[pgpool-general: 4501] Determining pgpool child memory footprint

[Kennedy, Robert]

We are attempting to size the memory for a pgpool-II 3.5.0 instance and would like to know what things a child will use private resident memory for above and beyond what is necessary for TCP connection and socket handling. We are not using memcache and are not using load balancing, failover, or replication. In this environment pgpool is strictly used to manage a flood of connections for a distributed application that cannot do its own pooling. There is only one <user,database> tuple used for all the connections and nearly all of the connections consist of a single row UPDATE.

In our testing we've seen widely varying amounts of resident memory usage for pgpool children and wonder what is driving it. We believe our various test cases are not so different in their behavior as to generate as much difference as we are seeing between test runs.

No memory appears to be leaking, and the children are being recycled without issue when the service is idle.

Cheers,
Robert Kennedy