Cache data in memory and update it regularly via a background service (.NET Core MinimalAPI example)
Suppose we have a backend that takes a long time to collect and return the necessary data. Then it would make sense for the backend to cache the data so that it doesn’t always have to recalculate it. There are mature solutions for this (see for example Caching in .NET), but I have made my own small example here anyway, for the following reasons: I want the data to be generated in the background and the business code should be as decoupled as possible from the rest. Secondly an immediate update of the data should be possible. And last but not least I wanted to implement my own small minimal API example.
Example
Suppose we had the following scenario: we want to implement an API that provides GitHub statistics. There are authors and projects in GitHub. The API should read all authors and projects that exist in GitHub. This takes some time, and should not be done on every API call, but only every few hours. The data is cached in memory and our backend does not collect the data every time, but works with the cached data.
The API should provide the following routes:
GET "/summary/authors"should return the number of all authors, and information about the actuality of the cached data.GET "/summary/projects"should do the same for the all projects.POST "/triggerCacheUpdate/authors"andPOST "/triggerCacheUpdate/projects"should cause the data to be recollected.
Below I explain the example implementation from the outside in. First I explain the implementation of the services, then how the data is collected and finally how the cache is updated in the background. The complete example can also be found on GitHub.
.NET Core Minimal API
In this example, I used .NET 6. First, using a tutorial on Microsoft Learn (Tutorial: Create a minimal API with ASP.NET Core), I created a REST service that displays “Hello World” and launched it once so that Visual Studio creates a certificate for it. Then I created the minimal APIs for the REST services I want. The tips from Minimal APIs quick reference were very helpful here.
In the end Program.cs looks like this:
using InMemCacheMinimalApi.Api;
using InMemCacheMinimalApi.BusinessLogic;
using InMemCacheMinimalApi.Cache;
var builder = WebApplication.CreateBuilder(args);
// General Services for caching (described in section 'Data generation')
CachedDataConfiguration.AddServices(builder.Services);
// Services for generating the data (described in section 'Data update')
builder.Services.AddSingleton<ICachedDataEntryProducer, AuthorsDataProducer>();
builder.Services.AddSingleton<ICachedDataEntryProducer, ProjectsDataProducer>();
// Services for the APIs (described in this section)
builder.Services.AddScoped<ISummaryService, SummaryService>();
builder.Services.AddScoped<ICacheInfoService, CacheInfoService>();
var app = builder.Build();
app.MapGet("/summary/{objectType}", (string objectType, ISummaryService service) => service.Get(objectType));
app.MapPost("/triggerCacheUpdate/{objectType}", (string objectType, ICacheInfoService service) => service.TriggerUpdate(objectType));
app.Run();
For the REST API GET "/summary/{objectType}" there is an interface ISummaryService and a matching implementation SummaryService. The SummaryService class reads the cached list of authors and creates a summary from it:
internal sealed class SummaryService : ISummaryService
{
private readonly ICachedDataRepository _cachedDataRepository;
public SummaryService(ICachedDataRepository cachedDataRepository)
{
_cachedDataRepository = cachedDataRepository;
}
public string Get(string objectType)
{
switch (objectType)
{
case "authors":
if (_cachedDataRepository.GetEntry(typeof(ListOfAuthors)) is ListOfAuthors listOfAuthors)
{
return $"{listOfAuthors.AuthorsCount} authors ({_cachedDataRepository.GetStateInfo(typeof(ListOfAuthors))})";
}
else
return $"No authors data found ({_cachedDataRepository.GetStateInfo(typeof(ListOfAuthors))})";
case "projects":
// Analogous to the authors case
}
}
}
The cache stores exactly one instance of a given object type (in this example typeof(ListAuthors)).
- The call
_cachedDataRepository.GetEntry(typeof(ListOfAuthors))can be used to read this instance from the cache. - The call to
_cachedDataRepository.GetStateInfo(typeof(ListOfAuthors))returns a state like this:(Cached data, 00:00:03 old, auto update in 00:01:57).This allows the user to see how old the cache entry is and when it will be automatically updated. - The cache was injected in the constructor via DI. I will explain this cache further below.
For the REST API POST "/triggerCacheUpdate/authors" there is another interface and a service. This service is used to trigger the cache update:
internal class CacheInfoService : ICacheInfoService
{
private readonly ICachedDataUpdater _cachedDataUpdater;
public CacheInfoService(ICachedDataUpdater cachedDataUpdater)
{
_cachedDataUpdater = cachedDataUpdater;
}
public string TriggerUpdate(string objectType)
{
switch (objectType)
{
case "authors":
_cachedDataUpdater.TriggerUpdate(typeof(ListOfAuthors));
return "Update is triggered.";
case "projects":
// Analogous to the authors case
}
}
}
The call _cachedDataUpdater.TriggerUpdate(typeof(ListOfAuthors)) is used to trigger the update of the author data. This is explained further below.
Data generation
The class ListOfAuthors contains the data of all authors:
internal sealed class ListOfAuthors : ICachedDataEntry
{
private readonly List<AuthorData>? _authors;
public ListOfAuthors(List<AuthorData>? authors)
{
_authors = authors;
}
public TimeSpan MaxCacheTime { get { return new TimeSpan(0, 2, 0); } }
public int AuthorsCount
{
get { return _authors != null ? _authors.Count : 0; }
}
}
To interact with the cache, the class must implement the ICachedDataEntry interface. This contains one single property MaxCacheTime. Defining this property, the class specifies how long the data should be cached (2 minutes in the example above).
A second class creates the data:
internal class ListOfAuthorsProducer : ICachedDataEntryProducer
{
public Type GeneratesDataType { get { return typeof(ListOfAuthors); } }
public async Task<ICachedDataEntry> GenerateDataAsync()
{
ListOfAuthors result;
// This is the code that has a long runtime
await Task.Delay(15 * 1000); // Simulate a time here
result = new ListOfAuthors(new List<AuthorData> { new AuthorData("Stanislaw", "Lem"), ... });
return result;
}
}
For interaction with the cache, this class must implement the interface ICachedDataEntryProducer. This interface has two members:
GeneratesDataTypespecifies which object type can be produced by this class (so heretypeof(ListOfAuthors)).GenerateDataAsyncis the method that collects all data and returns it. This method can take a long time.
Similary, there are the classes ListOfProjects and ListOfProjectsProducer for collecting the projects information.
The two producer classes must be provided in Program.cs for DI. Because they work in the background, they are registered as singletons:
// Services for generating the data
builder.Services.AddSingleton<ICachedDataEntryProducer, ListOfAuthorsProducer>();
builder.Services.AddSingleton<ICachedDataEntryProducer, ListOfProjectsProducer>();
Data update with the help of a Background Service
Entry point for the automatic update is the class CachedDataUpdateService, which is derived from the .NET BackgroundService base class.
internal class CachedDataUpdateService : BackgroundService
{
private readonly ICachedDataUpdater _cachedDataUpdater;
public CachedDataUpdateService(ICachedDataUpdater cachedDataUpdater)
{
_cachedDataUpdater = cachedDataUpdater;
}
protected override async Task ExecuteAsync(CancellationToken stoppingToken)
{
while (!stoppingToken.IsCancellationRequested)
{
await _cachedDataUpdater.RunNeededUpdatesAsync();
await Task.Delay(10000, stoppingToken);
}
}
}
The method CachedDataConfiguration.AddServices(builder.Services) (which is called in Program.cs) registers the CachedUpdateService (and all other internal caching classes):
public static class CachedDataConfiguration
{
public static void AddServices(IServiceCollection services)
{
services.AddHostedService<CachedDataUpdateService>();
services.AddSingleton<ICachedDataRepository, CachedDataRepository>();
services.AddSingleton<ICachedDataUpdater, CachedDataUpdater>();
services.AddSingleton<ICachedDataInternalRepository, CachedDataInMemCache>();
}
}
The CachedDataUpdateServices method ExecuteAsync is called as a background service and checks in an endless loop every 10 seconds if data has to be updated. For this it uses the DI injected CacheDataUpdater.
The implementation of CacheDataUpdater is a bit longer and is only shown here in parts.
First the constructor:
internal class CachedDataUpdater : ICachedDataUpdater
{
private readonly Dictionary<string, ICachedDataEntryProducer> _producers = new();
private readonly ICachedDataInternalRepository _cachedDataRepository;
public CachedDataUpdater(IEnumerable<ICachedDataEntryProducer> producers, ICachedDataInternalRepository cachedDataRepository)
{
_cachedDataRepository = cachedDataRepository;
foreach (var producer in producers)
{
_producers.Add(producer.GeneratesDataType.ToString(), producer);
}
}
// ...
}
As first argument the constructor receives a parameter IEnumerable<ICachedDataEntryProducer> producers. In our example the two above defined producers ListOfAuthorsProducer and ListOfProjectsProducer are passed by DI to the class. Each producer specifies via the property GeneratesDataType what kind of object it creates. The producers are stored in a dictionary according to this type.
For storing the objects an ICachedDataInternalRepository is injected into the constructor. This is the in memory storage for the objects to be cached. Essentially, this is a Dictionary with the following structure:
-
The type of the object to be cached is used as the dictionary key.
-
An object to be cached (i.e., for example, a
ListOfAuthorsobject) is wrapped in a class calledCachedData, which is stored as the dictionary value. The classCacheDatalooks like this:internal sealed class CachedData { public enum CacheState { OK, ShouldBeUpdated, InUpdate } private readonly DateTime _creationTime = DateTime.MinValue; public CachedData(ICachedDataEntry? entry) { Entry = entry; State = CacheState.ShouldBeUpdated; _creationTime = DateTime.MinValue; } public ICachedDataEntry? Entry { get; } public CacheState State { get; set; } public bool NeedsRefresh { get { return (DateTime.Now - _creationTime > Entry?.MaxCacheTime); } } }The
NeedsRefreshproperty determines whether the cached entry is already older than the maximum cache time requested by theEntry.
The RunNeededUpdatesAsync method calculates and caches the objects:
internal class CachedDataUpdater : ICachedDataUpdater
{
// ...
public async Task RunNeededUpdatesAsync()
{
foreach (var dictEntry in _producers)
{
var producer = dictEntry.Value;
string typeOfICachedDataEntry = producer.GeneratesDataType.ToString();
// A few things are omitted here...
CachedData cachedData = _cachedDataRepository.Get(typeOfICachedDataEntry);
if (cachedData == null || cachedData.State == CachedData.CacheState.ShouldBeUpdated || cachedData.AutoRefresh < TimeSpan.Zero)
{
// A few things are omitted here...
// This is where the real work happens, which can take some time to complete:
ICachedDataEntry newDataEntry = await producer.GenerateDataAsync();
var newData = new CachedData(newDataEntry)
{
State = CachedData.CacheState.OK
};
_cachedDataRepository.InsertOrUpdate(typeOfICachedDataEntry, newData);
}
}
}
// ...
}
The method RunNeededUpdatesAsync is called every 10 seconds by the background service described above. The code of this method is given here somewhat simplified, but the principle should be recognizable. At first it is determined which object type the producer can generate. Then the cache is checked to see if one of the following conditions applies: (a) the object is missing, (b) the object should be updated explicitly (cachedData.State == CachedData.CacheState.ShouldBeUpdated) or (c) if the object has reached its maximum cache time and needs to be refreshed (cachedData.NeedsRefresh). If one of these conditions is true, the GenerateDataAsync method of the producer is called and the recalculated result is stored in the cache.
The last method of CacheDataUpdate is the TriggerUpdate method:
internal class CachedDataUpdater : ICachedDataUpdater
{
public void TriggerUpdate(Type typeOfICachedDataEntry)
{
string key = typeOfICachedDataEntry.ToString();
// A few things are omitted here...
CachedData cachedData = _cachedDataRepository.Get(key);
cachedData.State = CachedData.CacheState.ShouldBeUpdated;
_cachedDataRepository.InsertOrUpdate(key, cachedData);
// A few things are omitted here...
}
}
A call to the REST API POST "/triggerCacheUpdate/authors" invokes (via intermediate steps) the TriggerUpdate method of the CacheDataUpdater. This method simply sets the cache status of the object in question to ShouldBeUpdated. The next time RunNeededUpdatesAsync is called, the object will then be updated.
Summary
The core idea of this solution is that a background service (CachedDataUpdateService) handles the complete generation, update and caching of the data. The REST APIs use only cached data and therefore can deliver results without much delay. The classes for generating the data (ListOfAuthorsProducer and ListOfProjectsProducer) are passed to the background service via dependency injection (see the constructor of CachedDataUpdater), they do not have to worry about background processing and caching. Thus this approach can be easily extended to include any number of data producers.
More details of this example can be found on GitHub.
Currently, I still don’t like the fact that my example has no automatic tests. That might be my next project…
Comment on this post ❤️
I am very interested in what readers think of this post and what ideas or questions they have. The easiest way to do this is to respond to my anonymous survey.