[pgpool-general: 7750] Error: sorry, too many clients

michail alexakis drmalex07 at gmail.com
Sun Oct 10 04:32:41 JST 2021


Hello PgPool

We are using PgPool 4.2.4 (from Alpine 3.13) as the frontend of a
master/standby PostgreSQL 10.14 cluster (2 synchronous standbys in
streaming replication mode). PgPool is operating as a load balancer and a
connection pool.

PgPool forks 90 pool workers (num_init_children=90, max_pool=2). All
PostgreSQL backend servers are configured properly to accept a maximum of
190 connections (so, num_init_children * max_pool is below this maximum
with a safety margin of 10).

The relevant (pool-specific) part of the configuration is:
num_init_children = 90
max_pool = 2
child_life_time = 5min
child_max_connections = 0
connection_life_time = 0
client_idle_limit = 0

After a while (5-10 days of operation), we face several errors of the type
of "Sorry, too many clients".  These errors are not originating from a
PostgreSQL backend (error logs are completely clear and pg_stat_activity
shows no more than 140 (from 190) active connections).

This problem has been around for some time now, and we have to resort to
restarting the entire PgPool service. Of course, this is not a perfect
solution because it disrupts some database clients (so, we also have to
cascade those restarts to some other services, and so on).

The majority of database clients are connecting from a Hikari pool (Java
Spring/Boot) which creates long-standing connections (i.e. not one
per-transaction).

Any light on this issue is highly appreciated!

Kind regards,
Michail Alexakis, Software Engineer
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.pgpool.net/pipermail/pgpool-general/attachments/20211009/e577ffca/attachment.htm>


More information about the pgpool-general mailing list