背景
公司有個渠道系統，專門對接三方渠道使用，沒有什麽業務邏輯，主要是轉換報文和參數校驗之類的工作，起著一個承上啓下的作用。
最近在優化接口的響應時間，優化了代碼之後，但是時間還是達不到要求；有一個詭異的100ms左右的耗時問題，在接口中打印了請求處理時間後，和調用方的響應時間還有差了100ms左右。比如程序裏記錄150ms，但是調用方等待時間卻爲250ms左右。
下面記錄下當時詳細的定位&解決流程（其實解決很簡單，關鍵在于怎麽定位並找到解決問題的方法）
定位過程
分析代碼
渠道系統是一個常見的spring-boot web工程，使用了集成的tomcat。分析了代碼之後，發現並沒有特殊的地方，沒有特殊的過濾器或者攔截器，所以初步排除是業務代碼問題
分析調用流程
出現這個問題之後，首先確認了下接口的調用流程。由于是內部測試，所以調用流程較少。
Nginx -反向代理-> 渠道系統
公司是雲服務器，網絡走的也是雲的內網。由于不明確問題的原因，所以用排除法，首先確認服務器網絡是否有問題。
先確認發送端到Nginx Host是否有問題：
[jboss@VM_0_139_centos ~]$ ping 10.0.0.139 PING 10.0.0.139 (10.0.0.139) 56(84) bytes of data. 64 bytes from 10.0.0.139: icmp_seq=1 ttl=64 time=0.029 ms 64 bytes from 10.0.0.139: icmp_seq=2 ttl=64 time=0.041 ms 64 bytes from 10.0.0.139: icmp_seq=3 ttl=64 time=0.040 ms 64 bytes from 10.0.0.139: icmp_seq=4 ttl=64 time=0.040 ms
從ping結果上看，發送端到Nginx主機的延遲是無問題的，接下來查看Nginx到渠道系統的網絡。
# 由于日志是沒問題的，這裏直接複制上面日志了 [jboss@VM_0_139_centos ~]$ ping 10.0.0.139 PING 10.0.0.139 (10.0.0.139) 56(84) bytes of data. 64 bytes from 10.0.0.139: icmp_seq=1 ttl=64 time=0.029 ms 64 bytes from 10.0.0.139: icmp_seq=2 ttl=64 time=0.041 ms 64 bytes from 10.0.0.139: icmp_seq=3 ttl=64 time=0.040 ms 64 bytes from 10.0.0.139: icmp_seq=4 ttl=64 time=0.040 ms
從ping結果上看，Nginx到渠道系統服務器網絡延遲也是沒問題的
既然網絡看似沒問題，那麽可以繼續排除法，砍掉Nginx，客戶端直接再渠道系統的服務器上，通過回環地址（localhost）直連，避免經過網卡/dns，縮小問題範圍看看能否複現（這個應用和地址是我後期模擬的，測試的是一個空接口）：
[jboss@VM_10_91_centos tmp]$ curl -w “@curl-time.txt” http://127.0.0.1:7744/send success http: 200 dns: 0.001s redirect: 0.000s time_connect: 0.001s time_appconnect: 0.000s time_pretransfer: 0.001s time_starttransfer: 0.073s size_download: 7bytes speed_download: 95.000B/s ———- time_total: 0.073s 請求總耗時
從curl日志上看，通過回環地址調用一個空接口耗時也有73ms。這就奇怪了，跳過了中間所有調用節點（包括過濾器&攔截器之類），直接請求應用一個空接口，都有73ms的耗時，再請求一次看看：
[jboss@VM_10_91_centos tmp]$ curl -w “@curl-time.txt” http://127.0.0.1:7744/send success http: 200 dns: 0.001s redirect: 0.000s time_connect: 0.001s time_appconnect: 0.000s time_pretransfer: 0.001s time_starttransfer: 0.003s size_download: 7bytes speed_download: 2611.000B/s ———- time_total: 0.003s
更奇怪的是，第二次請求耗時就正常了，變成了3ms。經查閱資料，linux curl是默認開啓http keep-alive的。就算不開啓keep-alive，每次重新handshake，也不至于需要70ms。
經過不斷分析測試發現，連續請求的話時間就會很短，每次請求只需要幾毫秒，但是如果隔一段時間再請求，就會花費70ms以上。
從這個現象猜想，可能是某些緩存機制導致的，連續請求因爲有緩存，所以速度快，時間長緩存失效後導致時間長。
那麽這個問題點到底在哪一層呢？tomcat層還是spring-webmvc呢？
光猜想定位不了問題，還是得實際測試一下，把渠道系統的代碼放到本地ide裏啓動測試能否複現
但是導入本地Ide後，在Ide中啓動後並不能複現問題，並沒有70+ms的延遲問題。這下頭疼了，本地無法複現，不能Debug，由于問題點不在業務代碼，也不能通過加日志的方式來Debug
這時候可以祭出神器Arthas了
Arthas分析問題
Arthas 是Alibaba開源的Java診斷工具，深受開發者喜愛。
當你遇到以下類似問題而束手無策時，Arthas可以幫助你解決：
- 這個類從哪個 jar 包加載的？爲什麽會報各種類相關的 Exception？
- 我改的代碼爲什麽沒有執行到？難道是我沒 commit？分支搞錯了？
- 遇到問題無法在線上 debug，難道只能通過加日志再重新發布嗎？
- 線上遇到某個用戶的數據處理有問題，但線上同樣無法 debug，線下無法重現！
- 是否有一個全局視角來查看系統的運行狀況？
- 有什麽辦法可以監控到JVM的實時運行狀態？
上面是Arthas的官方簡介，這次我只需要用他的一個小功能trace。動態計算方法調用路徑和時間，這樣我就可以定位時間在哪個地方被消耗了。
- trace 方法內部調用路徑，並輸出方法路徑上的每個節點上耗時
- trace 命令能主動搜索 class-pattern／method-pattern
- 對應的方法調用路徑，渲染和統計整個調用鏈路上的所有性能開銷和追蹤調用鏈路。
有了神器，那麽該追蹤什麽方法呢？由于我對Tomcat源碼不是很熟，所以只能從spring mvc下手，先來trace一下spring mvc的入口：
[arthas@24851]$ trace org.springframework.web.servlet.DispatcherServlet * Press Q or Ctrl+C to abort. Affect(class-cnt:1 , method-cnt:44) cost in 508 ms. `—ts=2019-09-14 21:07:44;thread_name=http-nio-7744-exec-2;id=11;is_daemon=true;priority=5;TCCL=org.springframework.boot.web.embedded.tomcat.TomcatEmbeddedWebappClassLoader@7c136917 `—[2.952142ms] org.springframework.web.servlet.DispatcherServlet:buildLocaleContext()
`—ts=2019-09-14 21:07:44;thread_name=http-nio-7744-exec-2;id=11;is_daemon=true;priority=5;TCCL=org.springframework.boot.web.embedded.tomcat.TomcatEmbeddedWebappClassLoader@7c136917 `—[18.08903ms] org.springframework.web.servlet.DispatcherServlet:doService() +—[0.041346ms] org.apache.commons.logging.Log:isDebugEnabled() #889 +—[0.022398ms] org.springframework.web.util.WebUtils:isIncludeRequest() #898 +—[0.014904ms] org.springframework.web.servlet.DispatcherServlet:getWebApplicationContext() #910 +—[1.071879ms] javax.servlet.http.HttpServletRequest:setAttribute() #910 +—[0.020977ms] javax.servlet.http.HttpServletRequest:setAttribute() #911 +—[0.017073ms] javax.servlet.http.HttpServletRequest:setAttribute() #912 +—[0.218277ms] org.springframework.web.servlet.DispatcherServlet:getThemeSource() #913 | `—[0.137568ms] org.springframework.web.servlet.DispatcherServlet:getThemeSource() | `—[min=0.00783ms,max=0.014251ms,total=0.022081ms,count=2] org.springframework.web.servlet.DispatcherServlet:getWebApplicationContext() #782 +—[0.019363ms] javax.servlet.http.HttpServletRequest:setAttribute() #913 +—[0.070694ms] org.springframework.web.servlet.FlashMapManager:retrieveAndUpdate() #916 +—[0.01839ms] org.springframework.web.servlet.FlashMap:<init>() #920 +—[0.016943ms] javax.servlet.http.HttpServletRequest:setAttribute() #920 +—[0.015268ms] javax.servlet.http.HttpServletRequest:setAttribute() #921 +—[15.050124ms] org.springframework.web.servlet.DispatcherServlet:doDispatch() #925 | `—[14.943477ms] org.springframework.web.servlet.DispatcherServlet:doDispatch() | +—[0.019135ms] org.springframework.web.context.request.async.WebAsyncUtils:getAsyncManager() #953 | +—[2.108373ms] org.springframework.web.servlet.DispatcherServlet:checkMultipart() #960 | | `—[2.004436ms] org.springframework.web.servlet.DispatcherServlet:checkMultipart() | | `—[1.890845ms] org.springframework.web.multipart.MultipartResolver:isMultipart() #1117 | +—[2.054361ms] org.springframework.web.servlet.DispatcherServlet:getHandler() #964 | | `—[1.961963ms] org.springframework.web.servlet.DispatcherServlet:getHandler() | | +—[0.02051ms] java.util.List:iterator() #1183 | | +—[min=0.003805ms,max=0.009641ms,total=0.013446ms,count=2] java.util.Iterator:hasNext() #1183 | | +—[min=0.003181ms,max=0.009751ms,total=0.012932ms,count=2] java.util.Iterator:next() #1183 | | +—[min=0.005841ms,max=0.015308ms,total=0.021149ms,count=2] org.apache.commons.logging.Log:isTraceEnabled() #1184 | | `—[min=0.474739ms,max=1.19145ms,total=1.666189ms,count=2] org.springframework.web.servlet.HandlerMapping:getHandler() #1188 | +—[0.013071ms] org.springframework.web.servlet.HandlerExecutionChain:getHandler() #971 | +—[0.372236ms] org.springframework.web.servlet.DispatcherServlet:getHandlerAdapter() #971 | | `—[0.280073ms] org.springframework.web.servlet.DispatcherServlet:getHandlerAdapter() | | +—[0.004804ms] java.util.List:iterator() #1224 | | +—[0.003668ms] java.util.Iterator:hasNext() #1224 | | +—[0.003038ms] java.util.Iterator:next() #1224 | | +—[0.006451ms] org.apache.commons.logging.Log:isTraceEnabled() #1225 | | `—[0.012683ms] org.springframework.web.servlet.HandlerAdapter:supports() #1228 | +—[0.012848ms] javax.servlet.http.HttpServletRequest:getMethod() #974 | +—[0.013132ms] java.lang.String:equals() #975 | +—[0.003025ms] org.springframework.web.servlet.HandlerExecutionChain:getHandler() #977 | +—[0.008095ms] org.springframework.web.servlet.HandlerAdapter:getLastModified() #977 | +—[0.006596ms] org.apache.commons.logging.Log:isDebugEnabled() #978 | +—[0.018024ms] org.springframework.web.context.request.ServletWebRequest:<init>() #981 | +—[0.017869ms] org.springframework.web.context.request.ServletWebRequest:checkNotModified() #981 | +—[0.038542ms] org.springframework.web.servlet.HandlerExecutionChain:applyPreHandle() #986 | +—[0.00431ms] org.springframework.web.servlet.HandlerExecutionChain:getHandler() #991 | +—[4.248493ms] org.springframework.web.servlet.HandlerAdapter:handle() #991 | +—[0.014805ms] org.springframework.web.context.request.async.WebAsyncManager:isConcurrentHandlingStarted() #993 | +—[1.444994ms] org.springframework.web.servlet.DispatcherServlet:applyDefaultViewName() #997 | | `—[0.067631ms] org.springframework.web.servlet.DispatcherServlet:applyDefaultViewName() | +—[0.012027ms] org.springframework.web.servlet.HandlerExecutionChain:applyPostHandle() #998 | +—[0.373997ms] org.springframework.web.servlet.DispatcherServlet:processDispatchResult() #1008 | | `—[0.197004ms] org.springframework.web.servlet.DispatcherServlet:processDispatchResult() | | +—[0.007074ms] org.apache.commons.logging.Log:isDebugEnabled() #1075 | | +—[0.005467ms] org.springframework.web.context.request.async.WebAsyncUtils:getAsyncManager() #1081 | | +—[0.004054ms] org.springframework.web.context.request.async.WebAsyncManager:isConcurrentHandlingStarted() #1081 | | `—[0.011988ms] org.springframework.web.servlet.HandlerExecutionChain:triggerAfterCompletion() #1087 | `—[0.004015ms] org.springframework.web.context.request.async.WebAsyncManager:isConcurrentHandlingStarted() #1018 +—[0.005055ms] org.springframework.web.context.request.async.WebAsyncUtils:getAsyncManager() #928 `—[0.003422ms] org.springframework.web.context.request.async.WebAsyncManager:isConcurrentHandlingStarted() #928
[jboss@VM_10_91_centos tmp]$ curl -w “@curl-time.txt” http://127.0.0.1:7744/send success http: 200 dns: 0.001s redirect: 0.000s time_connect: 0.001s time_appconnect: 0.000s time_pretransfer: 0.001s time_starttransfer: 0.115s size_download: 7bytes speed_download: 60.000B/s ———- time_total: 0.115s
本次調用，調用端時間花費115ms，但是從arthas trace上看，spring mvc只消耗了18ms，那麽剩下的97ms去哪了呢？
本地測試後已經可以排除spring mvc的問題了，最後也是唯一可能出問題的點就是tomcat
可是本人並不熟悉tomcat中的源碼，就連請求入口都不清楚，tomcat裏需要trace的類都不好找。。。
不過沒關系，有神器Arthas，可以通過stack命令來反向查找調用路徑，以org.springframework.web.servlet.DispatcherServlet作爲參數：
- stack 輸出當前方法被調用的調用路徑
很多時候我們都知道一個方法被執行，但這個方法被執行的路徑非常多，或者你根本就不知道這個方法是從那裏被執行了，此時你需要的是 stack 命令。
[arthas@24851]$ stack org.springframework.web.servlet.DispatcherServlet * Press Q or Ctrl+C to abort. Affect(class-cnt:1 , method-cnt:44) cost in 495 ms. ts=2019-09-14 21:15:19;thread_name=http-nio-7744-exec-5;id=14;is_daemon=true;priority=5;TCCL=org.springframework.boot.web.embedded.tomcat.TomcatEmbeddedWebappClassLoader@7c136917 @org.springframework.web.servlet.FrameworkServlet.processRequest() at org.springframework.web.servlet.FrameworkServlet.doGet(FrameworkServlet.java:866) at javax.servlet.http.HttpServlet.service(HttpServlet.java:635) at org.springframework.web.servlet.FrameworkServlet.service(FrameworkServlet.java:851) at javax.servlet.http.HttpServlet.service(HttpServlet.java:742) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:231) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.apache.tomcat.websocket.server.WsFilter.doFilter(WsFilter.java:52) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.RequestContextFilter.doFilterInternal(RequestContextFilter.java:99) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.HttpPutFormContentFilter.doFilterInternal(HttpPutFormContentFilter.java:109) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.HiddenHttpMethodFilter.doFilterInternal(HiddenHttpMethodFilter.java:81) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.CharacterEncodingFilter.doFilterInternal(CharacterEncodingFilter.java:200) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.apache.catalina.core.StandardWrapperValve.invoke(StandardWrapperValve.java:198) at org.apache.catalina.core.StandardContextValve.invoke(StandardContextValve.java:96) at org.apache.catalina.authenticator.AuthenticatorBase.invoke(AuthenticatorBase.java:496) at org.apache.catalina.core.StandardHostValve.invoke(StandardHostValve.java:140) at org.apache.catalina.valves.ErrorReportValve.invoke(ErrorReportValve.java:81) at org.apache.catalina.core.StandardEngineValve.invoke(StandardEngineValve.java:87) at org.apache.catalina.connector.CoyoteAdapter.service(CoyoteAdapter.java:342) at org.apache.coyote.http11.Http11Processor.service(Http11Processor.java:803) at org.apache.coyote.AbstractProcessorLight.process(AbstractProcessorLight.java:66) at org.apache.coyote.AbstractProtocol$ConnectionHandler.process(AbstractProtocol.java:790) at org.apache.tomcat.util.net.NioEndpoint$SocketProcessor.doRun(NioEndpoint.java:1468) at org.apache.tomcat.util.net.SocketProcessorBase.run(SocketProcessorBase.java:49) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at org.apache.tomcat.util.threads.TaskThread$WrappingRunnable.run(TaskThread.java:61) at java.lang.Thread.run(Thread.java:748)
ts=2019-09-14 21:15:19;thread_name=http-nio-7744-exec-5;id=14;is_daemon=true;priority=5;TCCL=org.springframework.boot.web.embedded.tomcat.TomcatEmbeddedWebappClassLoader@7c136917 @org.springframework.web.servlet.DispatcherServlet.doService() at org.springframework.web.servlet.FrameworkServlet.processRequest(FrameworkServlet.java:974) at org.springframework.web.servlet.FrameworkServlet.doGet(FrameworkServlet.java:866) at javax.servlet.http.HttpServlet.service(HttpServlet.java:635) at org.springframework.web.servlet.FrameworkServlet.service(FrameworkServlet.java:851) at javax.servlet.http.HttpServlet.service(HttpServlet.java:742) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:231) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.apache.tomcat.websocket.server.WsFilter.doFilter(WsFilter.java:52) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.RequestContextFilter.doFilterInternal(RequestContextFilter.java:99) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.HttpPutFormContentFilter.doFilterInternal(HttpPutFormContentFilter.java:109) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.HiddenHttpMethodFilter.doFilterInternal(HiddenHttpMethodFilter.java:81) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.springframework.web.filter.CharacterEncodingFilter.doFilterInternal(CharacterEncodingFilter.java:200) at org.springframework.web.filter.OncePerRequestFilter.doFilter(OncePerRequestFilter.java:107) at org.apache.catalina.core.ApplicationFilterChain.internalDoFilter(ApplicationFilterChain.java:193) at org.apache.catalina.core.ApplicationFilterChain.doFilter(ApplicationFilterChain.java:166) at org.apache.catalina.core.StandardWrapperValve.invoke(StandardWrapperValve.java:198) at org.apache.catalina.core.StandardContextValve.invoke(StandardContextValve.java:96) at org.apache.catalina.authenticator.AuthenticatorBase.invoke(AuthenticatorBase.java:496) at org.apache.catalina.core.StandardHostValve.invoke(StandardHostValve.java:140) at org.apache.catalina.valves.ErrorReportValve.invoke(ErrorReportValve.java:81) at org.apache.catalina.core.StandardEngineValve.invoke(StandardEngineValve.java:87) at org.apache.catalina.connector.CoyoteAdapter.service(CoyoteAdapter.java:342) at org.apache.coyote.http11.Http11Processor.service(Http11Processor.java:803) at org.apache.coyote.AbstractProcessorLight.process(AbstractProcessorLight.java:66) at org.apache.coyote.AbstractProtocol$ConnectionHandler.process(AbstractProtocol.java:790) at org.apache.tomcat.util.net.NioEndpoint$SocketProcessor.doRun(NioEndpoint.java:1468) at org.apache.tomcat.util.net.SocketProcessorBase.run(SocketProcessorBase.java:49) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624) at org.apache.tomcat.util.threads.TaskThread$WrappingRunnable.run(TaskThread.java:61) at java.lang.Thread.run(Thread.java:748)
從stack日志上可以很直觀的看出DispatchServlet的調用棧，那麽這麽長的路徑，該trace哪個類呢（這裏跳過spring mvc中的過濾器的trace過程，實際排查的時候也trace了一遍，但這詭異的時間消耗不是由這裏過濾器産生的）？
有一定經驗的老司機從名字上大概也能猜出來從哪裏下手比較好，那就是org.apache.coyote.http11.Http11Processor.service，從名字上看，http1處理器，這可能是一個比較好的切入點。
下面來trace一下：
[arthas@24851]$ trace org.apache.coyote.http11.Http11Processor service Press Q or Ctrl+C to abort. Affect(class-cnt:1 , method-cnt:1) cost in 269 ms. `—ts=2019-09-14 21:22:51;thread_name=http-nio-7744-exec-8;id=17;is_daemon=true;priority=5;TCCL=org.springframework.boot.loader.LaunchedURLClassLoader@20ad9418 `—[131.650285ms] org.apache.coyote.http11.Http11Processor:service() +—[0.036851ms] org.apache.coyote.Request:getRequestProcessor() #667 +—[0.009986ms] org.apache.coyote.RequestInfo:setStage() #668 +—[0.008928ms] org.apache.coyote.http11.Http11Processor:setSocketWrapper() #671 +—[0.013236ms] org.apache.coyote.http11.Http11InputBuffer:init() #672 +—[0.00981ms] org.apache.coyote.http11.Http11OutputBuffer:init() #673 +—[min=0.00213ms,max=0.007317ms,total=0.009447ms,count=2] org.apache.coyote.http11.Http11Processor:getErrorState() #683 +—[min=0.002098ms,max=0.008888ms,total=0.010986ms,count=2] org.apache.coyote.ErrorState:isError() #683 +—[min=0.002448ms,max=0.007149ms,total=0.009597ms,count=2] org.apache.coyote.http11.Http11Processor:isAsync() #683 +—[min=0.002399ms,max=0.00852ms,total=0.010919ms,count=2] org.apache.tomcat.util.net.AbstractEndpoint:isPaused() #683 +—[min=0.033587ms,max=0.11832ms,total=0.151907ms,count=2] org.apache.coyote.http11.Http11InputBuffer:parseRequestLine() #687 +—[0.005384ms] org.apache.tomcat.util.net.AbstractEndpoint:isPaused() #695 +—[0.007924ms] org.apache.coyote.Request:getMimeHeaders() #702 +—[0.006744ms] org.apache.tomcat.util.net.AbstractEndpoint:getMaxHeaderCount() #702 +—[0.012574ms] org.apache.tomcat.util.http.MimeHeaders:setLimit() #702 +—[0.14319ms] org.apache.coyote.http11.Http11InputBuffer:parseHeaders() #703 +—[0.003997ms] org.apache.coyote.Request:getMimeHeaders() #743 +—[0.026561ms] org.apache.tomcat.util.http.MimeHeaders:values() #743 +—[min=0.002869ms,max=0.01203ms,total=0.014899ms,count=2] java.util.Enumeration:hasMoreElements() #745 +—[0.070114ms] java.util.Enumeration:nextElement() #746 +—[0.010921ms] java.lang.String:toLowerCase() #746 +—[0.008453ms] java.lang.String:contains() #746 +—[0.002698ms] org.apache.coyote.http11.Http11Processor:getErrorState() #775 +—[0.00307ms] org.apache.coyote.ErrorState:isError() #775 +—[0.002708ms] org.apache.coyote.RequestInfo:setStage() #777 +—[0.171139ms] org.apache.coyote.http11.Http11Processor:prepareRequest() #779 +—[0.009349ms] org.apache.tomcat.util.net.SocketWrapperBase:decrementKeepAlive() #794 +—[0.002574ms] org.apache.coyote.http11.Http11Processor:getErrorState() #800 +—[0.002696ms] org.apache.coyote.ErrorState:isError() #800 +—[0.002499ms] org.apache.coyote.RequestInfo:setStage() #802 +—[0.005641ms] org.apache.coyote.http11.Http11Processor:getAdapter() #803 +—[129.868916ms] org.apache.coyote.Adapter:service() #803 +—[0.003859ms] org.apache.coyote.http11.Http11Processor:getErrorState() #809 +—[0.002365ms] org.apache.coyote.ErrorState:isError() #809 +—[0.003844ms] org.apache.coyote.http11.Http11Processor:isAsync() #809 +—[0.002382ms] org.apache.coyote.Response:getStatus() #809 +—[0.002476ms] org.apache.coyote.http11.Http11Processor:statusDropsConnection() #809 +—[0.002284ms] org.apache.coyote.RequestInfo:setStage() #838 +—[0.00222ms] org.apache.coyote.http11.Http11Processor:isAsync() #839 +—[0.037873ms] org.apache.coyote.http11.Http11Processor:endRequest() #843 +—[0.002188ms] org.apache.coyote.RequestInfo:setStage() #845 +—[0.002112ms] org.apache.coyote.http11.Http11Processor:getErrorState() #849 +—[0.002063ms] org.apache.coyote.ErrorState:isError() #849 +—[0.002504ms] org.apache.coyote.http11.Http11Processor:isAsync() #853 +—[0.009808ms] org.apache.coyote.Request:updateCounters() #854 +—[0.002008ms] org.apache.coyote.http11.Http11Processor:getErrorState() #855 +—[0.002192ms] org.apache.coyote.ErrorState:isIoAllowed() #855 +—[0.01968ms] org.apache.coyote.http11.Http11InputBuffer:nextRequest() #856 +—[0.010065ms] org.apache.coyote.http11.Http11OutputBuffer:nextRequest() #857 +—[0.002576ms] org.apache.coyote.RequestInfo:setStage() #870 +—[0.016599ms] org.apache.coyote.http11.Http11Processor:processSendfile() #872 +—[0.008182ms] org.apache.coyote.http11.Http11InputBuffer:getParsingRequestLinePhase() #688 +—[0.0075ms] org.apache.coyote.http11.Http11Processor:handleIncompleteRequestLineRead() #690 +—[0.001979ms] org.apache.coyote.RequestInfo:setStage() #875 +—[0.001981ms] org.apache.coyote.http11.Http11Processor:getErrorState() #877 +—[0.001934ms] org.apache.coyote.ErrorState:isError() #877 +—[0.001995ms] org.apache.tomcat.util.net.AbstractEndpoint:isPaused() #877 +—[0.002403ms] org.apache.coyote.http11.Http11Processor:isAsync() #879 `—[0.006176ms] org.apache.coyote.http11.Http11Processor:isUpgrade() #881
日志裏有一個129ms的耗時點（時間比沒開arthas的時候更長是因爲arthas本身帶來的性能消耗，所以生産環境小心使用），這個就是要找的問題點。
打問題點找到了，那怎麽定位是什麽導致的問題呢，又如何解決呢？
繼續trace吧，細化到具體的代碼塊或者內容。trace由于性能考慮，不會展示所有的調用路徑，如果調用路徑過深，只有手動深入trace，原則就是trace耗時長的那個方法：
[arthas@24851]$ trace org.apache.coyote.Adapter service Press Q or Ctrl+C to abort. Affect(class-cnt:1 , method-cnt:1) cost in 608 ms. `—ts=2019-09-14 21:34:33;thread_name=http-nio-7744-exec-1;id=10;is_daemon=true;priority=5;TCCL=org.springframework.boot.loader.LaunchedURLClassLoader@20ad9418 `—[81.70999ms] org.apache.catalina.connector.CoyoteAdapter:service() +—[0.032546ms] org.apache.coyote.Request:getNote() #302 +—[0.007148ms] org.apache.coyote.Response:getNote() #303 +—[0.007475ms] org.apache.catalina.connector.Connector:getXpoweredBy() #324 +—[0.00447ms] org.apache.coyote.Request:getRequestProcessor() #331 +—[0.007902ms] java.lang.ThreadLocal:get() #331 +—[0.006522ms] org.apache.coyote.RequestInfo:setWorkerThreadName() #331 +—[73.793798ms] org.apache.catalina.connector.CoyoteAdapter:postParseRequest() #336 +—[0.001536ms] org.apache.catalina.connector.Connector:getService() #339 +—[0.004469ms] org.apache.catalina.Service:getContainer() #339 +—[0.007074ms] org.apache.catalina.Engine:getPipeline() #339 +—[0.004334ms] org.apache.catalina.Pipeline:isAsyncSupported() #339 +—[0.002466ms] org.apache.catalina.connector.Request:setAsyncSupported() #339 +—[6.01E-4ms] org.apache.catalina.connector.Connector:getService() #342 +—[0.001859ms] org.apache.catalina.Service:getContainer() #342 +—[9.65E-4ms] org.apache.catalina.Engine:getPipeline() #342 +—[0.005231ms] org.apache.catalina.Pipeline:getFirst() #342 +—[7.239154ms] org.apache.catalina.Valve:invoke() #342 +—[0.006904ms] org.apache.catalina.connector.Request:isAsync() #345 +—[0.00509ms] org.apache.catalina.connector.Request:finishRequest() #372 +—[0.051461ms] org.apache.catalina.connector.Response:finishResponse() #373 +—[0.007244ms] java.util.concurrent.atomic.AtomicBoolean:<init>() #379 +—[0.007314ms] org.apache.coyote.Response:action() #380 +—[0.004518ms] org.apache.catalina.connector.Request:isAsyncCompleting() #382 +—[0.001072ms] org.apache.catalina.connector.Request:getContext() #394 +—[0.007166ms] java.lang.System:currentTimeMillis() #401 +—[0.004367ms] org.apache.coyote.Request:getStartTime() #401 +—[0.011483ms] org.apache.catalina.Context:logAccess() #401 +—[0.0014ms] org.apache.coyote.Request:getRequestProcessor() #406 +—[min=8.0E-4ms,max=9.22E-4ms,total=0.001722ms,count=2] java.lang.Integer:<init>() #406 +—[0.001082ms] java.lang.reflect.Method:invoke() #406 +—[0.001851ms] org.apache.coyote.RequestInfo:setWorkerThreadName() #406 +—[0.035805ms] org.apache.catalina.connector.Request:recycle() #410 `—[0.007849ms] org.apache.catalina.connector.Response:recycle() #411
一段無聊的手動深入trace之後………………
[arthas@24851]$ trace org.apache.catalina.webresources.AbstractArchiveResourceSet getArchiveEntries Press Q or Ctrl+C to abort. Affect(class-cnt:4 , method-cnt:2) cost in 150 ms. `—ts=2019-09-14 21:36:26;thread_name=http-nio-7744-exec-3;id=12;is_daemon=true;priority=5;TCCL=org.springframework.boot.loader.LaunchedURLClassLoader@20ad9418 `—[75.743681ms] org.apache.catalina.webresources.JarWarResourceSet:getArchiveEntries() +—[0.025731ms] java.util.HashMap:<init>() #106 +—[0.097729ms] org.apache.catalina.webresources.JarWarResourceSet:openJarFile() #109 +—[0.091037ms] java.util.jar.JarFile:getJarEntry() #110 +—[0.096325ms] java.util.jar.JarFile:getInputStream() #111 +—[0.451916ms] org.apache.catalina.webresources.TomcatJarInputStream:<init>() #113 +—[min=0.001175ms,max=0.001176ms,total=0.002351ms,count=2] java.lang.Integer:<init>() #114 +—[0.00104ms] java.lang.reflect.Method:invoke() #114 +—[0.045105ms] org.apache.catalina.webresources.TomcatJarInputStream:getNextJarEntry() #114 +—[min=5.02E-4ms,max=0.008531ms,total=0.028864ms,count=31] java.util.jar.JarEntry:getName() #116 +—[min=5.39E-4ms,max=0.022805ms,total=0.054647ms,count=31] java.util.HashMap:put() #116 +—[min=0.004452ms,max=34.479307ms,total=74.206249ms,count=31] org.apache.catalina.webresources.TomcatJarInputStream:getNextJarEntry() #117 +—[0.018358ms] org.apache.catalina.webresources.TomcatJarInputStream:getManifest() #119 +—[0.006429ms] org.apache.catalina.webresources.JarWarResourceSet:setManifest() #120 +—[0.010904ms] org.apache.tomcat.util.compat.JreCompat:isJre9Available() #121 +—[0.003307ms] org.apache.catalina.webresources.TomcatJarInputStream:getMetaInfEntry() #133 +—[5.5E-4ms] java.util.jar.JarEntry:getName() #135 +—[6.42E-4ms] java.util.HashMap:put() #135 +—[0.001981ms] org.apache.catalina.webresources.TomcatJarInputStream:getManifestEntry() #137 +—[0.064484ms] org.apache.catalina.webresources.TomcatJarInputStream:close() #141 +—[0.007961ms] org.apache.catalina.webresources.JarWarResourceSet:closeJarFile() #151 `—[0.004643ms] java.io.InputStream:close() #155
發現了一個值得暫停思考的點：
+—[min=0.004452ms,max=34.479307ms,total=74.206249ms,count=31] org.apache.catalina.webresources.TomcatJarInputStream:getNextJarEntry() #117
這行代碼加載了31次，一共耗時74ms；從名字上看，應該是tomcat加載jar包時的耗時，那麽是加載了31個jar包的耗時，還是加載了jar包內的某些資源31次耗時呢？
TomcatJarInputStream這個類源碼的注釋寫到：
The purpose of this sub-class is to obtain references to the JarEntry objectsfor META-INF/ and META-INF/MANIFEST.MF that are otherwise swallowed by theJarInputStream implementation.
大概意思也就是，獲取jar包內META-INF/，META-INF/MANIFEST的資源，這是一個子類，更多的功能在父類JarInputStream裏。
其實看到這裏大概也能猜到問題了，tomcat加載jar包內META-INF/，META-INF/MANIFEST的資源導致的耗時，至于爲什麽連續請求不會耗時，應該是tomcat的緩存機制（下面介紹源碼分析）
不著急定位問題，試著通過Arthas最終定位問題細節，繼續手動深入trace
[arthas@24851]$ trace org.apache.catalina.webresources.TomcatJarInputStream * Press Q or Ctrl+C to abort. Affect(class-cnt:1 , method-cnt:4) cost in 44 ms. `—ts=2019-09-14 21:37:47;thread_name=http-nio-7744-exec-5;id=14;is_daemon=true;priority=5;TCCL=org.springframework.boot.loader.LaunchedURLClassLoader@20ad9418 `—[0.234952ms] org.apache.catalina.webresources.TomcatJarInputStream:createZipEntry() +—[0.039455ms] java.util.jar.JarInputStream:createZipEntry() #43 `—[0.007827ms] java.lang.String:equals() #44
`—ts=2019-09-14 21:37:47;thread_name=http-nio-7744-exec-5;id=14;is_daemon=true;priority=5;TCCL=org.springframework.boot.loader.LaunchedURLClassLoader@20ad9418 `—[0.050222ms] org.apache.catalina.webresources.TomcatJarInputStream:createZipEntry() +—[0.001889ms] java.util.jar.JarInputStream:createZipEntry() #43 `—[0.001643ms] java.lang.String:equals() #46 #這裏一共31個trace日志，刪減了剩下的
從方法名上看，還是加載資源之類的意思。都已經到jdk源碼了，這時候來看一下TomcatJarInputStream這個類的源碼:
/** * Creates a new <code>JarEntry</code> (<code>ZipEntry</code>) for the * specified JAR file entry name. The manifest attributes of * the specified JAR file entry name will be copied to the new * <CODE>JarEntry</CODE>. * * @param name the name of the JAR/ZIP file entry * @return the <code>JarEntry</code> object just created */ protected ZipEntry createZipEntry(String name) { JarEntry e = new JarEntry(name); if (man != null) { e.attr = man.getAttributes(name); } return e; }
這個createZipEntry有個name參數，從注釋上看，是jar/zip文件名，如果能得到文件名這種關鍵信息，就可以直接定位問題了；還是通過Arthas，使用watch命令，動態監測方法調用數據
watch方法執行數據觀測
讓你能方便的觀察到指定方法的調用情況。能觀察到的範圍爲：返回值、抛出異常、入參，通過編寫 OGNL 表達式進行對應變量的查看。
watch 該方法的入參
[arthas@24851]$ watch org.apache.catalina.webresources.TomcatJarInputStream createZipEntry “{params[0]}” Press Q or Ctrl+C to abort. Affect(class-cnt:1 , method-cnt:1) cost in 27 ms. ts=2019-09-14 21:51:14; [cost=0.14547ms] result=@ArrayList[ @String[META-INF/], ] ts=2019-09-14 21:51:14; [cost=0.048028ms] result=@ArrayList[ @String[META-INF/MANIFEST.MF], ] ts=2019-09-14 21:51:14; [cost=0.046071ms] result=@ArrayList[ @String[META-INF/resources/], ] ts=2019-09-14 21:51:14; [cost=0.033855ms] result=@ArrayList[ @String[META-INF/resources/swagger-ui.html], ] ts=2019-09-14 21:51:14; [cost=0.039138ms] result=@ArrayList[ @String[META-INF/resources/webjars/], ] ts=2019-09-14 21:51:14; [cost=0.033701ms] result=@ArrayList[ @String[META-INF/resources/webjars/springfox-swagger-ui/], ] ts=2019-09-14 21:51:14; [cost=0.033644ms] result=@ArrayList[ @String[META-INF/resources/webjars/springfox-swagger-ui/favicon-16×16.png], ] ts=2019-09-14 21:51:14; [cost=0.033976ms] result=@ArrayList[ @String[META-INF/resources/webjars/springfox-swagger-ui/springfox.css], ] ts=2019-09-14 21:51:14; [cost=0.032818ms] result=@ArrayList[ @String[META-INF/resources/webjars/springfox-swagger-ui/swagger-ui-standalone-preset.js.map], ] ts=2019-09-14 21:51:14; [cost=0.04651ms] result=@ArrayList[ @String[META-INF/resources/webjars/springfox-swagger-ui/swagger-ui.css], ] ts=2019-09-14 21:51:14; [cost=0.034793ms] result=@ArrayList[ @String[META-INF/resources/webjars/springfox-swagger-ui/swagger-ui.js.map],
這下直接看到了具體加載的資源名，這麽熟悉的名字：swagger-ui，一個國外的rest接口文檔工具，又有國內開發者基于swagger-ui做了一套spring mvc的集成工具，通過注解就可以自動生成swagger-ui需要的接口定義json文件，用起來還比較方便，就是侵入性較強。
刪除swagger的jar包後問題，詭異的70+ms就消失了
<!–pom 裏刪除這兩個引用，這兩個包時國內開發者封裝的，swagger-ui並沒有提供java spring-mvc的支持包，swagger只是一個浏覽器端的ui+editor –> <dependency> <groupId>io.springfox</groupId> <artifactId>springfox-swagger2</artifactId> <version>2.9.2</version> </dependency> <dependency> <groupId>io.springfox</groupId> <artifactId>springfox-swagger-ui</artifactId> <version>2.9.2</version> </dependency>
那麽爲什麽swagger會導致請求耗時呢，爲什麽每次請求偶讀會加載swagger內部的靜態資源呢？
其實這是tomcat-embed的一個bug吧，下面詳細介紹一下該Bug
Tomcat embed Bug分析&解決
源碼分析過程實在太漫長，而且也不是本文的重點，所以就不介紹了， 下面直接介紹下分析結果
順便貼一張tomcat處理請求的核心類圖