[pgpool-general: 9198] pcp_recovery_node with snapshot isolation and large database

[Bill Carlson]

Hello,

Ran into the issue where attempting a pcp_recovery_node with mode 
snapshot_isolation fails due to active database connections. Have a 
large database that takes hours to pg_basebackup, can't afford closing 
the database connections for that long.

Any known practices to work around this?


Have two thoughts myself:

1) Run through recovery_stage_two process manually, first closing all 
connections, then running the related script. What else is done outside 
of the script, besides likely pcp_attach_node?

2) Add a method to close all user connections to the end of 
recovery_stage_one script. Add inverse to end of recovery_stage_two. 
Seems to me there are two big steps for this.
    a) Can't trust any pg_* tables unless going to individual nodes. How 
to close connections? Am I missing a    pgpool method that would all one 
to close user connections, as in not superuser?
    b) Cannot stop/start Postgres itself, as that would stop the recovery.


Thanks,

-- 
Bill Carlson

Carlson Computer Services
http://apgtm.com/

"Anything is Possible Given Time and Money"