Database Driver Design

This skill provides expert guidance on building production-quality database client libraries in Swift, covering wire protocol implementation, connection management, type-safe APIs, and integration with Swift Concurrency. Patterns are derived from exemplary implementations: valkey-swift and postgres-nio.

Agent Behavior Contract (Follow These Rules)

1. Prefer parameterized queries - Never concatenate user input into SQL/command strings
2. Use string interpolation for safety - Implement ExpressibleByStringInterpolation to convert values to bindings
3. Design commands as types - Each command/query should be a struct with associated response type
4. Implement state machines for protocols - Complex connection lifecycles need explicit state transitions
5. Support backpressure - Row/result streaming must respect consumer demand
6. Align actor executors - Use unownedExecutor to align with NIO event loops
7. Pool connections properly - Implement keep-alive, idle timeout, and graceful shutdown

Core Patterns

Command as Type Pattern

Define commands as types with associated response types for compile-time safety:

public protocol DatabaseCommand: Sendable, Hashable {
    associatedtype Response: Decodable
    static var name: String { get }
    func encode(into encoder: inout CommandEncoder)
}

public struct GET: DatabaseCommand {
    public typealias Response = String?
    public static var name: String { "GET" }
    public let key: String

    public func encode(into encoder: inout CommandEncoder) {
        encoder.encode(Self.name, key)
    }
}

SQL Injection Prevention via String Interpolation

public struct Query: ExpressibleByStringInterpolation {
    public var sql: String
    public var bindings: Bindings

    public struct StringInterpolation: StringInterpolationProtocol {
        var sql: String = ""
        var bindings: Bindings = Bindings()

        public mutating func appendLiteral(_ literal: String) {
            sql.append(literal)
        }

        public mutating func appendInterpolation<T: Encodable>(_ value: T) {
            bindings.append(value)
            sql.append("$\(bindings.count)")
        }
    }
}

// Usage: let query: Query = "SELECT * FROM users WHERE id = \(userId)"
// Result: sql = "SELECT * FROM users WHERE id = $1", bindings = [userId]

Actor with NIO Executor Alignment

Eliminate context switches by aligning actor executor with NIO event loop:

public final actor Connection: Sendable {
    nonisolated public let unownedExecutor: UnownedSerialExecutor

    init(channel: any Channel) {
        self.unownedExecutor = channel.eventLoop.executor.asUnownedSerialExecutor()
    }
}

State Machine with Actions Pattern

Manage complex protocol state with explicit transitions and actions:

struct ConnectionStateMachine {
    enum State {
        case idle
        case executing(QueryStateMachine)
        case closing
        case closed
        case modifying // Prevents COW during mutations
    }

    enum Action {
        case sendMessage(Message)
        case forwardResult(Result)
        case closeConnection
        case none
    }

    private var state: State = .idle

    mutating func handle(_ message: Message) -> Action {
        switch (state, message) {
        case (.idle, .query(let q)):
            state = .executing(QueryStateMachine(q))
            return .sendMessage(.parse(q))
        // ... other transitions
        }
    }
}

Length-Prefixed Binary Encoding

Write length-prefixed data with placeholder and backfill:

extension Encodable {
    func encodeRaw(into buffer: inout ByteBuffer) throws {
        // Write placeholder for length (4 bytes for Int32)
        let lengthIndex = buffer.writerIndex
        buffer.writeInteger(Int32(0))

        // Record position before encoding
        let startIndex = buffer.writerIndex

        // Encode the actual value
        try self.encode(into: &buffer)

        // Calculate and write actual length
        let length = buffer.writerIndex - startIndex
        buffer.setInteger(Int32(length), at: lengthIndex)
    }
}

Protocol Hierarchy for Encoding/Decoding

Design tiered protocols for different encoding guarantees:

// Base: runtime-determined type, may throw
public protocol ThrowingDynamicTypeEncodable: Sendable {
    func encode(into byteBuffer: inout ByteBuffer) throws
    var dataType: DataType { get }
}

// Non-throwing variant
public protocol DynamicTypeEncodable: ThrowingDynamicTypeEncodable {
    func encode(into byteBuffer: inout ByteBuffer)
}

// Static type info known at compile time
public protocol StaticTypeEncodable: ThrowingDynamicTypeEncodable {
    static var dataType: DataType { get }
}

// Non-throwing + static type info (most efficient)
public protocol NonThrowingEncodable: StaticTypeEncodable, DynamicTypeEncodable {}

Variadic Generic Row Decoding

Decode multiple columns type-safely using parameter packs:

extension Row {
    func decode<each T: Decodable>(
        _ types: (repeat each T).Type
    ) throws -> (repeat each T) {
        var index = 0
        return (repeat try decodeColumn((each T).self, at: &index))
    }
}

// Usage: let (id, name, email) = try row.decode((Int.self, String.self, String.self))

Backpressure-Aware Streaming

Implement adaptive buffer strategy for result streaming:

struct AdaptiveBuffer: BackPressureStrategy {
    var lowWatermark: Int
    var highWatermark: Int
    var currentTarget: Int

    mutating func didYield(bufferDepth: Int) -> Bool {
        // Shrink target if buffer too deep
        if bufferDepth > currentTarget * 2 {
            currentTarget = max(lowWatermark, currentTarget / 2)
        }
        return bufferDepth < currentTarget
    }

    mutating func didConsume(bufferDepth: Int) -> Bool {
        // Grow target if buffer drains completely
        if bufferDepth == 0 {
            currentTarget = min(highWatermark, currentTarget * 2)
        }
        return bufferDepth < currentTarget
    }
}

Connection Pool Integration

Conform connections to pool protocols:

extension Connection: PooledConnection {
    public typealias ConnectionID = Int
}

struct KeepAliveBehavior: ConnectionKeepAliveBehavior {
    typealias Connection = Connection

    let frequency: Duration

    func runKeepAlive(for connection: Connection) async throws {
        _ = try await connection.ping()
    }
}

final class ClientMetrics: ConnectionPoolObservabilityDelegate {
    func connectionCreated(id: Int) { /* metrics */ }
    func connectionLeased(id: Int) { /* metrics */ }
    func connectionReleased(id: Int) { /* metrics */ }
    func connectionClosed(id: Int) { /* metrics */ }
}

Depth-Limited Parsing

Prevent stack overflow with nested structures:

mutating func parseToken(maxDepth: Int = 100) throws -> Token {
    guard maxDepth > 0 else {
        throw ParsingError.tooDeeplyNested
    }

    switch tokenType {
    case .array:
        var elements: [Token] = []
        for _ in 0..<count {
            elements.append(try parseToken(maxDepth: maxDepth - 1))
        }
        return .array(elements)
    // ... other cases
    }
}

Quick Decision Tree

1. Implementing wire protocol encoding/decoding?
   - Use length-prefixed messages with placeholder + backfill
   - Define protocol tokens as enums with associated values
   - Implement depth limits for nested structures

2. Designing type-safe query API?
   - Use ExpressibleByStringInterpolation for injection prevention
   - Create Encodable/Decodable protocol hierarchies for type coercion
   - Use variadic generics for multi-column row decoding

3. Managing connections?
   - Use actors with NIO executor alignment
   - Implement hierarchical state machines
   - Support cancellation via request IDs

4. Implementing connection pooling?
   - Conform to PooledConnection protocol
   - Implement keep-alive behavior
   - Track metrics via observability delegate

Triage-First Playbook

• "SQL Injection vulnerability"
• Implement ExpressibleByStringInterpolation on query type

• Convert interpolated values to $1, $2... parameter placeholders

• "Type mismatch when decoding results"

• Define Decodable protocol with typed throws

• Wrap errors with column/cell context

• "Connection state corruption"

• Use hierarchical state machines with explicit transitions

• Add .modifying sentinel to prevent COW issues

• "Backpressure not working"

• Implement adaptive buffer strategy
• Signal demand through data source protocol

Best Practices Summary

1. Commands/Queries as types - Each operation is a struct with associated response type
2. String interpolation for safety - Prevent injection by design
3. Protocol hierarchies for encoding - Different guarantees (throwing, static types, etc.)
4. Length-prefixed wire formats - Write placeholder, encode, backfill length
5. Hierarchical state machines - Compose child machines for complex protocols
6. Adaptive backpressure - Dynamically adjust buffer targets based on consumer rate
7. Actor + NIO alignment - Eliminate context switches with unownedExecutor
8. Cell/Row wrappers - Rich metadata for better error messages
9. Prepared statement caching - Deduplicate concurrent preparations
10. Graceful shutdown - Drain pending operations before closing

Key Libraries for Reference

• valkey-swift (github.com/valkey-io/valkey-swift) - Excellent RESP3 protocol, cluster support, pub/sub
• postgres-nio (github.com/vapor/postgres-nio) - Excellent query safety, type coercion, state machines