Replicação do PostgreSQL: ID da linha do tempo fora de sequência, quando o antigo mestre se tornou escravo

Configurei a replicação master slave no PostgreSQL, seguindo este tutorial:
http://wiki.postgresql.org/wiki/Binary_Replication_Tutorial#Binary_Replication_in_6_Steps

Funcionou bem. Agora promovi o escravo e configurei o mestre original para ser um escravo do novo mestre, novamente seguindo o tutorial no link acima.

No entanto, a replicação não funciona mais. O novo escravo (ou seja, o antigo mestre) registra erros como:

FATAL:  terminating walreceiver process due to administrator command
LOG:  out-of-sequence timeline ID 1 (after 2) in log file 0, segment 1, offset 0

ou como:

LOG:  unexpected timeline ID 1 in log file 0, segment 1, offset 0

O que estou fazendo errado?

Detalhes a seguir

Quando promovi o ex-escravo, parece que começou uma nova linha do tempo, nº 2:

LOG:  received promote request
FATAL:  terminating walreceiver process due to administrator command
LOG:  record with zero length at 0/18B5148
LOG:  redo done at 0/18B50F0
LOG:  last completed transaction was at log time 2011-12-05 08:39:43.872041+00
LOG:  selected new timeline ID: 2
LOG:  archive recovery complete
LOG:  database system is ready to accept connections

Existem 2 linhas de tempo no pg_xlogdiretório do novo mestre (ou seja, o antigo escravo):

new-master$ tree pg_xlog
    pg_xlog
    ├── 000000010000000000000001
    ├── 000000020000000000000001
    ├── 00000002.history
    └── archive_status
        └── 00000002.history.ready

Desligo o novo master e o novo slave. No novo escravo, eu faço isso:

new-slave$ cd /var/lib/pgsql/9.1/data/
new-slave$ rm *
new-slave$ mkdir pg_xlog ; chmod 700 pg_xlog
new-slave$ rsync -a  --exclude pg_xlog --exclude postgresql.conf --exclude recovery.conf --exclude recovery.done --exclude postmaster.pid --exclude 'server.*'  \
     dw0azewdbpv11danny:/var/lib/pgsql/9.1/data/*  .
new-slave$ ... restore postgresql.conf etcetera from backup,
             ... change settings following instructions in tutorial

Eu começo o novo mestre e o novo escravo. Mas o escravo diz FATAL: encerrando o processo walreceiver :

LOG:  database system was interrupted; last known up at 2011-12-06 21:49:49 UTC
LOG:  creating missing WAL directory "pg_xlog/archive_status"
LOG:  entering standby mode
LOG:  streaming replication successfully connected to primary
LOG:  unexpected timeline ID 1 in log file 0, segment 1, offset 0
FATAL:  terminating walreceiver process due to administrator command
LOG:  unexpected timeline ID 1 in log file 0, segment 1, offset 0
LOG:  unexpected timeline ID 1 in log file 0, segment 1, offset 0
...

O pg_xlog no novo escravo:

new-slave$ tree pg_xlog/
pg_xlog/
├── 000000020000000000000001
└── archive_status

Eu deveria ter rsync pg_xlogtambém? Eu fiz isso: desligar mestre e escravo, rsync pg_xlog, iniciar escravo, iniciar mestre. Então a replicação parece funcionar: o novo escravo diz:

LOG:  database system was shut down at 2011-12-06 22:10:14 UTC
LOG:  entering standby mode
LOG:  consistent recovery state reached at 0/19A4420
LOG:  database system is ready to accept read only connections
LOG:  record with zero length at 0/19A4420
FATAL:  could not connect to the primary server: could not connect to server: Connection refused
                Is the server running on host "dw0azewdbpv11danny" (46.137.XX.YY) and accepting
                TCP/IP connections on port 5432?

LOG:  streaming replication successfully connected to primary

Mas quando faço qualquer coisa no novo mestre:

insert into moo (mää) values (23);

Em seguida, a replicação falha. O novo escravo diz:

LOG:  could not receive data from client: Connection reset by peer
LOG:  out-of-sequence timeline ID 1 (after 2) in log file 0, segment 1, offset 0
FATAL:  terminating walreceiver process due to administrator command
LOG:  out-of-sequence timeline ID 1 (after 2) in log file 0, segment 1, offset 0
LOG:  out-of-sequence timeline ID 1 (after 2) in log file 0, segment 1, offset 0

O mestre diz:

LOG:  database system is ready to accept connections
LOG:  autovacuum launcher started
LOG:  could not receive data from client: Connection reset by peer

Mas eu não reiniciei nenhuma conexão!

Se eu reiniciar o escravo, ele falha instantaneamente: FATAL: encerrando o processo do walreceiver :

LOG:  database system was shut down in recovery at 2011-12-06 22:24:37 UTC
LOG:  entering standby mode
LOG:  consistent recovery state reached at 0/19A4420
LOG:  database system is ready to accept read only connections
LOG:  redo starts at 0/19A4420
LOG:  record with zero length at 0/19A4510
LOG:  streaming replication successfully connected to primary
LOG:  out-of-sequence timeline ID 1 (after 2) in log file 0, segment 1, offset 0
FATAL:  terminating walreceiver process due to administrator command

Arquivos de configuração:

---- new master & slave: ----
wal_level = hot_standby
#archive_mode = off   (default)

---- new master: ----
max_wal_senders = 5
wal_keep_segments = 100

---- new slave: ----
hot_standby = on

---- recovery.conf, on the new slave: ----
standby_mode = 'on'
primary_conninfo = 'host=dw0azewdbpv11danny user=... password=...'

O que estou fazendo errado? Como posso transformar o antigo mestre em escravo?

(É melhor se eu não deletar todos os arquivos no novo escravo antes de mim rsync?
Ou devo executar initdbno novo escravo antes de mim rsync?
Você sabe se existe algum tutorial sobre como converter um mestre em um escravo? )

Atenciosamente, KajMagnus