Thousands of TCP Connections Remain Established

I’ve raised a ticket in Github, but would like to know if anyone else has hit this kind of problem or has advice. Essentially, a certain app that creates outgoing HTTP connections to a k8s service accumulates thousands of ESTABLISHED TCP connections, sometimes until the count exceeds 8000 and the linkerd container crashes.

The problem is specific to a single cluster even though we run common code and use many other clusters with the same cloud provider.

Linkerd Enterprise 2.15.5-0