Series
Azure Data Services: SQL, Cosmos DB, and Analytics
The first real decision in any Azure data design isn't which database to use. It's how much of the machinery underneath you want to be responsible for. Run SQL Server on a virtual machine and high availability, patching, backups, and scaling are all yours to build and keep working. Use Azure SQL Database and they arrive as platform behaviour you configure rather than construct. Almost every other decision — cost model, operational burden, how fast you can respond to a change in load — follows from that one.
This series works through the platform that choice opens up. It starts with classifying data and the managed-versus-unmanaged trade-off, then covers Azure SQL in depth: the three deployment options and what each gives up, the DTU and vCore purchasing models and the serverless tier that changes the billing question entirely, and the service tiers — which differ architecturally, not just on performance, in ways that decide what happens during a failure. It covers Cosmos DB's APIs, request units, and the cost multipliers that come with each step up in availability, and where Table Storage stops being sufficient. It works through caching, the four encryption mechanisms and what each one genuinely protects against, and the three scaling strategies ranked by how much complexity each pushes into application code. It finishes with the analytics stack: Data Factory and its integration runtimes, the data lake, Databricks and Synapse, and the hot, warm, and cold paths that separate real-time alerting from batch reporting.
The emphasis throughout is on the constraints that decide designs rather than the feature lists that describe them: why an elastic pool is the wrong shape when databases need to scale independently, why transparent data encryption doesn't protect data from a database administrator, why read scale-out can serve a stale read straight after a write, and why sharding is the one scaling decision with a one-way door attached.
It also covers several live changes worth knowing if you're building now rather than sitting an exam: Azure Cache for Redis retiring in favour of Azure Managed Redis, Microsoft Fabric emerging as the strategic direction for analytics, and the zone-redundancy and storage-limit figures that most training material still gets wrong.
12 parts
- 1 Azure Data Storage: Managed vs Unmanaged Strategies How to classify data before choosing a service, and what moving from a self-managed database to a managed one actually buys you — high availability, scaling, threat detection, and auto-tuning you'd otherwise build yourself.
- 2 Azure SQL: Database, Managed Instance, or Virtual Machine The three Azure SQL deployment options and what actually separates them — instance-scoped features, OS-level access, and administrative overhead — plus the high availability options each one does and doesn't support.
- 3 Azure SQL Purchasing Models: DTU vs vCore Choosing between the DTU and vCore purchasing models, what the serverless compute tier changes about billing, and why DTU costs the same whether or not anyone uses the database.
- 4🔒 Azure SQL High Availability and Service Tiers How General Purpose, Business Critical, and Hyperscale differ architecturally rather than just on performance — separated versus integrated storage, replica counts, and which tiers support zone redundancy.
- 5 Azure Cosmos DB: APIs, Request Units, and Availability Choosing a Cosmos DB API and capacity mode, how availability scales from single-region to multi-write, and the cost multipliers that come with each step up in redundancy.
- 6 Azure Table Storage vs Cosmos DB for Table Two services share one table data model and the same SDK operations, but differ on latency guarantees, throughput ceilings, indexing, and global distribution — here's which side of that line a workload belongs on.
- 7 Azure Managed Redis and Caching Patterns Azure Cache for Redis is retiring and Azure Managed Redis replaces it — the four new tiers, what changes for a migration, and the caching patterns worth knowing whichever service you're on.
- 8🔒 Azure SQL Encryption: TDE, Always Encrypted, and Masking The four mechanisms protecting data at rest, in transit, and in use — what each one actually covers, why Always Encrypted needs a special client, and where dynamic data masking stops being a security control.
- 9🔒 Scaling Azure SQL: Vertical, Read Scale-Out, and Sharding Three ways to scale a database and how much application code each one demands — vertical scaling is a platform setting, read scale-out and sharding push real complexity into your application.
- 10 Azure Data Factory and Integration Runtimes How Data Factory orchestrates ETL across disparate stores — pipelines, activities, datasets, linked services, and data flows — and which of the three integration runtimes an on-premises source requires.
- 11 Azure Data Lake Storage, Databricks, and Synapse What a data lake is on Azure, how Databricks and Synapse sit on top of it, and where Microsoft Fabric now fits — including the Data Lake versus Blob Storage comparison that no longer holds.
- 12 Hot, Warm, and Cold Data Paths on Azure Splitting incoming data by how fast results are needed, and where Azure Stream Analytics fits — data streams, event processing, and the three formats it accepts.