Distributed Read Cache direct database connection doesn't work with Omnibus HA Postgres setup
Similar to #3356 (closed) there's another potential single point of failure with the in-memory cache for reads distribution added recently.
The cache needs a direct connection to Postgres and this is where it can fail. If the Postgres database goes down for any reason then this feature will stop working that in turn will impact performance heavily. Solutions for Postgres failover elsewhere in the product aren't available as they are all pretty much based on a PgBouncer + Consul setup, which can't be used here as PgBouncer doesn't support LISTEN operations.
As such I think solutions should be explored for this as I fear Distributed Reads failing is something that may tend to be missed more as nothing will break outright.
cc @8bitlife