Snowflake Interview Q&A: Warehouse Sizing, Concurrency and Query Diagnosis
Prepare Snowflake warehouse interview answers with fair experiment design and a sample aggregation.
1. Would you resize the warehouse first when a query is slow?
First establish what is slow. Compare the same representative query, inspect execution versus waiting, review the amount of data read, and note other simultaneous work. A one-off elapsed time cannot identify a compute bottleneck by itself. Record the warehouse configuration and workload so a later comparison is meaningful.
An interviewer is looking for a diagnosis process, not a promise that bigger compute always fixes performance. A changed query, warm cache or a quieter workload can make a comparison misleading.
2. How does concurrency differ from a heavy single query?
Concurrency concerns multiple queries competing for resources. A heavy query concerns the work of one query. More warehouse capacity and additional clusters address different workload situations; availability of features depends on the account. Do not recommend a configuration without checking requirements and supported options.
3. Can you demonstrate aggregation without creating a table?
Use inline sample rows. This Snowflake SQL deliberately avoids changing stored objects and uses invented product quantities.
WITH lines AS (
SELECT column1::VARCHAR AS product,
column2::NUMBER(10,0) AS quantity
FROM VALUES ('notebook',2), ('pen',5), ('notebook',3)
)
SELECT product, SUM(quantity) AS units
FROM lines GROUP BY product ORDER BY product;Expected results are notebook 5 and pen 5. This is a correctness exercise, not a warehouse performance benchmark: three inline rows do not represent a large production workload. The SQL was reviewed against the Snowflake reference and not executed in a live Snowflake account. Running it requires a suitable authorized session and may use account resources.
4. How would you design a fair sizing experiment?
Choose a realistic query or workload, fix its inputs, record the settings, and decide beforehand which measurements count. Repeat enough to notice variability and distinguish queueing from execution. Include resource consumption and business latency, not just the fastest observed run. Account for cache effects and avoid presenting a synthetic tiny query as proof of production improvement.
5. What should you check about suspension and resumption?
Confirm actual auto-suspend and auto-resume settings, the activity pattern and the workload's latency requirements. Aggressive suspension can introduce repeated startup overhead; keeping a warehouse active when unused can consume resources unnecessarily. The appropriate setting follows the workload rather than a universal value copied from an interview answer. Check current account billing rules before calculating costs.
Practice interview follow-up
Add a pen row of 4: expect pen 9 and notebook 5. Now imagine the real report scans millions of rows and ten users start reports together. Write three observations you would collect before changing capacity. A strong answer might distinguish input volume, waiting time and simultaneous workload, then propose a controlled comparison instead of immediately upgrading.
Prepared by AI Leaf Global Learn for learners in Mudichur/Tambaram, Chennai, and online across India. Confirm current course and batch details. This guide makes no salary, placement, certification or universal price claim.
Snowflake course.
Official warehouse considerations.
Snowflake VALUES reference.