Premium Capacity Restart

[Toast36]

Hi All,

I was wondering if there are any guidlines for restarting a Premium Capacity.

Our solution became unresponisve recently which was resolved by a restart but I'm wondering if we should be doing a restart on a regular basis.

Thanks in advance

[nickyvv]

There are these guidelines on the MSFT docs.

Regularly restarting sounds like a bad idea to me, unless you have a very good reason. The underlying problems why you would want to restart would be more interesting to see and try to solve. 

Did this help you or did I answer your question?
Then please give kudos or mark my post as a solution!
My blog: nickyvv.com

[Toast36]

I'm not sure that answer the issue.

We are finding that the Capacity becomes un-responsive and can only be resolved by doing a re-start of the capacity. (This has happend twice over the last 3 weeks)

There are no tools available to us to be able to identify what the root cause of this are and so far Microsoft have only responded by saying it is a "Transient issue" which is less than helpful.

I'm trying to find out if there is either a tool which can give more information to what this "transient issue" might have been or to find out if doing a scheduled restart of the capacity is something that should be considered much in the same way that on-prem servers are often scheduled for a re-start as part of patching windows ect.

[nickyvv]

Do you use the Power BI Premium Capacity Metrics app? With this you should be able to find the root cause, being refreshes, query wait times or other things like dataflows or paginated reports taking up memory or CPU.

Did this help you or did I answer your question?
Then please give kudos or mark my post as a solution!
My blog: nickyvv.com

[Toast36]

Thanks.

Yes we do use the Metrics app which to be honest is not that helpful. Whilst it tells us there was a memory spike at the same time as the issue it does not allow any interrogation to understand what caused said memory spike. Neither does it help identify what report/user might have caused the memory spike.

At the point that the memory spike (max 50gb) has occured there were no dataset refreshes running though a fair few queries (approx 450 but all processing in under 5 seconds with the majority under 50ms) this is not unexpected usage from our consumer base. The interesting comparison is from earlier in that same week we can see there were more queries running, during a dataset refresh and the memory didn't go over 11.74gb.

We currently have Paginated reporting, dataflows and AI disabled as we found they used too much memory to give a stable reporting environment.

The metrics app it seems cannot help to explain this behaviour.