With 3scale 2.15 we have moved to forced async mode for 3scale Operator.
This Jira: https://issues.redhat.com/browse/THREESCALE-8185 points to an issue with async not being able to re-connect to Redis when a connection is dropped (temporarily due to network outage or due to Redis issues). The Jira got closed as it could not be reproduced, however, with 2.15 live, we are getting more and more reports that the issue occurs.
Customers are receiving 403 errors and a solution for them is to restart backend listener pods.

In this Jira: https://issues.redhat.com/browse/OHSS-39412 rh-ee-jlledo mentions that:

When the connection to redis is interrupted, it (backend-listener) doesn't crash but it never recovers the connection to redis.

The goal of this Jira is to provide login in backend to attempt to re-establish connection to Redis when the issue occurs.