The Bit Bucket

OneLake is really the foundation of Fabric’s storage model. The idea is simple but powerful: instead of having separate storage systems for each analytics tool or service, Fabric provides a single, unified data lake. This gives you one logical place where all of your data lives, and all the workloads in Fabric can share it.

Technically, OneLake is built on the open Delta Lake format. Underneath, it uses Parquet files for efficient columnar storage, but it adds transactional support on top. That means multiple processes can read and write to the same data in a consistent way, with guarantees around reliability and performance. It’s open, it’s proven, and it avoids the pitfalls of closed, proprietary formats.

2026-02-28

Recently, I saw a subtle coding issue related to the UNION operator. With SQL Server, the UNION operator combines two rowsets into a single rowset. If UNION ALL is used then all rows are returned. With just UNION without the ALL, only distinct rows are returned. All good so far.

One of the most common performance issues that I come across is where people have just used UNION where they should have used UNION ALL. That extra distinct operation is often not needed, yet often very expensive.

2026-02-27

One of the most powerful aspects of Fabric’s Real-Time Intelligence is how it connects to external sources. The mechanism for doing this is through Eventstreams. Eventstreams are essentially the pipelines that define where your data is coming in — the inputs — and where it’s going out — the outputs. In a later post, we’ll explore Eventstreams.

Fabric comes with a range of native connectors. These include direct connections to industry-standard technologies like Kafka, Azure Event Hubs, Azure IoT Hub, and other AMQP-based systems. That means if you already have investments in streaming infrastructure — whether in the cloud or on-premises — Fabric can plug into them without a lot of custom development.

2026-02-26

Do you want to start working with AI-related features in SQL Server 2025?

Our new course might be just what you need:

https://sqldownunder.com/courses/aif_modules

As these features are relatively new, the focus of this course is on design and architecture decisions that remain valid, even as individual features, syntax, and external AI services evolve. It does not aim to turn SQL Server into an AI platform, nor does it attempt to teach model training, data science, or vendor-specific AI tooling.

2026-02-25

We’ve looked at events generated by applications, databases, and storage systems — but it’s important to remember that Azure and Fabric themselves also generate events. These are sometimes called platform events, because they come from the infrastructure and services you’re running rather than from your business data directly.

Some examples are really practical. Azure might emit events when resources are scaled up or down — say, when a cluster automatically adds nodes to handle increased load. Fabric might generate events when a pipeline starts, completes, or fails. You’ll also see events related to service health, configuration changes, or security alerts.

2026-02-24

When you first make a connection to the new LocalDB edition of SQL Server Express, the system files, etc. that are required for a new version are spun up. (The system files such as the master database files, etc. end up in:

C:\\Users\\<username>\\AppData\\Local\\Microsoft\\Microsoft SQL Server Local DB\\Instances\\LocalDBApp1) 

Now that can take a while on a slower machine, so this means that the default connection timeout of 30 seconds that you’ll find in most client libraries, could well be exceeded.

2026-02-23

Another important category of real-time sources comes from cloud storage platforms like Azure Blob Storage and AWS S3. These aren’t just passive data stores — they can actually generate events whenever something happens to the data inside them.

For example, when a new file is uploaded into a data lake, the storage service can immediately raise an event. That event might then trigger an ingestion pipeline, start a transformation process, or kick off machine learning model scoring. The key here is that you don’t have to wait for a scheduled scan or a batch job to check for new files. The storage system itself notifies you the instant a change occurs.

2026-02-22

There is a hotel in Sydney that we often stayed at several times per year, either for training or mentoring/consulting work. I really like it. If you asked me why though, I would find it hard to list all the reasons why I like it. (There are some things I don’t like about it too but that’s a topic for another day).

But one of the things that had always surprised me, each and every time I ate breakfast there, is how sharp the knives were. It might seem an odd thing to notice but I’ve never noticed this in any other hotel that I’ve ever stayed at.

2026-02-21

This diagram highlights the difference between Change Data Capture (CDC) and Change Event Streams (CES) — two ways of turning databases into real-time event sources.

On the left side, we see CDC. Here the focus is on row-level changes. Every time a row is inserted, updated, or deleted in a table, that operation is captured and streamed out as an event. Those events typically feed into real-time replication or analytics pipelines. In other words, CDC makes sure your downstream systems always have fresh copies of your data. It’s like having a live feed of every transaction at the cash register. That’s why we use the analogy here of cash register receipts — you get a record of each sale exactly as it happens.

2026-02-20

Version 27 of our free SDU Tools for developers and DBAs is now released and winging their way out to our SDU Insiders.

You can find details on the tools here .

If you haven’t been using SDU Tools yet, I’d suggest downloading them and taking a look. At the very least, it can help when you’re trying to work out how to code something in T-SQL.

Along with the normal updates to SQL Server versions and builds, we’ve added the following new tools:

2026-02-19