It does feel like you might be bottlenecked by available RAM, but I'm not necessarily confident that it's responsible for your pooling/connection issues. That said, upgrading to something like a t3.medium instance might alleviate any responsiveness issues that you're seeing.
I'm still trying to understand the core of the issue, though. 
Are you running out of connections to Redshift or to Retool's application database? Both can probably be addressed by tweaking the right environment variables - such as the ones mentioned here - but we need to first identify the precise nature of the problem.