Caching in .NET

In this article, you'll learn about various caching mechanisms. Caching is the act of storing data in an intermediate-layer, making subsequent data retrievals faster. Conceptually, caching is a performance optimization strategy and design consideration. Caching can significantly improve app performance by making infrequently changing (or expensive to retrieve) data more readily available. This article introduces the two primary types of caching, and provides sample source code for both:

Important

There are two MemoryCache classes within .NET, one in the System.Runtime.Caching namespace and the other in the Microsoft.Extensions.Caching namespace:

While this article focuses on caching, it doesn't include the System.Runtime.Caching NuGet package. All references to MemoryCache are within the Microsoft.Extensions.Caching namespace.

All of the Microsoft.Extensions.* packages come dependency injection (DI) ready, both the IMemoryCache and IDistributedCache interfaces can be used as services.

In-memory caching

In this section, you'll learn about the Microsoft.Extensions.Caching.Memory package. The current implementation of the IMemoryCache is a wrapper around the ConcurrentDictionary<TKey,TValue>, exposing a feature-rich API. Entries within the cache are represented by the ICacheEntry, and can be any object. The in-memory cache solution is great for apps that run in a single server, where all the cached data rents memory in the app's process.

Tip

For multi-server caching scenarios, consider the Distributed caching approach as an alternative to in-memory caching.

In-memory caching API

The consumer of the cache has control over both sliding and absolute expirations:

Setting an expiration will cause entries in the cache to be evicted if they're not accessed within the expiration time allotment. Consumers have additional options for controlling cache entries, through the MemoryCacheEntryOptions. Each ICacheEntry is paired with MemoryCacheEntryOptions which exposes expiration eviction functionality with IChangeToken, priority settings with CacheItemPriority, and controlling the ICacheEntry.Size. Consider the following extension methods:

In-memory cache example

To use the default IMemoryCache implementation, call the AddMemoryCache extension method to register all the required services with DI. In the following code sample, the generic host is used to expose the ConfigureServices functionality:

using Microsoft.Extensions.Caching.Memory;
using Microsoft.Extensions.DependencyInjection;
using Microsoft.Extensions.Hosting;

IHost host = Host.CreateDefaultBuilder(args)
    .ConfigureServices(services => services.AddMemoryCache())
    .Build();

Depending on your .NET workload, you may access the IMemoryCache differently; such as constructor injection. In this sample, you use the IServiceProvider instance on the host and call generic GetRequiredService<T>(IServiceProvider) extension method:

IMemoryCache cache =
    host.Services.GetRequiredService<IMemoryCache>();

With in-memory caching services registered, and resolved through DI — you're ready to start caching. This sample iterates through the letters in the English alphabet 'A' through 'Z'. The record AlphabetLetter type holds the reference to the letter, and generates a message.

record AlphabetLetter(char Letter)
{
    internal string Message =>
        $"The '{Letter}' character is the {Letter - 64} letter in the English alphabet.";
}

The sample includes a helper function that iterates through the alphabet letters:

static async ValueTask IterateAlphabetAsync(
    Func<char, Task> asyncFunc)
{
    for (char letter = 'A'; letter <= 'Z'; ++ letter)
    {
        await asyncFunc(letter);
    }

    Console.WriteLine();
}

In the preceding C# code:

  • The Func<char, Task> asyncFunc is awaited on each iteration, passing the current letter.
  • After all letters have been processed, a blank line is written to the console.

To add items to the cache call one of the Create, or Set APIs:

var addLettersToCacheTask = IterateAlphabetAsync(letter =>
{
    MemoryCacheEntryOptions options = new()
    {
        AbsoluteExpirationRelativeToNow =
            TimeSpan.FromMilliseconds(MillisecondsAbsoluteExpiration)
    };

    _ = options.RegisterPostEvictionCallback(OnPostEviction);

    AlphabetLetter alphabetLetter =
        cache.Set(
            letter, new AlphabetLetter(letter), options);

    Console.WriteLine($"{alphabetLetter.Letter} was cached.");

    return Task.Delay(
        TimeSpan.FromMilliseconds(MillisecondsDelayAfterAdd));
});
await addLettersToCacheTask;

In the preceding C# code:

  • The variable addLettersToCacheTask delegates to IterateAlphabetAsync and is awaited.
  • The Func<char, Task> asyncFunc is argued with a lambda.
  • The MemoryCacheEntryOptions is instantiated with an absolute expiration relative to now.
  • A post eviction callback is registered.
  • An AlphabetLetter object is instantiated, and passed into Set along with letter and options.
  • The letter is written to the console as being cached.
  • Finally, a Task.Delay is returned.

For each letter in the alphabet, a cache entry is written with an expiration, and post eviction callback.

The post eviction callback writes the details of the value that was evicted to the console:

static void OnPostEviction(
    object key, object letter, EvictionReason reason, object state)
{
    if (letter is AlphabetLetter alphabetLetter)
    {
        Console.WriteLine($"{alphabetLetter.Letter} was evicted for {reason}.");
    }
};

Now that the cache is populated, another call to IterateAlphabetAsync is awaited, but this time you'll call IMemoryCache.TryGetValue:

var readLettersFromCacheTask = IterateAlphabetAsync(letter =>
{
    if (cache.TryGetValue(letter, out object? value) &&
        value is AlphabetLetter alphabetLetter)
    {
        Console.WriteLine($"{letter} is still in cache. {alphabetLetter.Message}");
    }

    return Task.CompletedTask;
});
await readLettersFromCacheTask;

If the cache contains the letter key, and the value is an instance of an AlphabetLetter it's written to the console. When the letter key is not in the cache, it was evicted and its post eviction callback was invoked.

Additional extension methods

The IMemoryCache comes with many convenience-based extension methods, including an asynchronous GetOrCreateAsync:

Put it all together

The entire sample app source code is a top-level program and requires two NuGet packages:

using Microsoft.Extensions.Caching.Memory;
using Microsoft.Extensions.DependencyInjection;
using Microsoft.Extensions.Hosting;

IHost host = Host.CreateDefaultBuilder(args)
    .ConfigureServices(services => services.AddMemoryCache())
    .Build();

IMemoryCache cache =
    host.Services.GetRequiredService<IMemoryCache>();

const int MillisecondsDelayAfterAdd = 50;
const int MillisecondsAbsoluteExpiration = 750;

static void OnPostEviction(
    object key, object letter, EvictionReason reason, object state)
{
    if (letter is AlphabetLetter alphabetLetter)
    {
        Console.WriteLine($"{alphabetLetter.Letter} was evicted for {reason}.");
    }
};

static async ValueTask IterateAlphabetAsync(
    Func<char, Task> asyncFunc)
{
    for (char letter = 'A'; letter <= 'Z'; ++ letter)
    {
        await asyncFunc(letter);
    }

    Console.WriteLine();
}

var addLettersToCacheTask = IterateAlphabetAsync(letter =>
{
    MemoryCacheEntryOptions options = new()
    {
        AbsoluteExpirationRelativeToNow =
            TimeSpan.FromMilliseconds(MillisecondsAbsoluteExpiration)
    };

    _ = options.RegisterPostEvictionCallback(OnPostEviction);

    AlphabetLetter alphabetLetter =
        cache.Set(
            letter, new AlphabetLetter(letter), options);

    Console.WriteLine($"{alphabetLetter.Letter} was cached.");

    return Task.Delay(
        TimeSpan.FromMilliseconds(MillisecondsDelayAfterAdd));
});
await addLettersToCacheTask;

var readLettersFromCacheTask = IterateAlphabetAsync(letter =>
{
    if (cache.TryGetValue(letter, out object? value) &&
        value is AlphabetLetter alphabetLetter)
    {
        Console.WriteLine($"{letter} is still in cache. {alphabetLetter.Message}");
    }

    return Task.CompletedTask;
});
await readLettersFromCacheTask;

await host.RunAsync();

record AlphabetLetter(char Letter)
{
    internal string Message =>
        $"The '{Letter}' character is the {Letter - 64} letter in the English alphabet.";
}

Feel free to adjust the MillisecondsDelayAfterAdd and MillisecondsAbsoluteExpiration values to observe the changes in behavior to the expiration and eviction of cached entries. The following is sample output from running this code, due to the non-deterministic nature of .NET events — there is no guarantee that your output will be identical.

A was cached.
B was cached.
C was cached.
D was cached.
E was cached.
F was cached.
G was cached.
H was cached.
I was cached.
J was cached.
K was cached.
L was cached.
M was cached.
N was cached.
O was cached.
P was cached.
Q was cached.
R was cached.
S was cached.
T was cached.
U was cached.
V was cached.
W was cached.
X was cached.
Y was cached.
Z was cached.

A was evicted for Expired.
C was evicted for Expired.
B was evicted for Expired.
E was evicted for Expired.
D was evicted for Expired.
F was evicted for Expired.
H was evicted for Expired.
K was evicted for Expired.
L was evicted for Expired.
J was evicted for Expired.
G was evicted for Expired.
M was evicted for Expired.
N was evicted for Expired.
I was evicted for Expired.
P was evicted for Expired.
R was evicted for Expired.
O was evicted for Expired.
Q was evicted for Expired.
S is still in cache. The 'S' character is the 19 letter in the English alphabet.
T is still in cache. The 'T' character is the 20 letter in the English alphabet.
U is still in cache. The 'U' character is the 21 letter in the English alphabet.
V is still in cache. The 'V' character is the 22 letter in the English alphabet.
W is still in cache. The 'W' character is the 23 letter in the English alphabet.
X is still in cache. The 'X' character is the 24 letter in the English alphabet.
Y is still in cache. The 'Y' character is the 25 letter in the English alphabet.
Z is still in cache. The 'Z' character is the 26 letter in the English alphabet.

Since the absolute expiration (MemoryCacheEntryOptions.AbsoluteExpirationRelativeToNow) is set, all the cached items will eventually be evicted.

Worker Service caching

One common strategy for caching data, is updating the cache independently from the consuming data services. The Worker Service template is a great example, as the BackgroundService runs independent (or in the background) from the other application code. When an application starts running that hosts an implementation of the IHostedService, the corresponding implementation (in this case the BackgroundService or "worker") start running in the same process. These hosted services are registered with DI as singletons, through the AddHostedService<THostedService>(IServiceCollection) extension method. Other services can be registered with DI with any service lifetime.

Important

The service lifetime's are very important to understand. When you call AddMemoryCache to register all of the in-memory caching services, the services are registered as singletons.

Photo service scenario

Imagine you're developing a photo service that relies on third-party API accessible via HTTP. This photo data doesn't change very often, but there is a lot of it. Each photo is represented by a simple record:

namespace CachingExamples.Memory;

public record Photo(
    int AlbumId,
    int Id,
    string Title,
    string Url,
    string ThumbnailUrl);

In the following example, you'll see several services being registered with DI. Each service has a single responsibility.

using CachingExamples.Memory;

using IHost host = Host.CreateDefaultBuilder(args)
    .ConfigureServices(services =>
    {
        services.AddMemoryCache();
        services.AddHttpClient<CacheWorker>();
        services.AddHostedService<CacheWorker>();
        services.AddScoped<PhotoService>();
        services.AddSingleton(typeof(CacheSignal<>));
    })
    .Build();

await host.StartAsync();

In the preceding C# code:

The PhotoService is responsible for getting photos that match a given criteria (or filter):

using Microsoft.Extensions.Caching.Memory;

namespace CachingExamples.Memory;

public sealed class PhotoService
{
    private readonly IMemoryCache _cache;
    private readonly CacheSignal<Photo> _cacheSignal;
    private readonly ILogger<PhotoService> _logger;

    public PhotoService(
        IMemoryCache cache,
        CacheSignal<Photo> cacheSignal,
        ILogger<PhotoService> logger) =>
        (_cache, _cacheSignal, _logger) = (cache, cacheSignal, logger);

    public async IAsyncEnumerable<Photo> GetPhotosAsync(Func<Photo, bool>? filter = default)
    {
        try
        {
            await _cacheSignal.WaitAsync();

            Photo[] photos =
                await _cache.GetOrCreateAsync(
                    "Photos", _ =>
                    {
                        _logger.LogWarning("This should never happen!");

                        return Task.FromResult(Array.Empty<Photo>());
                    });

            // If no filter is provided, use a pass-thru.
            filter ??= _ => true;

            foreach (Photo? photo in photos)
            {
                if (photo is not null && filter(photo))
                {
                    yield return photo;
                }
            }
        }
        finally
        {
            _cacheSignal.Release();
        }
    }
}

In the preceding C# code:

  • The constructor requires an IMemoryCache, CacheSignal<Photo>, and ILogger.
  • The GetPhotosAsync method:
    • Defines a Func<Photo, bool> filter parameter, and returns an IAsyncEnumerable<Photo>.
    • Calls and waits for the _cacheSignal.WaitAsync() to release, this ensures that the cache is populated before accessing the cache.
    • Calls _cache.GetOrCreateAsync(), asynchronously getting all of the photos in the cache.
    • The factory argument logs a warning, and returns an empty photo array - this should never happen.
    • Each photo in the cache is iterated, filtered and materialized with yield return.
    • Finally, the cache signal is reset.

Consumers of this service are free to call GetPhotosAsync method, and handle photos accordingly. No HttpClient is required as the cache contains the photos.

The CacheWorker is a subclass of BackgroundService:

using System.Net.Http.Json;
using Microsoft.Extensions.Caching.Memory;

namespace CachingExamples.Memory;

public class CacheWorker : BackgroundService
{
    private readonly ILogger<CacheWorker> _logger;
    private readonly HttpClient _httpClient;
    private readonly CacheSignal<Photo> _cacheSignal;
    private readonly IMemoryCache _cache;
    private readonly TimeSpan _updateInterval = TimeSpan.FromHours(3);

    private const string Url = "https://jsonplaceholder.typicode.com/photos";

    public CacheWorker(
        ILogger<CacheWorker> logger,
        HttpClient httpClient,
        CacheSignal<Photo> cacheSignal,
        IMemoryCache cache) =>
        (_logger, _httpClient, _cacheSignal, _cache) = (logger, httpClient, cacheSignal, cache);

    public override async Task StartAsync(CancellationToken cancellationToken)
    {
        await _cacheSignal.WaitAsync();
        await base.StartAsync(cancellationToken);
    }

    protected override async Task ExecuteAsync(CancellationToken stoppingToken)
    {
        while (!stoppingToken.IsCancellationRequested)
        {
            _logger.LogInformation("Updating cache.");

            try
            {
                Photo[]? photos =
                    await _httpClient.GetFromJsonAsync<Photo[]>(
                        Url, stoppingToken);

                if (photos is { Length: > 0 })
                {
                    _cache.Set("Photos", photos);
                    _logger.LogInformation(
                        "Cache updated with {Count:#,#} photos.", photos.Length);
                }
                else
                {
                    _logger.LogWarning(
                        "Unable to fetch photos to update cache.");
                }
            }
            finally
            {
                _cacheSignal.Release();
            }

            try
            {
                _logger.LogInformation(
                    "Will attempt to update the cache in {Hours} hours from now.",
                    _updateInterval.Hours);

                await Task.Delay(_updateInterval, stoppingToken);
            }
            catch (OperationCanceledException)
            {
                _logger.LogWarning("Cancellation acknowledged: shutting down.");
                break;
            }
        }
    }
}

Important

You need to override BackgroundService.StartAsync and call await _cacheSignal.WaitAsync() in order to prevent a race condition between the starting of the CacheWorker and a call to PhotoService.GetPhotosAsync.

In the preceding C# code:

  • The constructor requires an ILogger, HttpClient, CacheSignal<Photo>, and IMemoryCache.
  • The defines an _updateInterval of three hours.
  • The ExecuteAsync method:
    • Loops while the app is running.
    • Makes an HTTP request to "https://jsonplaceholder.typicode.com/photos", and maps the response as an array of Photo objects.
    • The array of photos is placed in the IMemoryCache under the "Photos" key.
    • The _cacheSignal.Release() is called, releasing any consumers who were waiting for the signal.
    • The call to Task.Delay is awaited, given the update interval.
    • After delaying for three hours, the cache is again updated.

The asynchronous signal is based on an encapsulated SemaphoreSlim instance, within a generic-type constrained singleton. The CacheSignal<T> relies on an instance of SemaphoreSlim:

namespace CachingExamples.Memory;

public sealed class CacheSignal<T>
{
    private readonly SemaphoreSlim _semaphore = new(1, 1);

    /// <summary>
    /// Exposes a <see cref="Task"/> that represents the asynchronous wait operation.
    /// When signaled (consumer calls <see cref="Release"/>), the 
    /// <see cref="Task.Status"/> is set as <see cref="TaskStatus.RanToCompletion"/>.
    /// </summary>
    public Task WaitAsync() => _semaphore.WaitAsync();

    /// <summary>
    /// Exposes the ability to signal the release of the <see cref="WaitAsync"/>'s operation.
    /// Callers who were waiting, will be able to continue.
    /// </summary>
    public void Release() => _semaphore.Release();
}

In the preceding C# code, the decorator pattern is used to wrap an instance of the SemaphoreSlim. Since the CacheSignal<T> is registered as a singleton, it can be used across all service lifetimes with any generic type — in this case, the Photo. It is responsible for signaling the seeding of the cache.

Distributed caching

In some scenarios, a distributed cache is required — such is the case with multiple app servers. A distributed cache supports higher scale-out than the in-memory caching approach. Using a distributed cache offloads the cache memory to an external process, but does require extra network I/O and introduces a bit more latency (even if nominal).

The distributed caching abstractions are part of the Microsoft.Extensions.Caching.Memory NuGet package, and there is even an AddDistributedMemoryCache extension method.

Caution

The AddDistributedMemoryCache should only be used in development and/or testing scenarios, and is not a viable production implementation.

Consider any of the available implementations of the IDistributedCache from the following packages:

Distributed caching API

The distributed caching APIs are a bit more primitive than their in-memory caching API counterparts. The key-value pairs are a bit more basic. In-memory caching keys are based on an object, whereas the distributed keys are a string. With in-memory caching, the value can be any strongly-typed generic, whereas values in distributed caching are persisted as byte[]. That's not to say that various implementations don't expose strongly-typed generic values but that would be an implementation detail.

Create values

To create values in the distributed cache, call one of the set APIs:

Using the AlphabetLetter record from the in-memory cache example, you could serialize the object to JSON and then encode the string as a byte[]:

DistributedCacheEntryOptions options = new()
{
    AbsoluteExpirationRelativeToNow =
        TimeSpan.FromMilliseconds(MillisecondsAbsoluteExpiration)
};

AlphabetLetter alphabetLetter = new(letter);
string json = JsonSerializer.Serialize(alphabetLetter);
byte[] bytes = Encoding.UTF8.GetBytes(json);

await cache.SetAsync(letter.ToString(), bytes, options);

Much like in-memory caching, cache entries can have options to help fine-tune their existence in the cache — in this case, the DistributedCacheEntryOptions.

Create extension methods

There are several convenience-based extension methods for creating values, that help to avoid encoding string representations of objects into a byte[]:

Read values

To read values from the distributed cache, call one of the get APIs:

AlphabetLetter? alphabetLetter = null;
byte[]? bytes = await cache.GetAsync(letter.ToString());
if (bytes is { Length: > 0 })
{
    string json = Encoding.UTF8.GetString(bytes);
    alphabetLetter = JsonSerializer.Deserialize<AlphabetLetter>(json);
}

Once a cache entry is read out of the cache, you can get the UTF8 encoded string representation from the byte[]

Read extension methods

There are several convenience-based extension methods for reading values, that help to avoid decoding byte[] into string representations of objects:

Update values

There is no way to actually update the values in the distributed cache with a single API call, instead values can have their sliding expirations reset with one of the refresh APIs:

If the actual value needs to be updated, you'd have to delete the value and then re-add it.

Delete values

To delete values in the distributed cache, call one of the remove APIs:

Tip

While there are synchronous versions of the aforementioned APIs, please consider the fact that implementations of distributed caches are reliant on network I/O. For this reason, it is preferred more often than not to use the asynchronous APIs.

See also