannot just jump back to the old master, as it is said here:

Once failover to the standby occurs, there is only a single server in operation. This is known as a degenerate state. The former standby is now the primary, but the former primary is down and might stay down. To return to normal operation, a standby server must be recreated, either on the former primary system when it comes up, or on a third, possibly new, system.

For this, you need a recovery.conf on the old master. This should be nearly the same as the present one, only the connection has to point elsewhere.

  • So, I tried to make my old primary as standby. I did '/usr/local/pgsql/bin/pg_basebackup -h -U replica -D /db/postgres/main -X s -P' to get the base contents and config files. Then I renamed recovery.done to recovery.conf, and I changed host-address at line primary_conninfo in the recovery.conf. Also I changed IP-address in the pg_hba.conf. After I started old master I still have no replication relation between two servers. If I drop some database on the new primary, I don't see changes on the old one May 29, 2015 at 8:32
  • @user2627744 Do any errors pop up in the logs? May 29, 2015 at 8:47
  • yes, it told me about absent replication slot. I added it by 'SELECT pg_create_physical_replication_slot('standby_slot')' at new primary and replication now work. So, as I understood, now I can make old master primary again by repeating my actions: stop new primary, touch trigger at old primary, pg_basebackup from old primary to new primary, edit config files, add replication slot and start new primary as standby again. I hope pgpool can do it without errors.. May 29, 2015 at 9:21
  • Great! What I was not sure about was if the replication slot itself is replicated - apparently, you have to create it after failover. May 29, 2015 at 9:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.