There are a number of places where we own TaskExecutors and ThreadPoolTaskExecutors by unique_ptr. One such example is in the replication coordinator. These are dangerous because a TaskExecutor might invoke shared_from_this in its internal implementation, and those calls will fail because nothing has already taken a shared_ptr to the object. TaskExecutor already does this, albeit in just one infrequently called place, but there's plans in SPM-3750 to use extend this pattern to core functionality of ThreadPoolTaskExecutor (i.e., scheduleRemoteCommand variants and scheduleWorkAt).

In general, we should consider it a bug to ever create or own an object that inherits from enable_shared_from_this in any other fashion than by shared_ptr because otherwise, we risk the internal implementation calling shared_from_this and failing. I recommend we write a clang tidy rule to prevent such cases and use that as a guide for removing them.