Forum Discussion
Using Import mode for power bi dataset in Analysis Services
Hi Anonymous , Thank you for reaching out to the Microsoft Community Forum.
There is no hidden row limit when importing a Power BI dataset over the XMLA/Analysis Services endpoint, the engine returns whatever the semantic model currently contains. The behaviour you observed (3.5M -> 5M after adding an index) is consistent with a calculated table or partition that was not processed/rebuilt; your edit forced the model to re-evaluate and repopulate the table, which restored the full 5M rows.
Connect to the same dataset with SSMS, DAX Studio or Tabular Editor and run the partitions DMV (SELECT [Name],[RowCount] FROM $SYSTEM.TMSCHEMA_PARTITIONS;) and a DAX row count (EVALUATE { COUNTROWS('YourTableName') }) to confirm what the model currently stores. If the model shows ~3.5M, run a Process/Table or Process/Partition (or TMSL Refresh/ProcessFull) against that calculated table or its partitions to force a full rebuild. If it shows ~5M but Desktop imports 3.5M, ensure your import is pointed at the exact Initial Catalog (dataset/version) and that your identity has full visibility. To prevent recurrence, schedule explicit processing after any source update or avoid relying on implicit query edits to trigger model processing for calculated tables.
DAX Queries - DAX | Microsoft Learn
Process Database, Table, or Partition (Analysis Services) | Microsoft Learn
Partitions in Analysis Services tabular models | Microsoft Learn
Super UserThanks for confirming.
In that case, there are two additional things you can check, as they often cause missing rows when connecting through XMLA:
1. Partition row count mismatch inside the dataset
Even if Incremental Refresh is off, sometimes the model ends up with multiple internal partitions.
You can verify this by running:
SELECT [Name], [RowCount]
FROM $SYSTEM.TMSCHEMA_PARTITIONS
This will tell us exactly how many rows the service has per partition and whether the total actually matches 5M inside the model.
---
2. Relationship or filter propagation issue
If your table is fact-like, a relationship with a dimension table (with Cross filter direction = single) might reduce visible rows.
This won’t show as a Power Query filter but will still reduce the rows returned.
You can test this by connecting to the dataset and running:
EVALUATE ROWCOUNT(FactTableName)
(Replace the table name.)
1. Where do i run this query:
SELECT [Name], [RowCount]
FROM $SYSTEM.TMSCHEMA_PARTITION
In my source dataset , the count shows 5 Mn when i take a Card Visual and put the count
2. There is no relationship. I am just importing this source on a new empty pbi file.