It is the middle of 2026, and with the current crop of technologist-oriented AI tools that are available, can’t you just submit a question like “analyze the following Trino query plan for actionable insights to improve performance and scalability …”? Absolutely, and while this is huge, it isn’t enough by itself.
The information returned to you assumes you understand how a modern MPP engine like Trino or Starburst actually works. The findings and recommendations will be expressed in technical jargon, including fragments, partitions, hashing, broadcasting, statistics, splits, stages, and exchanges. Understanding how a multi-stage job executes across a distributed compute engine is still required to validate the responses you received and decide which, if any, of the suggestions need to be actioned.
Build the knowledge
Starburst has you covered with on-demand videos and hands-on exercises to learn more about query plan analysis. Let’s start with our three-part video series.
Parallel processing fundamentals
The 1st of 3 videos starts out with the following high-level concepts.
- Understand how and why datasets are divided into splits
- Explain the decomposition of a query into tasks, stages, and exchanges
- Visualize multi-stage queries that tackle sorting, aggregation, and multiple types of joins
- See the conceptual visualization of how a parallelized query aligns with Trino’s generated visual plan at runtime
Query plan syntax
The middle video is focused on the particular syntax Trino’s cost-based optimizer (CBO) uses in the textual query plan. This video hits on topics such as the following.
- Optimizations available with table statistics
- Generating a textual and a visual query plan
- Understand the format of the query plan output to include fragments, exchanges, distribution, estimates, and performance
- Performance with, and without, accurate table statistics
Detailed plan analysis
The final video walks through query plans, leveraging background knowledge from the earlier videos, with the mindset of identifying potential performance issues and their remedies. You’ll see firsthand topics like those listed below.
- Column pruning, predicate pushdown, and partition pruning
- Trino’s approach to tackle sorting and aggregation across two stages
- Join optimizations, including broadcast, reordering, and dynamic filtering
Go hands-on
Starburst still has you covered with two more great assets. First, the Workshop: Query Plan Analysis Deep‑Dive with Starburst is available on demand and includes hands-on exercises.
Additionally, Exploring query plans is an available course from Starburst Academy that includes examples of federated queries joining data lake and relational database catalogs.
Additional help
In addition to their account team, Starburst customers have access to world-class support and professional services. Everyone has access to the Trino community chat and our Starburst forum – Questions and answers for the Starburst user community to ask questions and to find additional content, such as this post on how to see I/O wait time from your query plan output.
AI tooling will only continue to advance the goals of query plan analysis, improved performance, and scalability, but fully understanding the recommendations it surfaces will remain most beneficial when you understand the fundamentals of your compute engine and how it works with your data sources.
