Every member in a replica size will have a copy of the oplog in db.oplog.rs(). The reason for this is that if the primary steps down, one of the secondaries will get elected, and it needs to have an up-to-date version of the oplog for the new secondaries to track.

The oplog size is configurable, and we should set it to be as large as possible. The oplog size doesn't affect the memory usage, and can make or break the database in cases of operational issues.

The reason for this is that if the replication lag increases over time, we will eventually get to the point where the secondaries will fall so behind that the primary server won't be able to read from the primary's oplog, as the oldest entry in the primary's oplog will be later than the latest entry that was applied in our secondary server.

In general, the oplog should be at least one to two days' worth of operations. The oplog should be longer than the time it takes for the initial sync, for the same reason that was detailed previously.