Is there any middleware such as a NAT or firewall which might be evicting idle TCP connections silently after a some timeout period? We've seen this issue before, and we introduced a few new system properties to help alleviate this issue (docs).
For example: in the past, we had a customer deploy Ignition in AWS. The Ignition instance connected to Auth0 over the Internet through the AWS NAT Gateway. After logging in a few times in a short period of time, if no other logins occur for 350 seconds, the AWS NAT Gateway would close its end of the TCP connection without Ignition being aware until the next time a user tried to login (say 450 seconds later). The stack trace matched what you posted.