Solved: Re: How to do incremental refresh using datalake o...

jaryszek · ‎09-01-2025

Hello,

like in the topic, how to do incremental refresh using datalake one lake tables (Fabric).

I have tables coming from azure blob storage into one Lake in Fabric.
How to set up incremental refresh ? I have delta parquets tables with year, month and day in the name...

Best,
Jacek

v-agajavelly · ‎09-09-2025

Hi @jaryszek ,

Sorry for delay in responce. Good point in Direct Lake mode you’re right, Power Query isn’t available so you can’t set up incremental refresh the usual way. In that case the trick is to manage incrementality at the data source / lakehouse level.

Since your tables are delta parquet partitioned by year/month/day, you can control what gets landed into the Lakehouse table (for example via pipelines or notebooks).
Direct Lake will then pick up those new partitions automatically without a full reload.
If you need true “incremental refresh policy” (like RangeStart/RangeEnd filtering), that’s only supported in Import / DQ mode today, not Direct Lake.

So you can try bellow ways.
With Direct Lake = keep your data partitioned properly and let Fabric read the latest partitions.
With Import = use Power Query + incremental refresh policy.

Thanks for calling this out — it’s an important distinction between Direct Lake vs. Import.

Thanks,
Akhil.

View solution in original post

v-agajavelly · ‎09-09-2025

Hi @jaryszek ,

Sorry for delay in responce. Good point in Direct Lake mode you’re right, Power Query isn’t available so you can’t set up incremental refresh the usual way. In that case the trick is to manage incrementality at the data source / lakehouse level.

Since your tables are delta parquet partitioned by year/month/day, you can control what gets landed into the Lakehouse table (for example via pipelines or notebooks).
Direct Lake will then pick up those new partitions automatically without a full reload.
If you need true “incremental refresh policy” (like RangeStart/RangeEnd filtering), that’s only supported in Import / DQ mode today, not Direct Lake.

So you can try bellow ways.
With Direct Lake = keep your data partitioned properly and let Fabric read the latest partitions.
With Import = use Power Query + incremental refresh policy.

Thanks for calling this out — it’s an important distinction between Direct Lake vs. Import.

Thanks,
Akhil.

v-agajavelly · ‎09-01-2025

Hi @jaryszek ,

Thanks for raising this. Since your data is already in delta parquet with year/month/day folders, you can leverage Fabric’s incremental refresh at the semantic model level. Just create RangeStart / RangeEnd parameters in Power Query, filter on your date column, and then configure incremental refresh in the dataset settings. Fabric will push filters down to your OneLake delta table so only new partitions are scanned. This way you avoid reloading full history every time and only process the latest data. Thanks to our super users for sharing these best practices earlier they really make it easier to set up.

Regards,
Akhil.