👁 Image
Pandatech.DistributedCache 6.0.0

Pandatech.DistributedCache

A focused .NET library that implements Microsoft's HybridCache abstraction on top of Redis. Provides strongly typed caching with MessagePack serialization, distributed locking, stampede protection, tag-based invalidation, and rate limiting — in under 500 lines of code.

Targets net9.0 and net10.0 only. HybridCache graduated from preview in .NET 9; net8 is not supported.

Table of Contents

1. Features
2. Installation
3. Getting Started
4. Caching
5. Tag-Based Invalidation
6. HybridCache Extensions
7. Rate Limiting
8. Distributed Locking
9. String Extensions
10. Configuration Reference
11. MessagePack Serialization
12. Health Check

Features

• HybridCache implementation — backs Microsoft's HybridCache abstraction with Redis, no local L1 layer
• Stampede protection — concurrent GetOrCreateAsync calls on the same key are serialized; only one caller hits the factory
• Tag-based invalidation — group cache entries under one or more tags and invalidate them all in one call
• Distributed locking — Redis-backed IDistributedLockService with atomic acquire/release via Lua
• Rate limiting — business-logic-level rate limiting with per-action, per-identity counters
• MessagePack serialization — binary, compact, fast; enforced uniformly across all cache entries
• HybridCache extension methods — GetOrDefaultAsync, TryGetAsync, ExistsAsync
• String key helpers — PrefixWithAssemblyName and PrefixWith for structured, collision-safe key naming
• Redis health check — auto-registered with a 3-second timeout on AddDistributedCache

Installation

dotnet add package Pandatech.DistributedCache

Getting Started

One call in Program.cs wires everything up:

builder.AddDistributedCache(options =>
{
 options.RedisConnectionString = "localhost:6379"; // required
 options.ChannelPrefix = "myapp"; // optional, default: null
});

AddDistributedCache registers:

• IConnectionMultiplexer (singleton, with exponential reconnect)
• HybridCache → RedisDistributedCache (singleton)
• IRateLimitService → RedisRateLimitService (singleton)
• IDistributedLockService → RedisLockService (singleton)
• Redis health check with a 3-second timeout

Caching

Preparing a cached model

Decorate your model with [MessagePackObject] and implement ICacheEntity:

[MessagePackObject]
public class UserSessionCache : ICacheEntity
{
 [Key(0)] public Guid UserId { get; set; }
 [Key(1)] public string Role { get; set; } = string.Empty;
 [Key(2)] public DateTime ExpiresAt { get; set; }
}

ICacheEntity is a marker interface with no members. It exists to make the intent explicit at the type level.

GetOrCreateAsync

Inject HybridCache directly. If the key is absent, the factory runs once — concurrent callers block until the first writer is done (stampede protection):

public class SessionService(HybridCache cache)
{
 public async Task<UserSessionCache> GetSessionAsync(Guid userId, CancellationToken ct = default)
 {
 return await cache.GetOrCreateAsync(
 $"session:{userId}",
 async _ => await LoadFromDbAsync(userId, ct),
 new HybridCacheEntryOptions { Expiration = TimeSpan.FromMinutes(30) },
 tags: [$"user:{userId}"],
 cancellationToken: ct);
 }
}

SetAsync

await cache.SetAsync(
 $"session:{userId}",
 session,
 new HybridCacheEntryOptions { Expiration = TimeSpan.FromMinutes(30) },
 tags: [$"user:{userId}"],
 cancellationToken: ct);

If Expiration is omitted, DefaultExpiration from configuration is used (default: 15 minutes). Pass TimeSpan.MaxValue to store without an expiry.

RemoveAsync

await cache.RemoveAsync($"session:{userId}", ct);

Tag-Based Invalidation

Tags let you invalidate a group of related entries without knowing their individual keys. Calling RemoveByTagAsync writes a tombstone timestamp for that tag. The next read of any entry carrying that tag checks the tombstone — if the tag was updated after the entry was written, the entry is evicted and re-fetched.

// Invalidate all cache entries tagged with "user:{userId}"
await cache.RemoveByTagAsync($"user:{userId}", ct);

An entry can carry multiple tags:

tags: ["user:42", "tenant:7"]

Invalidating either tag is enough to evict the entry on next read.

HybridCache Extensions

Three extension methods on HybridCache cover the most common patterns that the base API handles awkwardly.

GetOrDefaultAsync

Returns a cached value or a caller-supplied default without writing anything to Redis:

var value = await cache.GetOrDefaultAsync("feature-flag:dark-mode", defaultValue: false, ct);

TryGetAsync

Returns whether the key exists alongside its value in one round-trip:

var (exists, session) = await cache.TryGetAsync<UserSessionCache>($"session:{userId}", ct);
if (!exists)
{
 // key is not in cache
}

ExistsAsync

Checks presence without deserializing the value:

var isActive = await cache.ExistsAsync<UserSessionCache>($"session:{userId}", ct);

All three extensions are implemented against the HybridCache abstraction, so they work with any compatible implementation — not just this one.

Rate Limiting

IRateLimitService applies business-logic rate limits per action type and identity. State is stored in Redis and is consistent across all instances of your service.

Define action types and configurations

public enum ActionType
{
 SmsOtp = 1,
 EmailOtp = 2,
 Login = 3
}

public static class RateLimits
{
 public static RateLimitConfiguration SmsOtp() => new()
 {
 ActionType = (int)ActionType.SmsOtp,
 MaxAttempts = 3,
 TimeToLive = TimeSpan.FromMinutes(10)
 };

 public static RateLimitConfiguration Login() => new()
 {
 ActionType = (int)ActionType.Login,
 MaxAttempts = 10,
 TimeToLive = TimeSpan.FromMinutes(15)
 };
}

Apply the limit

public class AuthService(IRateLimitService rateLimitService)
{
 public async Task<RateLimitState> RequestOtpAsync(string phoneNumber, CancellationToken ct = default)
 {
 var config = RateLimits.SmsOtp().SetIdentifiers(phoneNumber);
 var state = await rateLimitService.RateLimitAsync(config, ct);

 if (state.Status == RateLimitStatus.Exceeded)
 {
 // state.TimeToReset — how long until the window resets
 // state.RemainingAttempts — always 0 here
 throw new TooManyRequestsException($"Try again in {state.TimeToReset.TotalSeconds:0}s.");
 }

 // state.RemainingAttempts — how many calls are left in the window
 await SendSmsAsync(phoneNumber, ct);
 return state;
 }
}

SetIdentifiers takes a primary identifier (e.g. phone number) and an optional secondary identifier (e.g. tenant ID). The two together form a unique rate-limit key for that action type.

RateLimitState always contains:

Distributed Locking

IDistributedLockService is available for cases where you need explicit locking outside of the cache layer. The implementation uses SET NX for acquire and a Lua script for atomic release — the standard Redis lock pattern.

public class InventoryService(IDistributedLockService locks)
{
 public async Task DeductStockAsync(int productId, int quantity, CancellationToken ct = default)
 {
 var key = $"product:{productId}";
 var token = Guid.NewGuid().ToString();

 if (!await locks.AcquireLockAsync(key, token))
 {
 await locks.WaitUntilLockIsReleasedAsync(key, ct);
 // re-read state and decide what to do
 return;
 }

 try
 {
 // exclusive access to this product's stock
 }
 finally
 {
 await locks.ReleaseLockAsync(key, token);
 }
 }
}

String Extensions

Utilities for building structured, collision-safe Redis key names.

// Prefix with a literal string
"user:42".PrefixWith("myapp"); // → "myapp:user:42"

// Prefix with the calling assembly's name (resolved at call site)
"user:42".PrefixWithAssemblyName(); // → "MyService.Api:user:42"

// Batch prefix
new[] { "user:1", "user:2" }.PrefixWith("myapp"); // → ["myapp:user:1", "myapp:user:2"]
new[] { "user:1", "user:2" }.PrefixWithAssemblyName();

PrefixWithAssemblyName calls Assembly.GetCallingAssembly(), so it captures the assembly that actually calls the method — useful for shared utilities that should tag keys with the service that owns them.

Configuration Reference

All options except RedisConnectionString have sensible defaults and are optional.

Key naming

All cache keys are stored in Redis under the pattern:

DistributedCache[:{ChannelPrefix}]:{yourKey}

Tag tombstone keys follow:

DistributedCache[:{ChannelPrefix}]:tag:{tagName}

Lock keys append :lock to the prefixed cache key.

MessagePack Serialization

All cache values are serialized with MessagePack. This is not configurable — by design.

MessagePack is binary, compact (~50% of equivalent JSON), and significantly faster to serialize and deserialize than JSON or Protobuf in most .NET benchmarks. It also renders as a JSON-like view in most Redis desktop clients (e.g. Another Redis Desktop Manager), so debugging is not meaningfully harder than with JSON.

Enforcing a single serializer removes an entire class of subtle bugs (mismatched serializers between writers and readers, type name handling differences, DateTime encoding differences) and keeps the library surface small.

The trade-off: your cached models must carry [MessagePackObject] and [Key(n)] attributes. This is a one-time, mechanical annotation and does not affect your domain logic.

Health Check

AddDistributedCache automatically registers a Redis health check via AspNetCore.HealthChecks.Redis with a 3-second timeout. No additional configuration is needed.

If you expose a health endpoint:

app.MapHealthChecks("/health");

Redis connectivity is included in the response automatically. This integrates with Kubernetes liveness/readiness probes, load-balancer health checks, and any monitoring stack that speaks the ASP.NET Core health check protocol.

License

MIT