RedisClusterInvalidator subscriber does not recover properly on failure

In case of connection error there is a retry policy applied, by default trying to reconnect using an exponential delay.
For the RedisClusterInvalidator subscriber once reconnected the delay is not reinitialized, so next failure will use the previous delay until the timeout is reached.
–
Resolved by using a different retry policy on subscribers, the policy is try for ever with a constant backoff of 2s.