[pgpool-hackers: 897] Idea for 3.5: switch over script

[Tatsuo Ishii]

Now that PostgreSQL 9.5 has pg_rewind in core, I would like to propose
adding new directive "switchver_script", whose function is almost same
as failover_script except that the former is called only when switch
over is requested, which will be triggered by new command
"pcp_switchover".

Switchover_script will execute pg_rewind, thus has an advantage over
pg_basebackup or rsync (or whatever) used to recover new standby, in
terms of recovery duration because pg_rewinrd does not require copying
whole database cluster. The reason why pg_rewind cannot be used for
failover_script is, pg_rewind will not work for the primary server
crash case. That's the reason why we need to have a separate script
for the switch over case.

pcp_switchover works like this:

1) pcp_switchover --mode prepare --node node_number

   Prepare switchover. From now on, pgpool-II will call
   switchover_script rather than failover_script when a node goes
   down. "--node" specifies a node to be promoted after switchover. If
   this parameter is not specified, pgpool-II will automatically
   select a standby node which is the first live standby node in
   pgpool.conf.

2) After #1 is completed, the primary node is shutdown by an
   operator. This triggers switchover and pgpool-II executes the
   switchover_script. After switchover the node specified by --node
   option will be promoted.

3) pcp_switchover --mode stop

   If the operator decides to cancel the switchover work between
   "pcp_switchover --mode prepare" and shutting down the primary for
   some reason, this commands should be executed. This will clear
   "switchover mode" and any event after this will trigger
   failover_script as usual.

Comments and suggesions are welcome.
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp