Power BI Capacity - Background Operations Limits

[joeparkinson]

We have a P2 capacity, and previously, the background operations were limited to 25% of our total capacity, and this protected the users from anything we might have been doing in the backend. Suddenly, this limit has been removed and the background operations are exceeding 80% of our capacity. Is there a way to limit the background operations? As they are being smoothed out across 24 hours, it means our users essentially can't properly use the service for the rest of the day. This is causing huge issues

[lbendlin]

Welcome to the club.  Your CU consumption will be fine until it isn't.  All it takes is a single rogue dataflow that keeps running for a couple hours (ignoring the 5 hr limit for semantic models !!!) and once it completes - BOOM - you are hit with a huge debt that not only immediately brings your P2 over capacity but will also take a loong time to pay back (the 24 hours you mentioned).  We have a ticket open with Microsoft because this is just outrageous, and we are flying blind, always in fear of some large transaction finishing.  The entire charge model needs to change.  We need to be able to see what is in flight, and about to hit the CU , and we need to be able to pay back (burndown) sooner if we have spare CUs, rather than having to wait for 24 hrs.

Worst case scenario:  You have a large action completing, bringing you into rejection territory, first for interactive requests and then also for refreshes. You did not enable autoscale.  Result:  Paralysis.  Remedy: Move ALL workspaces off that capacity IMMEDIATELY. Issue:  You may not have a spare capacity to move stuff to.

Ask me how I know.

Please raise a Pro ticket as well.

[joeparkinson]

Thank you for the reply. This is what I feared. After posting this, I found a rogue gen 2 dataflow and used the API to delete it as you cant cancel them. Luckily, it cleared the issue immediately. We only have the one capacity, but are now investing in a Dev capacity so we can avoid this in the future

[lbendlin]

I don't think a "Dev"  capacity makes sense.  What we have is a "dog house"  capacity where all the misbehaving semantic models go.  All the innocent bystanders are moved out of that capacity.  The devs of the offending semantic models are treated to frequent emails about the CU stats, and to dev clinics. They are not allowed to move out of the dog house until they proved that they refactored in a reasonable way.

[joeparkinson]

My team are the only team that use Power BI in the company, so we have full control over all the objects so thats less of an issue here