Go env:
GOARCH =" AMD64"
GOBIN =" /根/"
GOEXE =""
GOHOSTARCH =" AMD64"
GOHOSTOS =" linux的"
GOOS =" linux的"
GOPATH =" /数据/工作区/ kubernetes"
GORACE =""
GOROOT ="在/ usr /本地/去"
GOTOOLDIR ="在/ usr /本地/去/包装/工具/ linux_amd64"
GO15VENDOREXPERIMENT =" 1"
CC =" GCC"
GOGCCFLAGS =" -fPIC -m64 -pthread -fmessage-length = 0"
CXX ="克++"
CGO_ENABLED =" 1"
转到版本:
go go go.6.6.3 linux / amd64
这个问题发生在高性能的“性能测试环境”kube-apiserver上。 kube-apiserver恐慌和退出:
fatal error: concurrent map read and map write
goroutine 77930636 [running]:
runtime.throw(0x2f4c4c0, 0x21)
/root/.gvm/gos/go1.6.3/src/runtime/panic.go:547 +0x90 fp=0xca67b477f0 sp=0xca67b477d8
runtime.mapaccess1_faststr(0x2a8e520, 0xc9e29000f0, 0x2c11220, 0xa, 0x433e360)
/root/.gvm/gos/go1.6.3/src/runtime/hashmap_fast.go:202 +0x5b fp=0xca67b47850 sp=0xca67b477f0
k8s.io/kubernetes/pkg/httplog.(*respLogger).Log(0xcbddf2ae70)
/data/gerrit/src/k8s.io/kubernetes/_output/local/go/src/k8s.io/kubernetes/pkg/httplog/log.go:180 +0x43d fp=0xca67b47af8 sp=0xca67b47850
k8s.io/kubernetes/pkg/apiserver.RecoverPanics.func1(0x7f099f157090, 0xcbddf2ae70, 0xcd7569e380)
/data/gerrit/src/k8s.io/kubernetes/_output/local/go/src/k8s.io/kubernetes/pkg/apiserver/handlers.go:174 +0x15d fp=0xca67b47b50 sp=0xca67b47af8
net/http.HandlerFunc.ServeHTTP(0xc821a4eac0, 0x7f099f157058, 0xca0f4eb450, 0xcd7569e380)
/root/.gvm/gos/go1.6.3/src/net/http/server.go:1618 +0x3a fp=0xca67b47b70 sp=0xca67b47b50
net/http.serverHandler.ServeHTTP(0xc8215a7b80, 0x7f099f157058, 0xca0f4eb450, 0xcd7569e380)
/root/.gvm/gos/go1.6.3/src/net/http/server.go:2081 +0x19e fp=0xca67b47bd0 sp=0xca67b47b70
net/http.(*conn).serve(0xc8b5d6b980)
/root/.gvm/gos/go1.6.3/src/net/http/server.go:1472 +0xf2e fp=0xca67b47f98 sp=0xca67b47bd0
runtime.goexit()
/root/.gvm/gos/go1.6.3/src/runtime/asm_amd64.s:1998 +0x1 fp=0xca67b47fa0 sp=0xca67b47f98
created by net/http.(*Server).Serve
/root/.gvm/gos/go1.6.3/src/net/http/server.go:2137 +0x44e
相应的源代码:
PKG / API服务器/ handlers.go
145 func RecoverPanics(handler http.Handler) http.Handler {
146 return http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
147 defer func() {
148 if x := recover(); x != nil {
149 http.Error(w, "apis panic. Look in log for details.", http.StatusInternalServerError)
150 glog.Errorf("APIServer panic'd on %v %v: %v\n%s\n", req.Method, req.RequestURI, x, debug.Stack())
151 }
152 }()
153 defer httplog.NewLogged(req, &w).StacktraceWhen(
httplog.StatusIsNot(
http.StatusOK,
http.StatusCreated,
http.StatusAccepted,
http.StatusBadRequest,
http.StatusMovedPermanently,
http.StatusTemporaryRedirect,
http.StatusConflict,
http.StatusNotFound,
http.StatusUnauthorized,
http.StatusForbidden,
errors.StatusUnprocessableEntity,
http.StatusSwitchingProtocols,
http.StatusRequestTimeout,
errors.StatusTooManyRequests,
),
170 ).Log()
// Dispatch to the internal handler
handler.ServeHTTP(w, req)
174 })
}
PKG / HTTPLOG / log.go:
159 func (rl *respLogger) Log() {
160 latency := time.Since(rl.startTime)
161 if glog.V(2) {
162 extraInfo := ""
163 if latency >= time.Millisecond*200 && latency < time.Second {
extraInfo = fmt.Sprintf("%d00.Millisecond", latency/(time.Millisecond*100))
} else if latency >= time.Second && latency < time.Minute { // Warning
extraInfo = fmt.Sprintf("%d.Second", latency/(time.Second))
} else if latency >= time.Minute { // nce will timeout
extraInfo = fmt.Sprintf("%d.Minutes", latency/(time.Minute))
}
method := rl.req.Method
if len(rl.req.Header["Detailed-Method"]) > 0 {
method = rl.req.Header["Detailed-Method"][0]
}
remoteIP := rl.getXForwardIPAdress(rl.req)
if !rl.hijacked {
//glog.InfoDepth(1, fmt.Sprintf("%s %s: (%v) %v%v%v [%s %s]", rl.req.Method, rl.req.RequestURI, latency, rl.status, rl.statusStack, rl.addedInfo, rl.req.Header["User-Agent"], rl.req.RemoteAddr))
180 glog.InfoDepth(1, fmt.Sprintf("%v %s %s: (%sms) %v%v [%s %s]-%s %v", rl.req.Header["X-Requestid"], method, rl.req.RequestURI, GetMilliLatency(latency), rl.status, rl.addedInfo, rl.req.Header["User-Agent"], remoteIP, extraInfo, rl.statusStack))
} else {
//glog.InfoDepth(1, fmt.Sprintf("%s %s: (%v) hijacked [%s %s]", rl.req.Method, rl.req.RequestURI, latency, rl.req.Header["User-Agent"], rl.req.RemoteAddr))
glog.InfoDepth(1, fmt.Sprintf("%v %s %s: (%sms) hijacked [%s %s]-%s", rl.req.Header["X-Requestid"], method, rl.req.RequestURI, GetMilliLatency(latency), rl.req.Header["User-Agent"], remoteIP, extraInfo))
}
}
}
在handler.ServeHTTP中,我更改了request.Header。我找不到原因:
(1)&#34; serveHTTP&#34;和&#34;推迟日志&#34;以连续的方式工作。所以不存在&#34;并发读写&#34;
(2)即使存在&#34;并发读写&#34;,也存在&#34; recover&#34; func处理恐慌,所以kube-apiserver不会退出。
这个问题耗费了我很多时间,谁能帮帮我?感谢
答案 0 :(得分:0)
我的猜测是,您已经踩到了这样一个事实:全局超时处理程序无法实际停止处理请求的goroutines。因此,当连接超时(1m,除非您已更改它),它将返回错误。如果goroutine实际上同时完成,它也可以尝试写入连接。我以为我们已经添加了一个锁来防止这会导致问题,但是标题并不总是受到锁的保护。如果您可以在干净的源代码树中重现这一点,请在Kubernetes github repo上提交一个问题。