Entity Framework Core Performance-Optimierung aus der Praxis

Transcript

1. Entity Framework Core Performance-Optimierung aus der Praxis Pawel Gerr @pawelgerr pawel.gerr@thinktecture.com
2. Entity Framework Core Performance-Optimierung aus der Praxis Example for today – Products DB Product group Products Prices Sellers Studio
3. Entity Framework Core Performance-Optimierung aus der Praxis My usual options for Entity Framework performance optimization 2nd level cache Hardware upgrade Materialized views Stored procedures Triggers Database Features Server Raw SELECT * FROM … Raw SQL Bulk operations Indices Entity Framework Requests reduction Query restructure … and some other (minor) features we will see later Use 3rd party libs if available CPU, Memory, I/O
4. Entity Framework Core Performance-Optimierung aus der Praxis Did „the change“ anything good? – My KPI Execution time (LINQ) SELECT * FROM … SELECT * FROM … SELECT * FROM … Number of requests Query statistics SELECT * FROM … “Shape” of SQL Execution plans Parse time Compile time Execution time (DB) Physical reads Logical reads Row count Round trip time SQL generation Execution time (DB) Materialization My key performance indicators
5. Entity Framework Core Performance-Optimierung aus der Praxis Reducing database requests
6. Entity Framework Core Performance-Optimierung aus der Praxis N+1 queries problem Execution of hundreds of requests for one use case Caused by: unintentional execution of queries in loops Possible solutions: • Better code structure (software architecture) • Avoid overgeneralization (shared/core projects) • Help each other to improve (review/feedback) • Know the insights of Entity Framework • Lazy loading • Limitations Reducing database requests Demos 1.1 + 1.2
7. Simplified case of indirect execution of database requests in a loop Library Main Entity Framework Core Performance-Optimierung aus der Praxis Multiple requests in loops Reducing database requests var products = LoadProducts(); var prices = new List<Price>(); foreach (var product in products) { var price = GetPrice(product.Id); prices.Add(price); } public Price GetPrice(Guid productId) { MyDbContext ctx = …; return ctx.Prices.FirstOrDefault(p => p.ProductId == productId); } Executed N times
8. Library Main Entity Framework Core Performance-Optimierung aus der Praxis LINQ queries are loops Reducing database requests var products = LoadProducts(); var prices = products.Select(product => GetPrice(product.Id)) .ToList(); public Price GetPrice(Guid productId) { MyDbContext ctx = …; return ctx.Prices.FirstOrDefault(p => p.ProductId == productId); }
9. Library Main Entity Framework Core Performance-Optimierung aus der Praxis Approach 1 – Use-case specific methods Reducing database requests var productsWithPrices = LoadProductsWithPrices(); public List<ProductWithPrices> LoadProductsWithPrices() { return ctx.Products .Include(p => p.Prices) // merely symbolic, fetching prices is usually more complex .ToList(); } Approach 1.1
10. Library Main var products = LoadProducts(); var productIds = products.Select(p => p.Id); var prices = GetPrices(productIds); public List<Price> GetPrices(IEnumerable<Guid> productIds) { return ctx.Prices .Where(p => productIds.Contains(p.ProductId)) .ToList(); } Entity Framework Core Performance-Optimierung aus der Praxis Approach 2 – load all required prices at once Reducing database requests Approach 1.2
11. Given: 100 products being in 5 studio Loading of all products “in one go” leads to 1 + 5 queries Library Main var productsByStudio = ctx.Products .Include(p => p.Studio) .ToLookup(p => p.Studio.Name); Entity Framework Core Performance-Optimierung aus der Praxis Lazy loading Reducing database requests public static ILookup<TKey, TSource> ToLookup<TSource, TKey>( this IEnumerable<TSource> source, Func<TSource, TKey> keySelector) { ... } Use eager loading to prevent unnecessary requests IEnumerable<T> always forces query execution Demo 1.3
12. LINQ Entity Framework Core Performance-Optimierung aus der Praxis Limitations of Entity Framework Core Access to navigational property in EF 2.2 leads to N+1 queries if query is not trivial • Loading of the first product of 5 studios leads to 1 + 5 queries • The issue has been fixed in EF 3.0 Reducing database requests var studios = ctx.Studios .Select(s => { Studio = s, FirstProduct = s.Products.FirstOrDefault() }) .ToList(); not trivial query
13. Entity Framework Core Performance-Optimierung aus der Praxis Finding the problem area Use profiling tools In case of Microsoft SQL Server: XEvent Profiler in SQL Server Management Studio SQL Server Profiler (deprecated, resource intensive) Azure Data Studio Logs coming from EF with EnableSensitiveDataLogging Reducing database requests Logging DbContextOptionsBuilder builder = ...; builder.UseSqlServer("...") // or any other database .UseLoggerFactory(loggerFactory) .EnableSensitiveDataLogging(); In development only because possible security leak!
14. Entity Framework Core Performance-Optimierung aus der Praxis Finding the problem area Look for repetitive lines and cycles Reducing database requests Repetive lines Cycles Demo: SQL Server Profiler
15. Entity Framework Core Performance-Optimierung aus der Praxis Reducing query complexity
16. Entity Framework Core Performance-Optimierung aus der Praxis Cartesian explosion problem Fetching a lot of data from multiple tables at once • Increased memory consumption and execution time • High I/O load Caused by: loading of multiple navigational collection-properties Possible solutions: • Reduce Includes (eager loading) • Reduce access to navigational properties in projections (i.e. Select) Reducing query complexity Query splitting Demo 2.1
17. Entity Framework Core Performance-Optimierung aus der Praxis Eager loading via Include EF 2.2 is loading collections separately EF 3.0 loads all data using 1 SQL statement • Bigger result sets: 100 products with 10 prices each and 2 sellers (100 * 10 * 2) rows > (100 + 1000 + 2) records • EF-forced ORDER BY clause produces considerable load Reducing query complexity Product Id Price Id Seller Id 1 1 1 1 1 2 1 1 3 1 2 1 1 2 2 1 2 3 SQL LINQ var products = ctx.Products .Include(p => p.Studio) .Include(p => p.Prices) .Include(p => p.Sellers) .ToList(); SELECT * FROM Products INNER JOIN Studios ... LEFT JOIN Prices ON ... LEFT JOIN Sellers ON ... ... ORDER BY Products.Id, Prices.Id, Sellers.Id Cartesian explosion
18. Before Entity Framework Core Performance-Optimierung aus der Praxis Using collections in projections Loading all data at once Statistics: • CPU: 31 • Duration: 75 • Reads: 5299 • Row Count: 12342 Reducing query complexity LINQ var studios = ctx.Studios .Select(s => new MyStudio { Id = s.Id, Name = s.Name, Infinity = s.Products.Where(p => p.Name.StartsWith("Infinity")), Endgame = s.Products.Where(p => p.Name.StartsWith("Endgame")) }) .ToList();
19. Before Usually, loading collection separately requires more code but less resources on database Statistics: • CPU: 31 • Duration: 75 • Reads: 5299 • Row Count: 12342 After LINQ Entity Framework Core Performance-Optimierung aus der Praxis Query splitting Reducing query complexity var studios = ctx.Studios.Select(p => new Studio() { Id = p.Id, Name = p.Name }).ToList(); var infinity = ctx.Products.Where(p => p.Name.StartsWith("Infinity")).ToList(); var endgame = ctx.Products.Where(p => p.Name.StartsWith("Endgame")).ToList(); var infinityLookup = infinity.ToLookup(p => p.StudioId); var endgameLookup = endgame.ToLookup(p => p.StudioId); foreach (var studio in studios) { studio.Infinity = infinityLookup[studio.Id]; sturio.Endgame = endgameLooup[studio.Id]; } Necessary, if change tracking is disabled 16 3 352 2322
20. Entity Framework Core Performance-Optimierung aus der Praxis Understanding queries
21. LINQ 2 LINQ 1 Entity Framework Core Performance-Optimierung aus der Praxis Which one is better? Understanding Queries var groups = ctx.ProductGroups .Select(g => new { g.Products.FirstOrDefault().Id, g.Products.FirstOrDefault().Name }) .ToList(); var groups = ctx.ProductGroups .Select(g => g.Products .Select(p => new { p.Id, p.Name }) .FirstOrDefault()) .ToList(); 2x “FirstOrDefault()” before selecting properties Selection of properties before “FirstOrDefault()”
22. SQL 2 SQL 1 Entity Framework Core Performance-Optimierung aus der Praxis Lets try with SQL … Understanding Queries SELECT ( SELECT TOP(1) p.Id FROM Products p WHERE g.Id = p.ProductGroupId ) AS FirstProductId, ( SELECT TOP(1) p.Name FROM Products p WHERE g.Id = p.ProductGroupId ) AS FirstProductName FROM ProductGroups g SELECT p.Id, p.Name FROM ProductGroups g LEFT JOIN ( SELECT Id, Name, ProductGroupId FROM ( SELECT Id, Name, ProductGroupId, ROW_NUMBER() OVER(PARTITION BY ProductGroupId ORDER BY Id) AS row FROM Products ) p WHERE row <= 1 ) p ON g.Id = p. ProductGroupId 2 sub-selects Window function “ROW_NUMBER”
23. Entity Framework Core Performance-Optimierung aus der Praxis Execution plans
24. Entity Framework Core Performance-Optimierung aus der Praxis Textual representation of all database operations • Type and the order of operations (JOIN, filter, projection, ...) • Indexes being used • Amount of data flowing between two operations • Costs of an operation and a subtree Some tools have built-in support for displaying execution plans as a graph Execution plans
25. Entity Framework Core Performance-Optimierung aus der Praxis Another try … Execution plans Execution plan 1 SQL 1 SELECT ( SELECT TOP(1) p.Id FROM Products p WHERE g.Id = p.GroupId ) AS FirstProductId, ( SELECT TOP(1) p.Name FROM Products p WHERE g.Id = p.GroupId ) AS FirstProductName FROM ProductGroups g
26. Entity Framework Core Performance-Optimierung aus der Praxis Execution plans Execution plan 2 SQL 2 SELECT p.Id, p.Name FROM ProductGroups g LEFT JOIN ( SELECT Id, Name, GroupId FROM ( SELECT Id, Name, GroupId, ROW_NUMBER() OVER(PARTITION BY GroupId ORDER BY Id) AS row FROM Products ) p WHERE row <= 1 ) p ON g.Id = p.GroupId
27. Entity Framework Core Performance-Optimierung aus der Praxis Comparing • Compare estimated subtree costs • Deviance: estimated vs actual number of rows • Take statistics (like reads) into consideration Execution plans Query 1 wins! Demo 3.1
28. Entity Framework Core Performance-Optimierung aus der Praxis Execution plans Crash course Table scan Clustered index scan Non-clustered index scan Clustered index seek Non-clustered index seek Scan No clustered index Seek Filter Filtering Filtering Filtering Key lookup Missing (include) columns? RID lookup If no clustered index Sort “Stop & Go” operator! Required? Missing index? Sorting in .NET cheaper? Parallelism Expensive High query complexity Missing index? Fuzzy search? Bad discriminator?
29. Entity Framework Core Performance-Optimierung aus der Praxis JOINs Execution plans DB may perform sort on its own • Resource-saving • Data sets must be ordered • The “default” • Kind-of 2 nested loops • Speed depends on: • Scan of data set A • Seek of data set B • Large and unsorted sets • 2 phases: • Builds hash table for set A • Matches hash values from B Should be looked at Merge join Nested loop join Hash match join
30. Entity Framework Core Performance-Optimierung aus der Praxis Learnings • Look for database requests in “loops” • Split complex queries if necessary • Learn to read execution plans - optimize queries accordingly Resources • Entity Framework Core: https://docs.microsoft.com/en-us/ef/core/ • Execution Plans: https://docs.microsoft.com/en-us/sql/relational-databases/performance/execution-plans • Webinar demos: https://github.com/thinktecture/ef-core-performance-webinar-2020 • My EF playground: https://github.com/PawelGerr/EntityFrameworkCore-Demos Ask me @pawelgerr pawel.gerr@thinktecture.com