Documenation update execution in swarm

[vigilanz]

When run in a swarm (not: standalone container), idle connections to the database get dropped after 900 seconds. Hitting all applications that have long-idling connections.

To try out:

• Start database container in swarm, expose port
• Connect via psql from the host
• Do nothing for > 900 seconds
• Send some query. Will fail due to broken connection.

This is a timeout in the system drivers below vip endpoint_mode. It goes away in endpoint_mode of dnsrr, but that does not work with exposed ports. One fix is to setup net.ipv4.tcp_keepalive_time on the host to something below 900 seconds. But not everyone can do that.

One other fix that seems to work is to replace the default launch command to something like this:

(docker-compose.yml):

mydb:
image: ...
command: postgres -c tcp_keepalives_idle=600 -c tcp_keepalives_interval=60 -c tcp_keepalives_count=60

Maybe add this note to documentation. It is annoying for someone new to postgres in docker.