In Oracle 12c a new feature was introduced to the database called "Result Cache Dynamic Sampling", enabled by default. If the shared pool resources are maxed out, the dynamic sampling SQL, originated by Oracle will go into an "Invalid" state, resulting in higher CPU consumption by the Oracle database.

Confluence does not use this cache, thus it is safe to disable the feature by setting the  _optimizer_ads_use_result_cache option to "false" on the Oracle 12c database. This is a stability measure to reduce potential CPU burden on the database server.

 How are dynamic sampling results managed for SQL plan directives?