Re: GMT FATAL: remaining connection slots are reserved for non-replication superuser connections, but I'm using pgBouncer for connection pooling

On Sat, Feb 25, 2017 at 4:19 AM, lisandro <rostagnolisandro(at)gmail(dot)com>
wrote:

> Hi there! Please tell me if this isn't the place to post my question, I'm
> new
> in the list.
>
> I'm using PostgreSQL 9.3, I have around 150 databases, and I use pgBouncer
> for connection pooling.
> My server is a VPS with 8cpus and 24gb of RAM.
>
> My current postgreSQL configuration (resumed) is this:
>
> listen_addresses = '*'
> port = 6543
> max_connections = 250
> shared_buffers = 2GB
> effective_cache_size = 6GB
> work_mem = 10485kB
> maintenance_work_mem = 512MB
> checkpoint_segments = 32
> checkpoint_completion_target = 0.7
> wal_buffers = 16MB
> default_statistics_target = 100
>
>
> In the other hand, my pgBouncer configuration (resumed) is this:
>
> listen_addr = localhost
> listen_port = 5432
> pool_mode = transaction
> server_reset_query = DISCARD ALL
> max_client_conn = 10000
> default_pool_size = 10
> min_pool_size = 2
> server_idle_timeout = 30
> ...

Note that pgBouncer pool size is per user/database pair. With these
settings and 150 databases I'm actually surprised that you aren't running
out of connections more often. Perhaps there are per-database settings that
haven't been shown. We are also missing info on reserve_pool_timeout,
max_db_connections, etc. which could all play a role, here.

Cheers,
Steve