# Help with pgpool settings, **URL:** \ **Category:** PostgreSQL\ **Created:** [May 30, 2024, 2:39pm UTC](https://forums.percona.com/t/help-with-pgpool-settings/30714 "2024-05-30T14:39:46Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![CaptTofu](https://sea1.discourse-cdn.com/flex019/user_avatar/forums.percona.com/capttofu/32/9842_2.png) [@CaptTofu](https://forums.percona.com/u/CaptTofu)\ **Post date:** [May 30, 2024, 2:39pm UTC](https://forums.percona.com/t/help-with-pgpool-settings/30714/1 "2024-05-30T14:39:46Z")
Hi all, longtime MySQL/ProxySQL guy here trying to come up to speed with PostgreSQL and pgpool. A bit steeper than proxysql was. I’m encountering a few issues and can’t find any discussion as to a situation where on my pool, I see a lot of “pgpool: wait for connection request” on the host. My setup: 7 Percona Postgresql 15, one primary, the rest standbys as backend 1 pgpool-II version 4.5.0 (percona) server I’m trying to find the optimal settings for connections, timeouts, etc. as we’ve had to do a bunch of testing to handle the connections needed for the app. I’ve ended up setting the pg backends to max connections of 1000, then pgpool, num\_init\_children 500, max\_pool 10, child\_max\_connections to 1000, client\_idle\_limit 60. Am I missing something obvious? Also, tps using pgbench is around 220 throught he pool, but direct to the master is 340. Is this a normal decrease, or yet another thing I’m missing? yes, I am using memcached query cache. Thank you! ps. Boxes are all 32 procs, 128 GB RAM, ubuntu 22.04 bhvye VMs, running on smartOS/smartCloud ---
**Author:** ![muhammad.usama](https://avatars.discourse-cdn.com/v4/letter/m/e9c0ed/32.png) [@muhammad.usama](https://forums.percona.com/u/muhammad.usama)\ **Post date:** [June 3, 2024, 8:24am UTC](https://forums.percona.com/t/help-with-pgpool-settings/30714/2 "2024-06-03T08:24:47Z")
7 Percona Postgresql 15, one primary, the rest standbys as backend 1 pgpool-II version 4.5.0 (percona) server > I’m trying to find the optimal settings for connections, timeouts, etc. as we’ve had to do a bunch of testing to handle the connections needed for the app. I’ve ended up setting the pg backends to max connections of 1000, then pgpool, num\_init\_children 500, max\_pool 10, child\_max\_connections to 1000, client\_idle\_limit 60. Am I missing something obvious? `num_init_children` determines the number of concurrent connections you want to support. Setting it to 500 allows for a maximum of 500 concurrent connections. This setting should be based on your application requirements. The `max_pool` value depends on the number of distinct [db-name, db-user] pairs the application uses to connect to the database service. Typically, applications use one or two db-users, so this value is usually kept below 5. A higher `max_pool` value increases the `max_connections` requirements on the backend. If you use the settings for `max_pool` and `num_init_children` mentioned above, the ideal setting for `max_connections` on PostgreSQL would be `num_init_children * max_pool` (500 \* 10 = 5000). So, I suggest adjusting the `max_pool` and `num_init_children` values or increasing the `max_connections` on PostgreSQL. For higher values of `num_init_children`, you should consider enabling `serialize_accept` in the Pgpool configuration. However, `serialize_accept` is very load and system-dependent, so I recommend benchmarking with and without `serialize_accept` under your expected load. Additionally, you should enable `dynamic_process_management` mode with these `num_init_children` settings. More information on process management mode can be found [here](https://www.pgpool.net/docs/latest/en/html/runtime-config-process-management.html#GUC-PROCESS-MANAGEMENT-MODE). Regarding `client_idle_limit`, this depends on your workload and application expectations. If your application does not hold connections and only creates a connection when needed, performs some tasks, and then disconnects, this timeout can be as low as 10 seconds. An idle connection from the application, in this case, would indicate a problem, as it has not released the connection after use. Lastly, if you are not setting `max_connections` on PostgreSQL to `num_init_children * max_pool`, then enable `connection_life_time` as well. > Also, tps using pgbench is around 220 throught he pool, but direct to the master is 340. Is this a normal decrease, or yet another thing I’m missing? Pgpool-II performs better than direct connection when database and workload size crosses a certain limit. See the blog for reference https://www.highgo.ca/2019/09/06/can-you-gain-performance-with-pgpool-ii-as-a-load-balancer/ Thanks Best regards Muhammad Usama ---
**Author:** ![CaptTofu](https://sea1.discourse-cdn.com/flex019/user_avatar/forums.percona.com/capttofu/32/9842_2.png) [@CaptTofu](https://forums.percona.com/u/CaptTofu)\ **Post date:** [June 3, 2024, 10:53am UTC](https://forums.percona.com/t/help-with-pgpool-settings/30714/3 "2024-06-03T10:53:17Z")
Thank you Muhammad, I will try these settings. Also, per our other issue wrt to the bug with pg\_catalog queries, I wanted to make sure you saw my reply that 4.3 [Postgresql.org](http://Postgresql.org)’s packages also have this bug. We built 4.5.2 and it doesn’t seem to have the bug.