Publish statistics in Atlas to analyze what is filling oplog
It would be very useful to be able to see metrics/statistics about the contents of oplog. There are open-source tools like oplog analyzer (https://github.com/mhelmstetter/oplog-analyzer) that can be used, but it's a hassle to have to install it and run it in the same datacenter where the database is running (for performance).
The statistics I'm most interested is what collections have most oplog documents, what kind of operations they have been and what is the total size that each collection currently utilizes from the oplog. This will help improving code to use less oplog.
We've seen cases where bad code that always reads the whole document, adds something to e.g. an array and then saves the whole document again, causing huge amounts of oplog and causing constant pressure on making the oplog bigger, thus making the whole DB size bigger.
We'd love to see this implemented.
This is a great idea, Kai. In the iterim, Is there anyway you can use the Profiler to see load by collection to infer what the oplog weight ought to be?