[pgpool-hackers: 2444] Dealing with logical replication

[Tatsuo Ishii]

Now that PostgreSQL 10 will introduce logical replication, Pgpool-II
may need to do something. Here are my thoughts.

1) Need to have new mode

We have streaming replication mode, native replication mode, master
slave mode and raw mode. Logical replication replication does not fit
into none of them because: 1) slave can be updated 2) not all table
are replicated. So we need a new mode, say, "logical replication
mode".

2) If #1 is our direction, pgpool_setup need to deal with it. Since
the tool is aiming for testing/debugging, probably

- Similar to the native replication mode (i.e. physically replicating
  databases in the initial processing).

- Recovery script or failover script is empty.

- Do we need a initial publication/subscription setting?

3) What is "logical replication mode" anyway?

In my mind it should behave like the streaming replication mode. It
may be required that only particular tables are load balanced. For
now, we could ask users to use *redirect directives. More sophisticated
load balancing could be invented, but I feel like it's too early to
design it (maybe we should start it in Pgpool-II 3.8?)

An interesting idea would be, "auto copying DDLs". As you might know,
logical replication does not replicate DDLs. It maybe convenient for
users to let DDLs be copied by Pgpool-II (same thing can be said to
TRUNCATE).

Comments are welcome.

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