Forum Discussion

amir_mm's avatar
amir_mm
Helper III
2 years ago
Solved

Incremental refresh and refresh in Analysis services

Hello,

I have defined an incremental refresh policy to refresh the last 6 months and archive the entire data (5 years) to reduce peak memory usage during a refresh. But, I don't see much difference in terms of the number of refresh failures or even the refresh duration. I checked the partitions in Analysis Services, and when a refresh ends successfully, I can see that only the partitions for the last 6 months have been updated (which aligns with my incremental refresh policy). However, when I refreshed the 2 tables with the incremental refresh policy in SSMS, I noticed that the entire dataset had been transferred. Yet, after checking the partitions, only the last 6 months had been processed.

This leads me to the impression that during the refresh, only the last 6 months are updated, but the entire dataset is processed. This could explain why I don't see much difference in terms of memory usage and refresh time.

 

Below, you can see that almost 9.5 million and 8.5 million rows have been transferred for these 2 tables, which represent the entire dataset (5 years of data) :

 

 

I would appreciate it if you could help me understand how this works.

Thanks!

 

  • When refreshing multiple partition you can influence the level of parallelism.  May want to tone that down a bit, or even go fully sequential.

5 Replies

  • Which refresh type did you specify?

    Do you have table dependencies (like Auto Date/Time)?  - Check the refresh log to see if it processes other tables when you asked for the processing of a particular table.

    • amir_mm's avatar
      amir_mm
      Helper III

      I did a "full" refresh, and it processed the entire data but refreshed the last 6 months partitions only.

      I noticed something very strange. We have a premium embedded capacity with 2 workspaces, one for production and one for developments (excatly same configurations), and I published a semantic model into these 2 workspaces. Now I'm constantly getting memory capacity error for the production workspace, while all refreshes completed successfully in Dev workspace. I checked the workspaces settings and I can't find any differences. 

      I did some research but did not find anything special.

       

      Thanks.

      • lbendlin's avatar
        lbendlin
        Super User

        When refreshing multiple partition you can influence the level of parallelism.  May want to tone that down a bit, or even go fully sequential.