Forum Discussion
Parallel refresh delay for tabular model in Power Bi Premium with multiple partition
Hello,
I have a tabular model in Power Bi Premium with multiple partition.
I am expecting the Power Bi tabular model to fire all the respective queries of all the partitions all at once to the database but its firing only few queries at one time and the remaining queries get fired to the database after much delay after the refresh is started.
Could anyone help with the reason and a possible solution?
As the article states you need to modify the TMSL script if you want to change the level of parallelism. Make sure to test on your actual scenario as you can overwhelm the system.
4 Replies
Hi KeshavAgrawalDi ,
I found a blog may provide some help：
The behavior you’re observing is due to the way Power BI handles parallelism during data refresh. By default, Power BI Premium can refresh a maximum of six objects in parallel. This means that if your model has more than six partitions, some of them will have to wait until others have finished refreshing before they can start.
Semantic models in a Premium capacity with the XMLA endpoint enabled for read/write operations allow more advanced refresh, partition management, and metadata only deployments through tool, scripting, and API support. In addition, refresh operations through the XMLA endpoint aren't limited to 48 refreshes per day, and the scheduled refresh time limit isn't imposed.
Best regards.
Community Support Team_CaitlynMy model is XMLA point enabled. I already went through the link which you have mentioned.You mean to say even with XMLA endpoint refresh that barrier of 6 objects in parallel cannot be increased? Could you if possible also provide me the link to the documentations where its written that 6 is the maximum parallel objects refresh power bi premium can handle?
As the article states you need to modify the TMSL script if you want to change the level of parallelism. Make sure to test on your actual scenario as you can overwhelm the system.
I think that's a good thing. This system is not built for massive parallelism, plus in a typical SSAS Tabular model you have dependencies that need to be choreographed anyway.
