Rueidis clears the local cache on disconnections.

Since the invalidations are delivered on the same client tcp connection, they are guaranteed to be delivered but may be delayed for various reasons. Rueidis uses periodic background pings to mitigate the delay on a tcp connection. It also requires you to specify a local TTL on the cache entry to further reduce the chance of getting staled data.

To monitor the delay, I think you can watch the client_longest_output_list metric to see if it keeps growing.