Neither AWS RDS PostgreSQL nor Heroku has a PGBouncer.
Heroku actually does seem to support/encourage it:
https://postgres.heroku.com/blog/past/2013/11/22/connection_... https://github.com/gregburek/heroku-buildpack-pgbouncer
Saw that reference in the linked post from Heroku in the original article.
Has anyone used this? Is it configurable, or do they just choose some reasonable defaults?
Works great for me. There's some minimal configuration explained in the pgbouncer docs. What the pgbouncer buildpack does is just route your DATABASE_URL through the local pgbouncer.
In my python app, I switched from an sqlalchemy connection pool per process (eg per gunicorn worker) to using a Null pool everwhere. No increase in latency; far fewer connections needed.
Or would you have to roll your own pgbouncer setup for these systems?