50-60ms per-query overhead in PgPool-II 4.7 vs direct connection

Hi PgPool team,

We run a payment processing system with PostgreSQL 18 and PgPool-II 4.7.0.
Each transaction executes ~75 queries. We have noticed significant overhead
compared to direct connection:

*Measurements:*

-

Direct to PostgreSQL: ~50ms per query
-

Through PgPool: ~100-110ms per query
-

Difference: +50-60ms per query

*Production impact:*

-

75 queries × 50ms extra = ~3.75 seconds overhead per transaction
-

Full transaction: 2-3 seconds with PgPool vs 0.5-1 second direct

*Current PgPool configuration:*
ssl = off
load_balance_mode = off
num_init_children = 1100
max_pool = 1
connection_cache = on
child_life_time = 300
connection_life_time = 300

*What we already tried (no improvement):*

-

ssl = off (helped a little)
-

num_init_children = 300-800 (no change)
-

child_life_time = 0 (no change)
-

connection_life_time = 1800 (no change)
-

max_pool = 4 (no change)
-

relcache_size = 1024 (no change)
-

use_remote_estimate and fetch_size on FDW (improved FDW queries)

*Question:*
Is there any parameter or configuration to reduce per-query overhead in
PgPool-II? Is it possible to achieve near-direct connection times while
keeping failover and load balancing?

Any guidance is greatly appreciated.

Best regards,

Ernesto Montilva

--