To Nha Notes | Jan. 12, 2023, 3:51 p.m.
Document deletion resulting from TTL index expiry is a best effort process. Documents are not guaranteed to be deleted within any specific period. Factors like instance size, instance resource utilization, document size, overall throughput, the number of indexes, and whether indexes and the working set fit in memory can all affect the timing of when expired documents are deleted by the TTL process.
When the TTL monitor deletes your documents, each deletion incurs I/O costs, which increases your bill. If throughput and TTL delete rates increase, you should expect a higher bill due to increased I/O usage. However, if you do not create a TTL index to delete documents, but instead segment documents into collections based on time and simply drop those collections when they are no longer needed, you will not incur any IO costs. This can be significantly more cost effective than using a TTL index.
For time-series workloads, you can consider creating rolling collections instead of a TTL index as rolling collections can be a more performant way to delete data and less I/O intensive. If you have large collections (especially collections over 1TB) or TTL deletion I/O costs are a concern, we recommend that you partition documents into collections based on time, and drop collections when the documents are no longer needed. You can create one collection per day or one per week, depending on your data ingest rate. While requirements will vary depending on your application, a good rule of thumb is to have more smaller collections rather than a few large collections. Dropping these collections does not incur I/O costs, and can be faster and more cost effective than using a TTL index.
https://docs.aws.amazon.com/documentdb/latest/developerguide/best_practices.html