If you’ve been following the releases announced at FabCon, you must have seen one of the bigger Fabric ones: the on-demand capacities.
Now, this blogpost is based on my personal, private preview, experiences. The documentation is not yet available. I’ve been told it won’t take long before that will be available; just as the option to actually deploy it will appear soon.
What is it?
You can compare it to Serverless SQL databases: when it’s not in use, it won’t cost anything. Storage will always incur some cost, but no compute.
First, be aware that there are no reservations (reserve plans soon) available. Second, pricing to use items will differ from the regular capacities. Some items may have the same price; others will be calculated at x2 or x3. We have to wait for the official documentation to find out what the actual prices will be.
Can I set boundaries?
You most certainly can. When you deploy on-demand capacity, you can set a maximum number of CUs for the available items. This doesn’t mean it will always run at that level; it will, however, cap it.
This is akin to the Spark settings you have today.
What are the use cases?
In my daily work, the main use case will be development. Why? Well, first of all, development isn’t a continuous process. Not only is it not a 24/7 process, sometimes it’s just one day in a week. If you have a dedicated capacity, chances are that it’s being there doing nothing. And sure, you can pause it, but how many times did you forget to do that?
And you can also automate the pausing using Azure Automation for instance. We’ve done that, but also seen that someone worked late and all of a sudden everything stopped working.
It’s annoying. Using the F0 solves this. Also, if you have a light environment where the ETL/ELT runs an hour or two during the night, then refreshes import semantic models, and sleeps for the rest of the day, F0 is a great match.
Cost?
Yes, it will still cost money. How much? Good question that I’ve also asked Microsoft. Their answer is to wait for the official docs as, right now, it’s complicated because there are a number of dependencies. In general, I’d wager that the examples I’ve given above are a great match for on-demand capacities. There will be a point where dedicated capacities (reserved preferably) will make more sense.
Homework?
Yes, when it’s available, go and try it out. Make sure not everyone can add workspaces to the capacity, and keep a sharp eye on cost and usage to find if it’s a match for you.
And, let me know what your experiences are, I’d love to learn from you!
