ha_status

Returns the last-seen HA drain state for this cluster (e.g. “drained”, “draining”, “not_drained”), or None if unknown or not applicable (e.g. a non-HA cluster).

A cluster that is running but reports “draining” is up but NOT ready to serve queries/DML; the routing predicate GPUdb.__is_cluster_usable() treats such a cluster as unusable. This is mutable state, refreshed during discovery and failover.