Error promoting slave on cascading replication using replication slots - Mailing list pgsql-general

Hi,

I'm configuring a cascading replication environment, with replication
slots, but I'm having a problem when the master goes down and I promote
a slave. All servers start from a cluster created from scratch, with
default config options. The process that I'm using to set up the
cascading replication it is:

1 - On master:
wal_level = hot_standby
max_wal_senders = 3
max_wal_replication_slots = 3
hot_standby = on

2 - On slave1:
Stop Server
Apply the same configuration from above
Erase the old cluster
Run pg_basebackup -v -P -R -X stream -c fast -h IP -U postgres -D PGDATA

3 - On master:
pg_create_physical_replication_slot('NAME')

4 - On slave1:
Add the primary_slot_name to recovery.conf
Start cluster

Everything run smoothly, according to with "SELECT * FROM
pg_stat_replication" and "SELECT * FROM pg_replication_slots". The steps
2, 3 and 4 are repeated on slave2 wich points to slave1. The problem
happens when I stop the master, and run a

pg_ctl -D /var/lib/postgresql/9.4/main promote

on slave1. At this point, slave2 throws the following log, and stops
receiving WAL through the replication slot:

2015-12-17 11:23:06 BRST [944-2] LOG:  replication terminated by primary
server
2015-12-17 11:23:06 BRST [944-3] DETAIL:  End of WAL reached on timeline
1 at 0/30001A0.
2015-12-17 11:23:06 BRST [944-4] LOG:  fetching timeline history file
for timeline 2 from primary server
2015-12-17 11:23:06 BRST [937-7] LOG:  record with zero length at 0/30001A0
2015-12-17 11:23:06 BRST [944-5] LOG:  restarted WAL streaming at
0/3000000 on timeline 1
2015-12-17 11:23:06 BRST [944-6] LOG:  replication terminated by primary
server
2015-12-17 11:23:06 BRST [944-7] DETAIL:  End of WAL reached on timeline
1 at 0/30001A0.
2015-12-17 11:23:11 BRST [944-8] LOG:  restarted WAL streaming at
0/3000000 on timeline 1
2015-12-17 11:23:11 BRST [944-9] LOG:  replication terminated by primary
server

I found a instruction to add the following line to recovery.conf:
recovery_target_timeline = 'latest'

When this line is added, slave2 keeps its replication with slave 1:
2015-12-17 13:37:54 BRST [868-2] LOG:  replication terminated by primary
server
2015-12-17 13:37:54 BRST [868-3] DETAIL:  End of WAL reached on timeline
1 at 0/3001358.
2015-12-17 13:37:54 BRST [868-4] LOG:  fetching timeline history file
for timeline 2 from primary server
2015-12-17 13:37:54 BRST [863-7] LOG:  new target timeline is 2
2015-12-17 13:37:54 BRST [863-8] LOG:  record with zero length at 0/3001358
2015-12-17 13:37:54 BRST [868-5] LOG:  restarted WAL streaming at
0/3000000 on timeline 2

My question is: is this the right procedure, or am I missing something?

Best regards,

--
Álvaro Nunes Melo    Atua Sistemas de Informação
alvaro@atua.com.br   http://www.atua.com.br
(54) 9976-0106       (54) 3045-8100