GPnP Initialization, GPnP Get Item and GPnP Termination waits in session trace

Systems are 12c Exadata.

I have a single session repeatedly executing a query.  The plan for the query includes "cell smart scan".  The session spends a lot of time waiting on GPnP waits, and its trace file includes repeatedly (about every second) accessing the gpnp daemon to get the ASM-Profile.DiscoveryString.

I've seen this on another 12c Exadata system once before with an update statement, that was repeatedly executed by sessions which disconnected and reconnected between each execution.  Always connecting to the same node.  A restart of the node cleared up the GPnP waits, and they have not come back for two weeks.