Starburst sells a commercial distribution of Trino, the distributed SQL query engine that grew out of Presto at Facebook, and its academy is consequently in the pleasant position of teaching something genuinely valuable that exists independently of the product being sold. A large proportion of what you learn here applies to any Trino deployment, and Trino is used widely enough that this is real, portable knowledge rather than vendor-specific configuration trivia. The core value is understanding federated query, which is the idea that you can run a single SQL statement across data sitting in object storage, a relational database, a data warehouse and a streaming system without physically moving any of it first. This sounds like magic when you first encounter it and the training does a good job of explaining why it is not.
The engine pushes down what it can, and predicate pushdown is the difference between a query that returns in seconds and a query that drags several hundred gigabytes across a network before filtering it. Connectors differ enormously in what they support. Join ordering across heterogeneous sources is genuinely difficult and the optimiser needs statistics that frequently do not exist. The training covers these mechanics properly rather than selling the capability and leaving the performance cliffs for you to discover in production, which is a real credit given that the marketing version of federated query is considerably more cheerful.
The lakehouse architecture content is also solid. Table formats like Iceberg, why schema evolution and time travel matter operationally, how partitioning decisions made early determine query performance forever afterwards, and the separation of storage from compute that makes the whole model work economically. This is current, useful architecture knowledge and it is not Starburst-specific. The certification pricing deserves specific praise because it stands out sharply against everything else I have reviewed lately.
The Starburst Certified Practitioner exam costs forty-nine dollars per attempt, is forty questions, and is valid for three years, with the preparatory training free and explicitly optional. Compare that to three hundred and fifty dollars for a Zendesk admin certification or several hundred for various analytics credentials and the difference in posture is obvious. Forty-nine dollars covers the cost of running an exam and deters casual attempts without turning the credential into a profit centre. I would like considerably more vendors to price this way, and I mention it in every review where it applies because it is the clearest signal of whether a vendor education team is measured on learning outcomes or on revenue.
The honest weakness is the line between open source and commercial. Some of what the training teaches is pure Trino and works anywhere. Some requires Starburst's commercial additions, the management layer, the security and governance features, the additional connectors and the performance work that sits on top. The courses do not always mark which is which, and a learner could reasonably finish a module and be unclear about whether the capability just demonstrated is available in the free engine.
Given how much credibility the academy earns from teaching open source fundamentals, being sloppy about this boundary is a small self-inflicted wound. The prerequisites are real. You need SQL well beyond the basics, since the interesting material is about query planning and you cannot reason about a plan you cannot read, and you need enough distributed systems intuition that words like shuffle and partition mean something concrete. Data analysts who write SQL but have never thought about how it executes will find the middle of the course steep.
Four. Honest pricing, portable knowledge, narrow but real audience.