I am of the impression that the problem of having waiting clients does not depend from server_idle_timeout
How is the load on both ends of pgbouncer? High? Low? No activity? A lot of content switch perhaps?
I think that a pool_size of 120 is very high, unless you have an XXL database server. Too high numbers will bring lower performances.
Also i think it would be of interest to know the values you are using for:
default_pool_size min_pool_size and reserve_pool_size
and of main importance is to actually know what are the active connections doing. Are those all active? Did perhaps somebody forgot to close the connection to the db in your application and they are only waiting to be timed out?
About 'DISCARD ALL', that is the default 'server_reset_query'.
regards,
fabio pardi
On 10/10/18 17:03, Nicola Contu wrote:
Hello,
we are running pgbouncer 1.9.1 connected to postgres 10.5
Sometimes we are seeing a lot of waiting connections with this query :
DISCARD ALL
This is our pgbouncer config :
[databases]
dev = host=10.10.10.1 port=5432 dbname=dev auth_user=pgbouncer pool_size=120
[pgbouncer]
listen_port = 6543
listen_addr = *
auth_type = md5
auth_file = /etc/pgbouncer/users.txt
auth_query = select uname,phash from user_lookup($1)
logfile = /var/log/pgbouncer.log
pidfile = /home/postgres/pgbouncer.pid
admin_users = admin
user = postgres
max_db_connections = 220
log_connections = 0
log_disconnections = 0
Do you think this can depend on the server_idle_timeout default config value?