Aiming to 9.6

Features we need to implement

• New configuration file for this feature (or new GUC is in postgresql.conf)
• Parsing configuration file.
• Only 1-level configuration with 'priority' method and its logic. (Maybe also 1-level 'quorum' method)
• New system catalog.

Replication Type

priority

The each standbys has priority according to definition order, starting from left. If we want to set the set of standbys as a priority group, we can set by surrounding with '[' and ']'. The master server waits for "N" standbys having higher priority. priority 1 is highest priority.

quorum

The each standbys always has same priority 1, which means every standbys are equal priority. If we want to set the set of standbys as a quorum group, we can set by surrounding with '(' and ')'. The master server waits for "N" standbys from quorum group.

Syntax

N(node1, node2, ...):group1

The comma separated node list surrounded by '( ... )' represents the standby will be handled as a quorum group. N means that the number of nodes/group to wait for in this group. Using ":" immediately after set of node name, we name the group name to it.

N[node1, node2, ...]:group1

The comma separated node list surrounded by '[ ... ]' represents the standby will be handled as a priority group. N means that the number of nodes/group to wait for in this group. Using ":" immediately after set of node name, we name the group name to it.

Example

If we configure

'2[local1, 1(london1, london2):london, [tokyo1, tokyo2]:tokyo]'.

it means:

• The 'london' group has two nodes; london1, london2, and is handled as quorum group. It means that 'london' group is active only when at least 1 standby returns ACK to master.
• The 'tokyo' group also has two nodes; 'tokyo1', 'tokyo2', and is handled as quorum group. It means that 'tokyo' group is active only when 'tokyo1' returns ACK to master. If 'tokyo1' is down, the master waits for 'tokyo2'.
• The 'main' group has one node(local1) and two groups(london, tokyo), and is handled as priority group.
• The master can do COMMIT when ACK is returned from local1 node and london group, because local1 and london priority are higher than tokyo group.

System View

We have a new system view, says pg_stat_replication_group, to easily confirm the status of sync replication including each group.

The definition is,

Table "public.pg_stat_replication_group"
     Column     |  Type   | Modifiers 
----------------+---------+-----------
 name           | text    | 
 sync_type      | text    | 
 wait_num       | integer | 
 sync_priority  | integer | 
 sync_state     | text    | 
 member         | text[]  | 
 level          | integer | 
 write_location | pg_lsn  | 
 flush_location | pg_lsn  | 
 apply_location | pg_lsn  | 

column	description
name	node name or group name, or "main" meaning top level node.
sync_type	'priority' or 'quorum' for group node, otherwise NULL.
wait_num	number of nodes/groups to wait for in this group.
sync_priority	priority of node/group in this group.
sync_state	'sync' or 'potential' or 'quorum'.
member	array of members for group node, otherwise NULL.
level	nested level. "root" node is level 0.
write_location	group/node calculated LSN according to configuration.
flush_location	same as above
apply_location	same as above