Amit Saxena

> The reason why you don't have the problem with Heroku looks be that keep-alive isn't used between the Heroku router and your app (Puma): https://devcenter.heroku.com/articles/http-routing#http-versions-supported Also thanks for this!...

> If you are running Puma behind a load balancer, the load balancer should ideally be configured with one keep alive connection per worker thread. This is the absolute best...

I tried different configurations and below are my observations: 1. `max_fast_inline 0` result in slightly higher latencies than without keep-alive for me but mostly works fine. The code says this...

> Try using [least_outstanding_requests](https://registry.terraform.io/providers/hashicorp/aws/latest/docs/resources/lb_target_group#load_balancing_algorithm_type) @MehdiZonjy That's what we used during the load tests. > See if you could enable [ALB access logs](https://docs.aws.amazon.com/elasticloadbalancing/latest/application/load-balancer-access-logs.html). I'm interested in the various request|target|response_processing_time values. They...