Query Folding with Databricks.Catalogs Connector

[Anonymous]

Hi all,

I'm using Databricks.Catalogs connector to get data from Databricks Database Tables in the import mode (using the online Dataflows). 

The problem is that my query is not folded even though I do not use any complex transformation logic. However, when I inspect the Query plan for the last step of the query, I can see only nodes which are folded.

What may be the causes behind the query not folding in this scenario?

Thank you.

Boris.

[badac002]

Bumping this topic. Can someone point me to a guide on query folding in Power Query with the Azure Databricks connector?

The behavior mentioned in the above post is still present in 2024 in Power Query Online. In my opinion, it is unacceptable that simply selecting a table from Unity Catalog yields a "Will not fold" indicator. Even in Power BI Desktop, users will not see the "View Native Query" option, which would indicate the query is folding. I should not have to instruct users to use the Azure Databricks connector and then edit their M code to switch to Databricks.Query (which has no documentation other than this random LinkedIn post https://www.linkedin.com/uas/login?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Fpulse%2Fquery-f...

I can't be the only only one having trouble with this. I am trying to instill best practices with my community, and it is difficult when the Azure Databricks connector behaves so differently from others. Can anyone point me to a proper guide on how users can be sure their queries are folding with Databricks?

[Stachu]

It seems that when the native query is provided in the connection UI the fold is also not breaking. PowerBI wraps the Databricks.Catalog in Value.NativeQuery, like this:

let
    Source = Value.NativeQuery(Databricks.Catalogs(server, http, [Catalog="samples", Database=null, EnableAutomaticProxyDiscovery=null]){[Name="samples",Kind="Database"]}[Data], "select * from nyctaxi.trips", null, [EnableFolding=true])
in
    Source

When analysed using the Diagnose feature it shows the source query is folded even though the View Native Query option is greyed out, but I must say I didn't verify it on the Databricks end

[jakaihammuda]

Yep, im struggling to get it working. 

[Anonymous · Accepted solution]

I found a great solution and does not break Databricks Query Folding 

use Databricks.Query instead of Databricks.Catalogs

If you want to lightweight your loading process

create a parameter  like date_filter

let
    Source = Databricks.Query("warehouses" , [] ),
    part1 = "select * from TYPE YOUR TABLE",
    part2 = " WHERE snapshot_date > "&"'"&date_filter&"'",
    MyQuery = Source(part1 & part2)
in
    MyQuery

[cristiano_lopes]

Done:

= let
Source= Databricks.Query("xxxxx.cloud.databricks.com", "/sql/1.0/warehouses/xxxxxxx", [Catalog="hive_metastore", Database="schema_xxxx", EnableAutomaticProxyDiscovery=null]),

hive_metastore_Database = Fonte("SELECT * FROM hive_metastore.schema_xxxx.tablename T")
in
hive_metastore_Database

[Anonymous]

Hello gustavoleo_RJ,

thank you. This seems to help with my issue indeed.

For some reason I can't find the documentation of the Databricks.Query in the M language reference. Do you know if it is there?

Thank you.

Kind regards,

Boris.

[v-jingzhang]

Hi @Anonymous 

Can you show the M code of this query? You can open Advanced Editor and find all code there. Or you can use Query folding indicators to check which steps are preventing your query from folding. Reorder them if necessary to increase folding.

Best Regards,
Community Support Team _ Jing

[Anonymous]

Hello Jing,

thank you for your message.

• the M code is:

let

Source = Databricks.Catalogs(#"Server Hostname", #"HTTP Path"),

  Navigation = Source{[Name = #"Database Name", Kind = "Database"]}[Data],

  #"Navigation 1" = Navigation{[Name = #"Schema Name", Kind = "Schema"]}[Data],

  #"Navigation 2" = #"Navigation 1"{[Name = #"Table Name", Kind = "Table"]}[Data]

in

  #"Navigation 2"

Unfortunately, I'm strungling to understand which step migh prevent the query folding, as there is no other data transformation besides the unpacking of the content of the [Data] columns into the final table.

Kind regards,

Boris.