Hello,
I had an issue with the behavior of pg_basebackup command. I was convinced previously that pg_basebackup command always
madethe binary copy of the database cluster files of the postgres master. However, I recently noticed that
pg_basebackupdid not copy the the replication slot object of the master, in comparison with the fact that the
copy-command-based-backupdid copy the replication slot object. Is this difference designed on purpose ?
Considering the difference mentioned above, I'm wandering that is there anything else that the pg_basebackup would NOT
copyfrom the master ?
In other words, what is the no-copying rules of pg_basebackup ?
Any help will be greatly appreciated. Thanks.
Best Regards.
P.S. The details of my test is shown as follows.
1. Edit the postgresql.conf of the master to add replication slot
## ~(Skipping the edit operation of postgresql.conf)~
wing@master:~$ cat /mnt/data/pgdata/master/postgresql.conf | grep "max_replication"
max_replication_slots = 2 # max number of replication slots
2. Create one replication slot on master.
testdb=# select * from pg_create_physical_replication_slot('slot1');
slotname | xlog_position
----------+---------------
slot1 |
(1 row)
testdb=# SELECT * FROM pg_replication_slots;
slot_name | plugin | slot_type | datoid | database | active | active_pid | xmin | catalog_xmin | restar
t_lsn
-----------+--------+-----------+--------+----------+--------+------------+------+--------------+-------
------
slot1 | | physical | | | f | | | | 7/AD00
0220
(1 row)
3. Execute the pg_basebackup to make the base backup on the slave machine as slave1
## ~(Skipping the preparation operation of the pg_basebackup )~
wing@ubslave:~$ pg_basebackup -h {Master IP} -p 5432-U wing -F p -P -x -R -D /home/wing/pgdata/slave1 -l
baseup_slave1.log
4. Confirm the replication slot settings of slave1
## ~(Skipping the operation for starting the slave1 cluster )~
## The 'max_replication_slots' settings of postgresql.conf equals with the one on master.
wing@ubslave:~$ cat /home/wing/pgdata/slave1/postgresql.conf | grep "max_replication"
max_replication_slots = 2 # max number of replication slots
## However, there is NO RECORD in “pg_replication_slots” of slave1.
testdb=# SELECT * FROM pg_replication_slots;
slot_name | plugin | slot_type | datoid | database | active | active_pid | xmin | catalog_xmin | restar
t_lsn
-----------+--------+-----------+--------+----------+--------+------------+------+--------------+-------
------
(0 rows)
5. Use the “pg_start_backup -> copy -> pg_stop_backup” method to make the base backup on the slave machine as slave2
## 5.1 execute pg_start_backup() on master
## 5.2 scp the data files from master to slave2.
## 5.3 execute pg_stop_backup() on master
6. Confirm the replication slot settings of slave2
## ~(Skipping the operation for starting the slave2 cluster )~
## The 'max_replication_slots' settings of postgresql.conf equals with the one on master.
wing@ubslave:~$ cat /home/wing/pgdata/slave2/postgresql.conf | grep "max_replication"
max_replication_slots = 2 # max number of replication slots
## Differing from slave1, slave2 had a record in “pg_replication_slots” which is similar to the one of master.
testdb=# SELECT * FROM pg_replication_slots;
slot_name | plugin | slot_type | datoid | database | active | active_pid | xmin | catalog_xmin | restar
t_lsn
-----------+--------+-----------+--------+----------+--------+------------+------+--------------+-------
------
slot1 | | physical | | | f | | | | 7/AB00
C540
(1 row)
Both master and slave are running PostgreSQL 9.5.0 on Ubuntu Server 14.04(64 bit)
--
Best Regards
---------------------------------------------------
FNST) Yi Yi
---------------------------------------------------
