Add support for Microsoft Fabric

[mattiasthalen]

Microsoft Fabric is becoming more and more popular with our clients. And most of the time it's new projects or migrations. I.e., the perfect time to implement SQLMesh.

I know that there are some issues with their lakehouse entity and creating schemas / tables. And that's why dbt only work on their warehouse entity.

Anyway, I hope this is something you'd be willing to look into!

[treysp]

Hello and thanks for the suggestion. Unfortunately, Fabric seems to have a number of limitations relative to SQL Server, and some of those are used in core SQLMesh operations. If there is sufficient demand, it may be possible to add Fabric once those limitations are addressed.

[tobymao]

@treysp which limitations do you think are critical? it seems like if the user doesn't do certain operations, we won't hit these limitations?

[treysp]

The limits on ALTER TABLE ADD/ALTER/DROP COLUMN were the main ones that concerned me.

But you're right that those limitations would be feature-specific (primarily on forward-only), so there's no inherent problem if we just tell people the affected features aren't supported.

Sorry for the premature closing @mattiasthalen!

[mattiasthalen]

@treysp No worries!

I'm curious though, are those there for the warehouse object as well? I was under the impression that it only was on the lakehouse.

[treysp]

Totally possible I'm misinterpreting that page - this text made me think it was platform-wide:

Applies to: ✅ SQL analytics endpoint and Warehouse in Microsoft Fabric

This article covers the T-SQL language syntax capabilities of Microsoft Fabric, when querying the SQL analytics endpoint or Warehouse.

[mattiasthalen]

@treysp you are absolutely right, I missed that part!

It sure does have it quirks, e.g., supporting UUID as a data type in the lakehouse/warehouse, but not in PowerBI. So much for consistency 😅

[pycal]

+1

our company has to use fabric, and if a connector for fabric is created we will use sqlmesh.

[tobymao]

we're planning to work on this soon after the holidays

[erindru]

Update on this:

SQLMesh needs some fundamental constructs in order to work properly, namely - schemas, tables and views.

It turns out the Lakehouse entity in Fabric does not support schemas by default. However, if we enable the preview feature for schemas - it stops supporting views. This is documented as a limitation in the public preview of schemas.

We need schemas for the physical layer and views for the virtual layer, so SQLMesh cannot work with Lakehouse until Microsoft resolves this. According to their documentation:

They'll be resolved in the coming releases before General Availability.

So we can revisit this after schema-enabled Lakehouses becomes generally available and theoretically support views.

Of course, the Warehouse endpoints that are TSQL-only are unaffected.

I'm curious - is there any demand for Warehouse / TSQL or are people mostly wanting to use Lakehouse / Spark ?

[mattiasthalen]

IIRC, dbt uses Warehouse for this reason. And you can reference the lakehouse tables as sources (external table)...

So our workflow is that the bronze layer is the lakehouse, where the EL lands. Then we do all our silver/gold (transform) in the warehouse.

That's how I'd want to use SQLMesh as well.

And tbh, while being able to use a lakehouse would be nice, it's not a deal breaker. A deal breaker would be not to be able to use SQLMesh at all 😉

So to answer your question, (pretty) please proceed with supporting the warehouse endpoint! ❤️