Microsoft Fabric: A Few Years On

It's been a quick hop since Fabric was unleashed on the world and the familiar, eye-blasting yellow of Power BI began to be mysteriously replaced by the calmer, more distinguished green tones of it's mothership. I say a quick hop. It's been a while now. Blazing into life at almost exactly the same time as Synapse Link for F&O, it wasn't exactly "back burnered", but it would be fair to say, given the number of estates we were managing for F&O, that it was certainly not the immediate priority!

That said, the tinkerer in me can never be supressed, and it wasn't long before I was pulling my hair out, wondering where some of the features I'd come to love about Synapse Analytics were. It badged itself as effectively being the successor (at least in marketing) to Synapse. It badged itself as being the new world. Why wasn't it doing the things I wanted?

We have come a long, long way from that time.

I think Fabric really came of age in late 2024, probably a couple months before the first birthday of its general availability. Having been constantly revisiting the technology before every implementation to determine the maturity, there was a sudden realisation that it was probably worth any extra difficulties in the short term in order to ensure that we weren't working with tech that's ageing out.

I maintain that the migration from Synapse to Fabric is likely to be very, very straight-forward, even for an organisation doing it 'cold'. Notebooks are notebooks, and while there will be a small number of changes needed (thinking particularly about the mssparkutils and sempy libraries in particular), the likelihood is that any well-written notebook is probably ready to run on Fabric with a few tweaks. Further to that, any organisation that's endeavoured to wrap its processes in functions has probably got an even easier job - simply changing the code in one place.

The remainder, whose dependencies are so deep that they need wholesale re-programming, are likely to form a very, very small percentage of the overall uptake.

So how is Fabric better now?

Well, there are a few things I love. Firstly, the 'sempy' library. Short for semantic Python, sempy is a super smooth way to collect environmental information without any fuss. You can wrap it into functions that help a notebook determine where it is and what its dependencies should be, and then just let it work away. It's a genuinely awesome little library to work with and has been a real shining light in the last year of Fabric work.

Next, there's something that's really only relevant to the ex-Synapse developers in the room - but CLUSTER STARTUP IS ACTUALLY WITHIN A REASONABLE TIMEFRAME. The 3/4-minute cluster start wait time in Synapse was a disaster for productivity - especially when paired with auto-shutdown. You'd duck out to attend a call. A long-running notebook would be cracking on. You'd end up coming off the call, grabbing a bite to eat, then coming back to find out that you need to add a filter or sort to your notebook and the cluster has already timed out. Another 4-minute wait for it to restart, followed by the cluster run-time. Nightmare. This is, however, a nightmare no longer - Fabric boasts much more Databricks-like cluster spool up times - often under 10 seconds. It's so satisfying after all that time, not having to wait any more.

The addition of tabs to allow quick navigation between notebooks (more consistent with Synapse and Databricks) is another huge win. The Power BI interface was a new idea, but as the breadth of functionality increases and the number of createable artifacts follows, it becomes almost impossible to keep track in such a small space on the left hand bar. The tabular approach up top is consistent with loads of user experiences nowadays - it's great to see that included.

Finally. Version history. One of my biggest gripes with Fabric as it stands is the mechanisms with which it integrates with Git - however, given the cloud-apps style approach for developers (you can see another developer typing - very like Word Online), I've always considered Git to be only partially necessary - we want the history to allow reversion. Well. That's something that could be solved by version history. This will allow (particularly smaller, closer-knit) teams to develop in the absence of Git - meaning that code changes can be enacted as they are committed, and reverted afterwards in the event of any failure.

I genuinely believe some of these improvements have positioned Fabric as the #1, unquestioned, must-have unified data platform in the world. Other platforms, like Snowflake or Databricks, are excellent in their own right, with a great many benefits (love that slick Databricks interface), but they simply don't "unify" in the same way. Fabric IS medallion architecture now. Fabric IS Power BI. Everybody is headed to the same place, for a range of reasons, and Fabric handles that beautifully.

Vive le Fabrique.