[pgpool-hackers: 2004] Re: Proposal to make backend node failover mechanism quorum aware

[Tatsuo Ishii]

> You can also understand "auto-failover" as preserving the connection. Right
> now if you have a connection with a master and a stand-by node, the pgpool
> "show pool_nodes" gives you this...
> 
> [ec2-user at ip-172-25-1-32 bin]$ ./edb-psql  edb -p 9999 -U ec2-user
> 
> edb-psql (9.6.1.6, server 9.5.3.8)
> 
> Type "help" for help.
> 
> edb=# show pool_nodes;
> 
>  node_id |  hostname   | port | status | lb_weight |  role   | select_cnt |
> load_balance_node | replication_delay
> 
> ---------+-------------+------+--------+-----------+---------+------------+-------------------+-------------------
> 
>  0       | 172.25.1.33 | 5444 | up     | 0.500000  | primary | 0          |
> false             | 0
> 
>  1       | 172.25.1.34 | 5444 | up     | 0.500000  | standby | 0          |
> true              | 0
> (2 rows)
> 
> if you bring down the stand-by node, the health-check will kick in and if
> you run the same query again it will fail.

That's only happens if the client uses the standby.

> What i mean by auto-failover is if the master (primary server) is
> available, the auto-failover feature should failover to the master instead
> of throwing an error.
> 
> Does that make sense?

To make it possible we need to think about followings:

1) We need to migrate all the session properties set in the previous
   session to the new master connection.

2) If the client was receiving data row from Pgpool-II, it's hard to
   continue it in the new session.

3) If the client was in the middle of a transaction, it's hard to
   continue it in the new session.

These are not trivial, and I think it's not worth the problem, to be
honest.

Best regards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp