A production SQL query has become slow. How would you read its execution plan, choose a likely fix, and prove that the change helped?

Read an execution plan for a slow SQL query

Uses an actual execution plan, row-flow evidence, runtime measurements, and representative parameters to diagnose a slow query without guessing.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I would identify the exact query and representative parameter values, record duration, CPU, logical reads, and returned row count, then capture the actual execution plan in a safe environment or use an already captured production plan. I would follow the row flow and compare estimated rows with actual rows while checking access paths, joins, sorts, lookups, spills, and warnings. A large estimate mismatch or unexpected row explosion often points to the first useful hypothesis. The plan does not prescribe the fix; it narrows what I should test, such as rewriting a predicate, updating statistics, changing query shape, or adding a carefully chosen index.

Full answer below

Practice focus

Actual versus estimated plans, row estimates, scans and seeks, joins, lookups, spills, statistics, parameter sensitivity, validation, and production-safe tuning.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Why this question matters
Execution-plan questions separate evidence-led database tuning from generic advice to add an index. A useful diagnosis explains how rows flow through the chosen plan, why the optimizer made that choice, and how a proposed change will be validated without changing query results.
I would identify the exact query and representative parameter values, record duration, CPU, logical reads, and returned row count, then capture the actual execution plan in a safe environment or use an already captured production plan. I would follow the row flow and compare estimated rows with actual rows while checking access paths, joins, sorts, lookups, spills, and warnings. A large estimate mismatch or unexpected row explosion often points to the first useful hypothesis. The plan does not prescribe the fix; it narrows what I should test, such as rewriting a predicate, updating statistics, changing query shape, or adding a carefully chosen index.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
I would not simply pick the operator with the largest displayed cost. I would trace where the actual row count first diverges sharply from the estimate and how that affects later operators. For example, applying a function to can turn an indexed range seek into a large scan; a key lookup may be harmless for ten rows but expensive when repeated hundreds of thousands of times; and an underestimated sort or hash join may spill to temp storage because it received too little memory. Each observation leads to a specific experiment, and I would rerun the same representative workload to confirm identical results and compare elapsed time, CPU, and logical reads before accepting the change.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Adding the first suggested index without checking existing indexes or write cost.
Treating estimated cost percentages as measured runtime.
Ignoring the first large estimated-versus-actual row mismatch.
Calling every scan bad even when most of the table is needed.
Measuring only duration once and not verifying result correctness, CPU, or logical reads.

A production SQL query has become slow. How would you read its execution plan, choose a likely fix, and prove that the change helped?

Practice focus

Answer path

Why this question matters

Review signals

Follow-up prompts

Continue the section

Tech stacks

Skill areas

Suggest an improvement