max_conns and max_conn_lifetime breaks db.Ping

[matteosuppo]

$ export DATABASE_URL="postgres://*:*@*:5432/auth?max_conn_lifetime=1m&max_conns=20&max_idle_conns=4"
$ ./hydra host --dangerous-force-http                                                                                                                                   
INFO[0000] Connecting with postgres://*:*@*:5432/auth?max_conn_lifetime=1m&max_conns=20&max_idle_conns=4 
ERRO[0000] An error occurred                             error="Could not connect to SQL: pq: unrecognized configuration parameter "max_idle_conns""

Investigating I discovered that it's this call that fails: https://github.com/ory/hydra/blob/master/config/backend_connections.go#L41

Unfortunately the custom parameters hydra uses to control the number of connections appear to break the connection since they're not recognized by the database library.

Not sure if this happens with all postgres databases. I tried with our production one and a docker one summoned with

docker run --rm --name postgres -e POSTGRES_PASSWORD=password -p 5432:5432 postgres -d 5

[aeneasr]

Oh ok so the issue is that we need to remove those parameters from the database DSN as apparently there is some validation going on in postgres!

[matteosuppo]

I can attempt a pull request in the next few days

[aeneasr]

I'm already working on it :)

[matteosuppo]

Uhm, it doesn't seem to have been fixed:

➜  hydra git:(master) ✗ docker run -e "DATABASE_URL=$DATABASE_URL" -e "SYSTEM_SECRET=$SYSTEM_SECRET" --rm --name my-hydra -p 4444:4444 oryd/hydra:v0.8.3-http -h
time="2017-05-24T08:48:42Z" level=info msg="Connecting with postgres://*:*@*:5432/auth?" 
time="2017-05-24T08:48:43Z" level=error msg="An error occurred" error="Could not connect to SQL: pq: unrecognized configuration parameter "max_conn_lifetime"" 
time="2017-05-24T08:48:43Z" level=info msg="Retrying in 0.100000 seconds..." 

Am I doing something wrong?

[aeneasr]

No, my bad, I updated the log but not the actual URL that get's connected. I also wrote test cases but unfortunately the postgres instance I use does not seem to care.

[aeneasr]

@matteosuppo what postgres version are you running by the way?