Say you have a problem which is hard to reproduce — e.g. your service gets slow in the early morning but the app-level metrics don’t show anything unusual, and maybe the problem duration is too short for someone to easily catch it in action.

You could set a trigger based on CPU or memory load so e.g. the next spike will capture a few dumps over a set period of time. You don’t have to deal with a ton of data from a simple periodic trigger or having to try to time an irregular event — nothing world-changing but a nice time-saver.

Bleating about how wonderful Microsoft are while they work on locking everyone in to Azure in the background.