Here are some interview questions related to Azure SQL Database query tuning and system performance tuning, along with sample answers:

1. What are common causes of poor performance in Azure SQL Database?

Answer:

• Inefficient Queries: Poorly written SQL queries, lack of proper indexing, or too many joins.
• Lack of Indexes: Missing indexes or unused indexes that increase query cost.
• High Resource Utilization: Excessive CPU, memory, or IO usage.
• Parameter Sniffing: Cached execution plans that don’t perform well for all parameters.
• Concurrency Issues: Blocking, deadlocks, and contention for resources.
• Suboptimal Query Plans: Poor execution plans due to outdated statistics or query structure.

2. How do you identify slow-running queries in Azure SQL Database?

Answer:

• Use Query Performance Insights in the Azure portal to find high-CPU or long-running queries.
• Query the sys.dm_exec_query_stats DMV to review query execution statistics.
• Check sys.dm_db_index_operational_stats for index usage and performance issues.
• Enable Query Store to capture query execution details and identify regressions.

3. What steps would you take to optimize a query in Azure SQL Database?

Answer:

1. Analyze Execution Plan:
   • Use the “Actual Execution Plan” to identify costly operations like table scans or sorts.
2. Index Optimization:
   • Create missing indexes or modify existing ones.
   • Use sys.dm_db_missing_index_details to identify missing indexes.
3. Rewrite Query:
   • Simplify joins, subqueries, or complex logic.
   • Replace SELECT * with specific columns.
4. Parameterization:
   • Use parameterized queries to leverage plan reuse and avoid parameter sniffing.
5. Statistics Update:
   • Run UPDATE STATISTICS to ensure query optimizer has up-to-date information.
6. Use Hints:
   • Use query hints (e.g., OPTION (RECOMPILE) or FORCE INDEX) judiciously.
7. Partitioning:
   • Partition large tables to improve performance of data reads and writes.

4. What tools or features are available for performance tuning in Azure SQL Database?

Answer:

• Query Performance Insight: Visualize query performance metrics and identify slow queries.
• Query Store: Monitor query execution plans and identify regressions.
• Dynamic Management Views (DMVs):
  • sys.dm_exec_query_stats for query performance.
  • sys.dm_db_index_usage_stats for index usage.
• Intelligent Insights: Detect and diagnose performance issues automatically.
• Azure SQL Database Advisor:
  • Provides recommendations for index creation or removal.
• Automatic Tuning:
  • Features like “Force Plan” or “Create Index” can optimize queries automatically.

5. How do you handle parameter sniffing issues in Azure SQL Database?

Answer:

• Use OPTION (RECOMPILE) to generate a new execution plan for each query execution.
• Use OPTIMIZE FOR UNKNOWN to create a generic plan.
• Implement query hints to guide the query optimizer.
• Rewrite the query to use local variables, which prevent parameter sniffing.

6. How do you monitor and tune system-level performance in Azure SQL Database?

Answer:

• Monitor CPU/IO Utilization:
  • Use Azure Monitor and set alerts for high resource usage.
• Scaling Resources:
  • Scale up/down compute tier (e.g., DTU or vCore models).
• Elastic Pools:
  • Use elastic pools to manage resource usage across multiple databases.
• Throttling:
  • Avoid throttling by optimizing queries and scaling appropriately.
• Resource Governance:
  • Use Resource Governor to allocate resources effectively for critical workloads.

7. What are DTUs and vCores in Azure SQL Database? How do they impact performance?

Answer:

• DTUs (Database Transaction Units):
  • A blended measure of compute, memory, and IO resources.
  • Higher DTU levels provide better performance for resource-intensive workloads.
• vCores (Virtual Cores):
  • Provides dedicated CPU cores, memory, and storage, offering better customization and transparency.
• Impact:
  • DTU model is simpler to manage but less flexible.
  • vCore model is suitable for workloads requiring predictable performance.

8. How do you resolve blocking and deadlocks in Azure SQL Database?

Answer:

• Blocking:
  • Identify blocking queries using sys.dm_tran_locks.
  • Optimize queries to minimize lock contention.
  • Use NOLOCK or READ COMMITTED SNAPSHOT isolation levels where appropriate.
• Deadlocks:
  • Enable Deadlock Graph in SQL Server Profiler or Azure Monitor logs.
  • Identify the root cause and refactor the queries.
  • Minimize lock time and avoid circular dependencies in queries.

9. What is the Query Store, and how do you use it for performance tuning?

Answer:

• Query Store is a feature in Azure SQL Database that captures query performance and execution plans.
• It helps:
  • Identify slow or regressed queries.
  • Compare historical query performance.
  • Force stable query execution plans for consistent performance.
• Enable it using: ALTER DATABASE CURRENT SET QUERY_STORE = ON; Query insights can be accessed in the Azure portal or via DMVs like sys.query_store_plan.

10. How would you identify and fix missing or unused indexes in Azure SQL Database?

Answer:

• Identify Missing Indexes: SELECT * FROM sys.dm_db_missing_index_details;
• Identify Unused Indexes: SELECT * FROM sys.dm_db_index_usage_stats WHERE user_seeks = 0 AND user_scans = 0;
• Fix:
  • Create missing indexes as suggested by the DMVs.
  • Drop unused indexes that consume storage and degrade write performance.

11. How do you handle high CPU utilization in Azure SQL Database?

Answer:

• Identify high CPU queries using Query Store or sys.dm_exec_query_stats.
• Optimize queries (e.g., indexes, joins, WHERE clauses).
• Update statistics or rebuild fragmented indexes.
• Scale up the database to a higher tier or vCore level if necessary.

12. How do you monitor disk IO performance in Azure SQL Database?

Answer:

• Use Azure Monitor to track IO-related metrics like read/write latency and throughput.
• Query DMVs like: SELECT * FROM sys.dm_io_virtual_file_stats(NULL, NULL);
• Optimize by:
  • Avoiding large table scans.
  • Partitioning large tables or using clustered indexes.

Mohammad Gufran Jahangir