Foreword
Microsoft.AspNetCore.ConcurrencyLimiter after AspNetCore3.0 increase for incoming requests are queued treatment, avoid running thread pool.
Our daily development may often do to a web server configuration and number of connections, the size of the request queue, then today we look at how to implement a concurrent queue length limit by the amount and form of middleware.
Queue Policy
Add Nuget
Install-Package Microsoft.AspNetCore.ConcurrencyLimiter
public void ConfigureServices(IServiceCollection services)
{
services.AddQueuePolicy(options =>
{
//最大并发请求数
options.MaxConcurrentRequests = 2;
//请求队列长度限制
options.RequestQueueLimit = 1;
});
services.AddControllers();
}
public void Configure(IApplicationBuilder app, IWebHostEnvironment env)
{
//添加并发限制中间件
app.UseConcurrencyLimiter();
app.Run(async context =>
{
Task.Delay(100).Wait(); // 100ms sync-over-async
await context.Response.WriteAsync("Hello World!");
});
if (env.IsDevelopment())
{
app.UseDeveloperExceptionPage();
}
app.UseHttpsRedirection();
app.UseRouting();
app.UseAuthorization();
app.UseEndpoints(endpoints =>
{
endpoints.MapControllers();
});
}
By the above simple configuration, we will be able to introduce him to our code, so do the value limits, and the queue length; so the question is, how he achieve it?
public static IServiceCollection AddQueuePolicy(this IServiceCollection services, Action<QueuePolicyOptions> configure)
{
services.Configure(configure);
services.AddSingleton<IQueuePolicy, QueuePolicy>();
return services;
}
QueuePolicy uses SemaphoreSlim semaphore design, SemaphoreSlim , Semaphore (Semaphore) support for simultaneous multi-threading into the protected code, object specifies the maximum number of tasks at initialization time, when the thread requests access to a resource, the semaphore is decremented, and when they are released , and increments the count of the semaphore.
/// <summary>
/// 构造方法(初始化Queue策略)
/// </summary>
/// <param name="options"></param>
public QueuePolicy(IOptions<QueuePolicyOptions> options)
{
_maxConcurrentRequests = options.Value.MaxConcurrentRequests;
if (_maxConcurrentRequests <= 0)
{
throw new ArgumentException(nameof(_maxConcurrentRequests), "MaxConcurrentRequests must be a positive integer.");
}
_requestQueueLimit = options.Value.RequestQueueLimit;
if (_requestQueueLimit < 0)
{
throw new ArgumentException(nameof(_requestQueueLimit), "The RequestQueueLimit cannot be a negative number.");
}
//使用SemaphoreSlim来限制任务最大个数
_serverSemaphore = new SemaphoreSlim(_maxConcurrentRequests);
}
ConcurrencyLimiterMiddleware Middleware
/// <summary>
/// Invokes the logic of the middleware.
/// </summary>
/// <param name="context">The <see cref="HttpContext"/>.</param>
/// <returns>A <see cref="Task"/> that completes when the request leaves.</returns>
public async Task Invoke(HttpContext context)
{
var waitInQueueTask = _queuePolicy.TryEnterAsync();
// Make sure we only ever call GetResult once on the TryEnterAsync ValueTask b/c it resets.
bool result;
if (waitInQueueTask.IsCompleted)
{
ConcurrencyLimiterEventSource.Log.QueueSkipped();
result = waitInQueueTask.Result;
}
else
{
using (ConcurrencyLimiterEventSource.Log.QueueTimer())
{
result = await waitInQueueTask;
}
}
if (result)
{
try
{
await _next(context);
}
finally
{
_queuePolicy.OnExit();
}
}
else
{
ConcurrencyLimiterEventSource.Log.RequestRejected();
ConcurrencyLimiterLog.RequestRejectedQueueFull(_logger);
context.Response.StatusCode = StatusCodes.Status503ServiceUnavailable;
await _onRejected(context);
}
}
Every time we requested will first call _queuePolicy.TryEnterAsync(), after entering the first turn on a private lock method lock, then went on to determine the total amount requested if ≥ (request queue limit size + the maximum number of concurrent requests) , if the current exceeds the quantity, then I direct throw, give you a 503 status;
if (result)
{
try
{
await _next(context);
}
finally
{
_queuePolicy.OnExit();
}
}
else
{
ConcurrencyLimiterEventSource.Log.RequestRejected();
ConcurrencyLimiterLog.RequestRejectedQueueFull(_logger);
context.Response.StatusCode = StatusCodes.Status503ServiceUnavailable;
await _onRejected(context);
}
The question is, I'm not here to say that if you set the size of it, I did not give this request your server questioning pressure, then you give me something about it.
await _serverSemaphore.WaitAsync();Asynchronous waiting to enter the semaphore if no thread is granted signal the amount of access, then enter the code execution protection; otherwise, this thread will wait here until the semaphore is released
lock (_totalRequestsLock)
{
if (TotalRequests >= _requestQueueLimit + _maxConcurrentRequests)
{
return false;
}
TotalRequests++;
}
//异步等待进入信号量，如果没有线程被授予对信号量的访问权限，则进入执行保护代码；否则此线程将在此处等待，直到信号量被释放为止
await _serverSemaphore.WaitAsync();
return true;
}
After a successful return to the side and then processed so middleware, _queuePolicy.OnExit();called by the call _serverSemaphore.Release();release signal lights, and then the total number of requests is decremented
Stack Strategy
Let's look at another way, the stack strategy, how he did it? Take a look. And then add the code on how to use it.
public void ConfigureServices(IServiceCollection services)
{
services.AddStackPolicy(options =>
{
//最大并发请求数
options.MaxConcurrentRequests = 2;
//请求队列长度限制
options.RequestQueueLimit = 1;
});
services.AddControllers();
}
Through the above configuration, we can perform for our application out appropriate policy. Let's look at how he achieved it below
public static IServiceCollection AddStackPolicy(this IServiceCollection services, Action<QueuePolicyOptions> configure)
{
services.Configure(configure);
services.AddSingleton<IQueuePolicy, StackPolicy>();
return services;
}
This can be seen by StackPolicythe class to do the strategy. To take a look at the main method
/// <summary>
/// 构造方法(初始化参数)
/// </summary>
/// <param name="options"></param>
public StackPolicy(IOptions<QueuePolicyOptions> options)
{
//栈分配
_buffer = new List<ResettableBooleanCompletionSource>();
//队列大小
_maxQueueCapacity = options.Value.RequestQueueLimit;
//最大并发请求数
_maxConcurrentRequests = options.Value.MaxConcurrentRequests;
//剩余可用空间
_freeServerSpots = options.Value.MaxConcurrentRequests;
}
When we request through the middleware call _queuePolicy.TryEnterAsync(), the first will determine whether we have a number of access requests, if _freeServerSpots> 0, then return true to us directly, let middleware directly to the next step, if we set the current queue = the queue size, then we need to cancel a previous request; behind every request to cancel reservations are canceled before;
public ValueTask<bool> TryEnterAsync()
{
lock (_bufferLock)
{
if (_freeServerSpots > 0)
{
_freeServerSpots--;
return _trueTask;
}
// 如果队列满了，取消先前的请求
if (_queueLength == _maxQueueCapacity)
{
_hasReachedCapacity = true;
_buffer[_head].Complete(false);
_queueLength--;
}
var tcs = _cachedResettableTCS ??= new ResettableBooleanCompletionSource(this);
_cachedResettableTCS = null;
if (_hasReachedCapacity || _queueLength < _buffer.Count)
{
_buffer[_head] = tcs;
}
else
{
_buffer.Add(tcs);
}
_queueLength++;
// increment _head for next time
_head++;
if (_head == _maxQueueCapacity)
{
_head = 0;
}
return tcs.GetValueTask();
}
}
When we requested to call _queuePolicy.OnExit();the stack, and then descending request length
public void OnExit()
{
lock (_bufferLock)
{
if (_queueLength == 0)
{
_freeServerSpots++;
if (_freeServerSpots > _maxConcurrentRequests)
{
_freeServerSpots--;
throw new InvalidOperationException("OnExit must only be called once per successful call to TryEnterAsync");
}
return;
}
// step backwards and launch a new task
if (_head == 0)
{
_head = _maxQueueCapacity - 1;
}
else
{
_head--;
}
//退出,出栈
_buffer[_head].Complete(true);
_queueLength--;
}
}
to sum up
Based on the characteristics of the stack structure, in practical applications, usually only performs two operations on the stack:
- Add elements to the stack, this process is called "push" (or push onto the stack);
- Extracting the specified element from the stack, this process is called "popped" (or popping);
Queue storage structure of the following two ways:
- The queues: queue structure implemented on the basis sequence table;
- Chain queues: queue structure implemented on the basis of the list;