[引子]HttpClientFactory日誌不好用，自己擴充套件一個？
前言
.NetCore2.1新推出HttpClientFactory工廠類， 替代了早期的HttpClient， 並新增了彈性Http呼叫機制 (整合Policy元件)。
替換的初衷還是簡單說下：
① using(var client= new HttpClient()) 呼叫Dispose()方法，並不會很快釋放底層Socket連線， 同時新建Socket需要時間，這在高併發場景下Socket耗盡。 傳送門
② 由於①很多人會想到用單例或靜態類構建HttpClient例項，但是這裡還有一個坑，HttpClient 會忽略DNS的變化。 傳送門
HttpClientFactory 以一種模組化、可命名、彈性可預期的方式重建了HttpClient的使用方式。
現在的HttpClientFactory以依賴注入的方式整合到.NETCore 框架：
// 擷取自Startup.cs檔案服務配置部分
public void ConfigureServices(IServiceCollection services)
{
services.AddHttpClient("bce-request", x =>
x.BaseAddress = new Uri(Configuration.GetSection("BCE").GetValue<string>("BaseUrl")))
.ConfigurePrimaryHttpMessageHandler(_ => new BceAuthClientHandler()
{
AccessKey = Configuration.GetSection("BCE").GetValue<string>("AccessKey"),
SerectAccessKey = Configuration.GetSection("BCE").GetValue<string>("SecretAccessKey"),
AllowAutoRedirect = true,
UseDefaultCredentials = true
})
.SetHandlerLifetime(TimeSpan.FromHours(12))
.AddPolicyHandler(GetRetryPolicy(3));
}
static IAsyncPolicy<HttpResponseMessage> GetRetryPolicy(int retry)
{
var retryPolicy = HttpPolicyExtensions
.HandleTransientHttpError()
.OrResult(msg => msg.StatusCode == System.Net.HttpStatusCode.NotFound)
.WaitAndRetryAsync(retry, retryAttempt => TimeSpan.FromSeconds(Math.Pow(2, retryAttempt)));
return retryPolicy;
}使用時從 IHttpClientFactory工廠建立所需HttpClient例項，發起業務端請求。
觀察Info級別日誌：
19/12/04 11:06:46 [Info].[System.Net.Http.HttpClient.bce-request.LogicalHandler].[18}].[] Start processing HTTP request GET http://localhost:5000/v1/eqid/b827a9400004132a000000065dc26470 19/12/04 11:06:46 [Info].[System.Net.Http.HttpClient.bce-request.ClientHandler].[18}].[] Sending HTTP request GET http://localhost:5000/v1/eqid/b827a9400004132a000000065dc26470 19/12/04 11:06:46 [Info].[System.Net.Http.HttpClient.bce-request.ClientHandler].[34}].[] Received HTTP response after 174.5088ms - OK 19/12/04 11:06:46 [Info].[System.Net.Http.HttpClient.bce-request.LogicalHandler].[34}].[] End processing HTTP request after 211.1478ms - OK
一切都是那麼自然，優雅。
頭腦風暴
觀察上面單次請求的日誌，由外層LogicHandler和內層ClientHandler 日誌頭組成。 這樣的日誌可以想象到有2個問題：
① 在高併發使用HttpClient，日誌條數眾多，沒有類似TraceId 這樣的機制定位 某次HttpClient呼叫的完整日誌。
② 若是微服務/ 分散式呼叫，可能還有 將本次HttpClient呼叫日誌與後置api日誌 結合分析的需求， 這個日誌也支援不了。
因此本文打算重新自定義HttpClientFactory日誌處理器(給請求的全部日誌設定TraceId)，實際上CustomLoggingHttpMessageHandler只是一個引子，掌握如何擴展才是關鍵。
結合我給出的典型用法來看IHttpClientFactory元件原理：
示例中System.Net.Http.HttpClient.bce-request.LogicalHandler，System.Net.Http.HttpClient.bce-request.ClientHandler 日誌頭即是來自LoggingScopeHttpMessageHandler ，LoggingHttpMessageHandler 兩個處理器，
給出手繪的UML類圖：
本次要擴充套件的入口便是 IHttpMessageHandlerFilter介面， 核心是自定義DelegatingHandler日誌處理器
+ https://github.com/aspnet/HttpClientFactory/blob/master/src/Microsoft.Extensions.Http/Logging/LoggingHttpMessageHandlerBuilderFilter.cs
程式設計實踐
如以上分析，
P1 實現 IHttpMessageHandlerFilter介面，在介面中移除預設的兩個日誌處理器；
public class TraceIdLoggingMessageHandlerFilter : IHttpMessageHandlerBuilderFilter
{
private readonly ILoggerFactory _loggerFactory;
public TraceIdLoggingMessageHandlerFilter(ILoggerFactory loggerFactory)
{
_loggerFactory = loggerFactory ?? throw new ArgumentNullException(nameof(loggerFactory));
}
public Action<HttpMessageHandlerBuilder> Configure(Action<HttpMessageHandlerBuilder> next)
{
if (next == null)
{
throw new ArgumentNullException(nameof(next));
}
return (builder) =>
{
// Run other configuration first, we want to decorate.
next(builder);
var outerLogger =_loggerFactory.CreateLogger($"System.Net.Http.HttpClient.{builder.Name}.LogicalHandler");
builder.AdditionalHandlers.Clear();
builder.AdditionalHandlers.Insert(0,new CustomLoggingScopeHttpMessageHandler(outerLogger));
};
}
}
P2 實現帶有TraceId能力的HttpClient 日誌處理器， 並加入到 IHttpMessageHandlerFilter介面實現類；
public class CustomLoggingScopeHttpMessageHandler : DelegatingHandler
{
private readonly ILogger _logger;
public CustomLoggingScopeHttpMessageHandler(ILogger logger)
{
_logger = logger ?? throw new ArgumentNullException(nameof(logger));
}
protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage request,
CancellationToken cancellationToken)
{
if (request == null)
{
throw new ArgumentNullException(nameof(request));
}
using (Log.BeginRequestPipelineScope(_logger, request))
{
Log.RequestPipelineStart(_logger, request);
var response = await base.SendAsync(request, cancellationToken);
Log.RequestPipelineEnd(_logger, response);
return response;
}
}
private static class Log
{
private static class EventIds
{
public static readonly EventId PipelineStart = new EventId(100, "RequestPipelineStart");
public static readonly EventId PipelineEnd = new EventId(101, "RequestPipelineEnd");
}
private static readonly Func<ILogger, HttpMethod, Uri, string, IDisposable> _beginRequestPipelineScope =
LoggerMessage.DefineScope<HttpMethod, Uri, string>(
"HTTP {HttpMethod} {Uri} {CorrelationId}");
private static readonly Action<ILogger, HttpMethod, Uri, string, Exception> _requestPipelineStart =
LoggerMessage.Define<HttpMethod, Uri, string>(
LogLevel.Information,
EventIds.PipelineStart,
"Start processing HTTP request {HttpMethod} {Uri} [Correlation: {CorrelationId}]");
private static readonly Action<ILogger, HttpStatusCode,string,Exception> _requestPipelineEnd =
LoggerMessage.Define<HttpStatusCode,string>(
LogLevel.Information,
EventIds.PipelineEnd,
"End processing HTTP request - {StatusCode}, [Correlation: {CorrelationId}]");
public static IDisposable BeginRequestPipelineScope(ILogger logger, HttpRequestMessage request)
{
var correlationId = GetCorrelationIdFromRequest(request);
return _beginRequestPipelineScope(logger, request.Method, request.RequestUri, correlationId);
}
public static void RequestPipelineStart(ILogger logger, HttpRequestMessage request)
{
var correlationId = GetCorrelationIdFromRequest(request);
_requestPipelineStart(logger, request.Method, request.RequestUri, correlationId, null);
}
public static void RequestPipelineEnd(ILogger logger, HttpResponseMessage response)
{
var correlationId = GetCorrelationIdFromRequest(response.RequestMessage);
_requestPipelineEnd(logger, response.StatusCode, correlationId, null);
}
private static string GetCorrelationIdFromRequest(HttpRequestMessage request)
{
string correlationId = Guid.NewGuid().ToString();
if (request.Headers.TryGetValues("X-Correlation-ID", out var values))
correlationId = values.First();
else
request.Headers.Add("X-Correlation-ID",correlationId);
return correlationId;
}
}
}
以上TraceId的實現思路，參考了我前一篇博文《被忽略的TraceId，可以用起來了》的思路，為每次HttpClient呼叫過程設定 全域性唯一的GUID標記， 後置api服務可酌情修改以上程式碼處理。
其中寫入日誌的程式碼Copy自HttpClientFactory原始碼。
P3 在DI框架中替換原有的 IHttpMessageHandlerFilter 實現:
services.Replace(ServiceDescriptor.Singleton<IHttpMessageHandlerBuilderFilter, TraceIdLoggingMessageHandlerFilter>());
發起兩次HttpClient請求， 輸出的日誌如下:
19/12/04 12:59:22 [Info].[System.Net.Http.HttpClient.bce-request.LogicalHandler].[17}].[] Start processing HTTP request GET http://localhost:5000/v1/eqid/ad78deef00444ed7000000035de704e8 [Correlation: 03de676d-680e-4a92-aef5-749bcc3ba499] 19/12/04 12:59:22 [Info].[System.Net.Http.HttpClient.bce-request.LogicalHandler].[4}].[] End processing HTTP request - OK, [Correlation: 03de676d-680e-4a92-aef5-749bcc3ba499] 19/12/04 12:59:48 [Info].[System.Net.Http.HttpClient.bce-request.LogicalHandler].[17}].[] Start processing HTTP request GET http://localhost:5000/v1/eqid/8ea0c3b66b60f0ff100000005de704fb [Correlation: 6f14393a-3a2b-45c4-a9b4-0b4ab874ef1d] 19/12/04 12:59:48 [Info].[System.Net.Http.HttpClient.bce-request.LogicalHandler].[42}].[] End processing HTTP request - OK, [Correlation: 6f14393a-3a2b-45c4-a9b4-0b4ab874ef1d]
可以看到每次請求的開始和結束都帶上了設定的GUID TraceId。
值得提醒的是：
① 這個TraceId 可以使用你業務上獨具一格的標記，這樣在排查時， 能根據上游業務更好的追蹤日誌。
② 現在這個TraceId位於LogMessage，實際上可以為nlog自定義LogoutRenderer，將該TraceId放在顯著位置，便於ETL等日誌整合框架過濾。
That's All， 本次為解決HttpClientFactory日誌無追蹤機制的探索，思考 + 實踐 + UML製圖。
實現CustomLoggingScopeHttpMessageHandler只是擴充套件HttpClientFactory能力的一個引子，如何擴充套件HttpClientFactory能力才是關鍵，希望能被大家一些啟發。