AGIBuild/dotnet-linq-optimization

dotnet-linq-optimization

LINQ performance patterns for .NET applications. Covers the critical distinction between IQueryable<T> server-side evaluation and IEnumerable<T> client-side materialization, compiled queries for EF Core hot paths, deferred execution pitfalls, LINQ-to-Objects allocation patterns and when to drop to manual loops, and Span-based alternatives for zero-allocation processing.

Out of scope: EF Core DbContext lifecycle, migrations, interceptors, and connection resiliency -- see [skill:dotnet-efcore-patterns]. Strategic data architecture (repository patterns, read/write split, N+1 governance) -- see [skill:dotnet-efcore-architecture]. Span<T> and Memory<T> fundamentals -- see [skill:dotnet-performance-patterns]. Microbenchmarking setup -- see [skill:dotnet-benchmarkdotnet].

Cross-references: [skill:dotnet-efcore-patterns] for compiled queries in EF Core context and DbContext usage, [skill:dotnet-performance-patterns] for Span<T>/Memory<T> foundations and ArrayPool patterns, [skill:dotnet-benchmarkdotnet] for measuring LINQ optimization impact.

---

IQueryable vs IEnumerable Materialization

The most impactful LINQ performance decision is where evaluation happens: on the database server (IQueryable<T>) or in application memory (IEnumerable<T>).

The Problem

// DANGEROUS: Materializes entire table into memory, then filters in C#
IEnumerable<Order> orders = dbContext.Orders;
var recent = orders.Where(o => o.CreatedAt > cutoff).ToList();
// SQL: SELECT * FROM Orders  (no WHERE clause!)

// CORRECT: Filter executes on the database server
IQueryable<Order> orders = dbContext.Orders;
var recent = orders.Where(o => o.CreatedAt > cutoff).ToList();
// SQL: SELECT ... FROM Orders WHERE CreatedAt > @cutoff

When Materialization Happens

| Operation | Effect | |-----------|--------| | ToList(), ToArray(), ToDictionary() | Executes query, loads results into memory | | foreach / await foreach | Executes query, streams results | | AsEnumerable() | Switches from server to client evaluation | | Count(), Any(), First(), Single() | Executes query, returns scalar | | Where(), Select(), OrderBy() on IQueryable | Builds expression tree (no execution) | | Where(), Select(), OrderBy() on IEnumerable | Deferred in-memory evaluation |

Common Mistakes

// MISTAKE 1: AsEnumerable() before filtering
var results = dbContext.Orders
    .AsEnumerable()           // <-- switches to client evaluation
    .Where(o => o.Total > 100)  // runs in memory, not SQL
    .ToList();

// MISTAKE 2: Calling a C# method in IQueryable predicate
var results = dbContext.Orders
    .Where(o => IsHighValue(o))  // Cannot translate to SQL; throws or falls back
    .ToList();

// FIX: Use expression-compatible predicates or call after materialization
var results = dbContext.Orders
    .Where(o => o.Total > 100)   // SQL-translatable
    .AsEnumerable()
    .Where(o => IsHighValue(o))  // C# logic after materialization
    .ToList();

// MISTAKE 3: Projecting too many columns
var names = dbContext.Orders.ToList().Select(o => o.CustomerName);
// Loads ALL columns, then picks one in memory

// FIX: Project before materializing
var names = dbContext.Orders.Select(o => o.CustomerName).ToList();
// SQL: SELECT CustomerName FROM Orders

Detection Checklist

• Any AsEnumerable() or cast to IEnumerable<T> before Where/Select is a potential server-bypass
• EF Core logs Microsoft.EntityFrameworkCore.Query at Warning level when it falls back to client evaluation
• Enable ConfigureWarnings(w => w.Throw(RelationalEventId.MultipleCollectionIncludeWarning)) during development

---

Compiled Queries for EF Core Hot Paths

Compiled queries eliminate the per-call expression tree compilation overhead. For queries executed thousands of times per second, this can reduce overhead significantly.

Standard Compiled Query

public sealed class OrderRepository(AppDbContext db)
{
    // Compiled once, reused across all calls
    private static readonly Func<AppDbContext, Guid, Task<Order?>>
        s_findById = EF.CompileAsyncQuery(
            (AppDbContext ctx, Guid id) =>
                ctx.Orders.FirstOrDefault(o => o.Id == id));

    private static readonly Func<AppDbContext, DateTime, IAsyncEnumerable<Order>>
        s_findRecent = EF.CompileAsyncQuery(
            (AppDbContext ctx, DateTime cutoff) =>
                ctx.Orders
                    .Where(o => o.CreatedAt > cutoff)
                    .OrderByDescending(o => o.CreatedAt));

    public Task<Order?> FindByIdAsync(Guid id) =>
        s_findById(db, id);

    public IAsyncEnumerable<Order> FindRecentAsync(DateTime cutoff) =>
        s_findRecent(db, cutoff);
}

When to Use Compiled Queries

| Scenario | Use compiled query? | |----------|-------------------| | High-frequency lookups (auth, caching) | Yes | | Admin dashboard queries (low frequency) | No -- overhead is negligible | | Queries with dynamic predicates (user search) | No -- cannot parameterize shape | | Queries with Include() that varies | No -- includes change expression tree shape |

Limitations

• Compiled queries cannot use dynamic Include() or conditional Where() clauses that change the expression tree shape
• Parameters must be simple types (no complex objects or collections)
• EF.CompileAsyncQuery returns Task<T> for single results or IAsyncEnumerable<T> for collections

---

Deferred Execution Pitfalls

LINQ uses deferred execution: query operators build a pipeline that executes only when results are consumed. This is powerful but creates subtle bugs.

Multiple Enumeration

// BUG: Enumerates the database query twice
IQueryable<Order> query = dbContext.Orders.Where(o => o.Status == Status.Active);

var count = query.Count();         // Executes SQL (1st query)
var items = query.ToList();        // Executes SQL again (2nd query)

// FIX: Materialize once
var items = dbContext.Orders
    .Where(o => o.Status == Status.Active)
    .ToList();

var count = items.Count;           // In-memory, no SQL

Closure Capture in Loops

// BUG: All queries capture the same loop variable 'i' by reference
var queries = new List<IQueryable<Order>>();
for (int i = 0; i < statuses.Length; i++)
{
    queries.Add(dbContext.Orders.Where(o => o.Status == statuses[i]));
    // 'i' is captured by reference -- all queries use final value of i
}

// FIX: Copy to a local variable inside the loop body
for (int i = 0; i < statuses.Length; i++)
{
    var localStatus = statuses[i];
    queries.Add(dbContext.Orders.Where(o => o.Status == localStatus));
}

Note: C# 5+ foreach loop variables are scoped per iteration and do not exhibit this bug. The for loop index variable is shared across iterations, making this a common pitfall when building deferred LINQ queries in a loop.

Deferred Execution in Method Returns

// DANGEROUS: Returns an unevaluated query -- caller may not realize
// the DbContext could be disposed before enumeration
public IEnumerable<Order> GetActiveOrders()
{
    return dbContext.Orders.Where(o => o.Status == Status.Active);
    // Not evaluated yet -- DbContext may be disposed when caller iterates
}

// SAFE: Materialize before returning
public async Task<List<Order>> GetActiveOrdersAsync(CancellationToken ct)
{
    return await dbContext.Orders
        .Where(o => o.Status == Status.Active)
        .ToListAsync(ct);
}

---

LINQ-to-Objects Allocation Patterns

LINQ operators on in-memory collections allocate iterators, delegates, and intermediate collections. For hot paths processing thousands of items per second, these allocations can cause GC pressure.

Allocation Sources

| Operation | Allocations | |-----------|------------| | Where(), Select() | Iterator object + delegate | | ToList(), ToArray() | New collection + possible resizing | | OrderBy() | Full copy for sorting | | GroupBy() | Dictionary + grouping objects | | SelectMany() | Iterator + inner iterators | | Lambda capture of local variable | Closure object per captured scope |

When LINQ Allocation Matters

LINQ allocations are negligible for most code. Optimize only when:

• Processing is on a hot path (called thousands of times per second)
• BenchmarkDotNet shows significant Allocated bytes
• GC metrics (Gen0 collections/sec) indicate pressure

Manual Loop Alternatives

// LINQ: Allocates iterator + delegate + List<T>
var result = items
    .Where(x => x.IsActive)
    .Select(x => x.Name)
    .ToList();

// Manual loop: Single List<T> allocation, no iterator/delegate overhead
var result = new List<string>(items.Count);
foreach (var item in items)
{
    if (item.IsActive)
    {
        result.Add(item.Name);
    }
}

// LINQ: Allocates iterator + delegate + bool boxing (Any)
var hasActive = items.Any(x => x.IsActive);

// Manual loop: Zero allocations beyond the enumerator
var hasActive = false;
foreach (var item in items)
{
    if (item.IsActive)
    {
        hasActive = true;
        break;
    }
}

Reducing Allocations Without Abandoning LINQ

Before dropping to manual loops, consider these intermediate steps:

// 1. Use Array.Find / Array.Exists for arrays (no iterator allocation)
var first = Array.Find(items, x => x.IsActive);
var exists = Array.Exists(items, x => x.IsActive);

// 2. Pre-size collections when count is known
var result = new List<string>(items.Length);
result.AddRange(items.Where(x => x.IsActive).Select(x => x.Name));

// 3. Use static lambdas to avoid delegate allocation (C# 9+)
var result = items.Where(static x => x.IsActive).ToList();
// Note: static lambdas prevent accidental closure capture
// but the delegate is already cached by the compiler for
// non-capturing lambdas; the main benefit is enforcement

---

Span-Based Alternatives for Collection Processing

For the highest-performance scenarios, Span<T> and ReadOnlySpan<T> enable stack-based, zero-allocation processing. These APIs are not LINQ-compatible but cover common patterns.

Span Search and Filter

// Zero-allocation contains check on an array
ReadOnlySpan<int> values = stackalloc int[] { 1, 2, 3, 4, 5 };
bool found = values.Contains(3);

// Zero-allocation index search
int index = values.IndexOf(4);

MemoryExtensions for String Processing

// Zero-allocation split and iterate
ReadOnlySpan<char> csv = "alice,bob,charlie";
foreach (var segment in csv.Split(','))
{
    ReadOnlySpan<char> value = csv[segment];
    // Process each value without allocating strings
}

// Zero-allocation trim and compare
ReadOnlySpan<char> input = "  hello  ";
bool match = input.Trim().SequenceEqual("hello");

When to Use Span Over LINQ

| Scenario | Approach | |----------|----------| | Parsing CSV/log lines in a tight loop | ReadOnlySpan<char> + Split | | Searching sorted arrays | Span<T>.BinarySearch | | Processing byte buffers from I/O | ReadOnlySpan<byte> slicing | | General business logic on collections | LINQ (readability over micro-optimization) |

See [skill:dotnet-performance-patterns] for comprehensive Span<T>/Memory<T> patterns and ArrayPool<T> usage.

---

Query Optimization Patterns

Projection Before Materialization

Always select only the columns you need:

// BAD: Loads entire entity graph
var orders = await dbContext.Orders
    .Include(o => o.Lines)
    .Include(o => o.Customer)
    .ToListAsync(ct);

var summaries = orders.Select(o => new
{
    o.Id,
    o.Customer.Name,
    Total = o.Lines.Sum(l => l.Price * l.Quantity)
});

// GOOD: Project in the query -- single SQL with computed columns
var summaries = await dbContext.Orders
    .Select(o => new
    {
        o.Id,
        CustomerName = o.Customer.Name,
        Total = o.Lines.Sum(l => l.Price * l.Quantity)
    })
    .ToListAsync(ct);

Pagination with Keyset (Seek) Method

// Offset pagination: O(N) -- server must skip rows
var page = await dbContext.Orders
    .OrderBy(o => o.Id)
    .Skip(pageSize * pageNumber)
    .Take(pageSize)
    .ToListAsync(ct);

// Keyset pagination: O(1) -- index seek
var page = await dbContext.Orders
    .Where(o => o.Id > lastSeenId)
    .OrderBy(o => o.Id)
    .Take(pageSize)
    .ToListAsync(ct);

Batch Operations

// BAD: N UPDATE statements (one per tracked entity change)
foreach (var order in orders)
{
    order.Status = OrderStatus.Archived;
}
await dbContext.SaveChangesAsync(ct);
// Generates N individual UPDATE statements in a single round-trip

// GOOD: EF Core 7+ ExecuteUpdateAsync (single SQL statement)
await dbContext.Orders
    .Where(o => o.CreatedAt < cutoff)
    .ExecuteUpdateAsync(
        s => s.SetProperty(o => o.Status, OrderStatus.Archived),
        ct);

---

Agent Gotchas

1. Do not cast IQueryable<T> to IEnumerable<T> before filtering -- this silently switches from server-side SQL evaluation to client-side in-memory evaluation, potentially loading entire tables. Check for AsEnumerable(), explicit casts, or method signatures that accept IEnumerable<T>.
2. Do not return IQueryable<T> from repository methods -- callers can compose additional operators, but the DbContext may be disposed before enumeration. Return materialized collections (List<T>) or use IAsyncEnumerable<T>.
3. Do not optimize LINQ allocations without benchmarks -- LINQ iterator overhead is negligible for most business logic. Use [skill:dotnet-benchmarkdotnet] [MemoryDiagnoser] to prove allocations matter before replacing LINQ with manual loops.
4. Do not use compiled queries with dynamic predicates -- compiled queries cache the expression tree shape. If the query shape changes per call (conditional includes, dynamic filters), the compiled query throws or produces wrong results.
5. Do not enumerate a deferred query multiple times -- each enumeration re-executes the underlying source (database query, network call). Materialize with ToList() when the result will be consumed more than once.
6. Do not use Skip()/Take() for deep pagination -- offset pagination is O(N) on the database. Use keyset (seek) pagination with a Where clause on the last-seen key for consistent performance regardless of page depth.

---

References

• EF Core query evaluation
• EF Core compiled queries
• EF Core efficient querying
• LINQ execution model (deferred vs immediate)
• MemoryExtensions class
• EF Core ExecuteUpdate and ExecuteDelete
• Keyset pagination in EF Core