[pgpool-hackers: 256] transaction pooling mode?

[Tatsuo Ishii]

Hi,

It seems pgbouncer's "transaction pooling mode" is pretty attractive
for those who need huge number of connections(like over a thousand) to
PostgreSQL. It gives an illusion that users can have many connections
to PostgreSQL while the actual number of connections stays low.
Transaction pooling mode is different from pgpool's connection pooling
in that it returns connection to PostgreSQL to the connection pool
when a transaction ends(either by commit or abort). On the other hand,
pgpool does not return the connection to pool until the client
disconnects to pgpool.

I think implementing the transaction pooling mode is not terribly hard
in pgpool. Pgpool has single connection data structure to
frontend. Creating a connection structure array for frontend
connection will make it possible to implement the transaction pooling
mode.

BTW, pgpool has the on memory query cache, which is not available in
pgbouncer. If pgpool has both the transaction pooling mode and the
query cache, it could be a very attractive solution for those who need
to handle many concurrent connections with short life, mostly read
only transactions type.

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