The evergreen patch shows that there could be a view cycle when we perform reads on the secondary node. Since collMod command won't succeed on the primary if it forms a cycle, this could be a bug for secondary reads.

Update:
Basically, view catalog could reload (loading view definitions from system.views to in-memory data structure) whenever a read comes and there is discrepancy between persistent version and in-memory version of view definition, even during oplog application. Thus, secondary reads could see inconsistent state of view catalog.