While Atlas provides the 'test failover' feature, this only takes effect on the primary of each shard. Customers who run large sharded clusters on Atlas require testing the effects of a full maintenance patch that would restart all mongod's and mongos's, config servers, etc. This feature is a request to improve the chaos engineering features of Atlas.

At present, customers are being advised to change the TLS version of the cluster which could force restarts of all mongod's, mongos's and config servers. While this may be acceptable in non-production environments, it is unlikely in production environments.

I believe there's an internal ticket for this already:
https://jira.mongodb.org/browse/PRODTRIAGE-3411