Merge pull request #72 from andyxning/enrich_info_about_nodename

detail how node-problem-detector get node name in README

Author: Random-Liu

README.md

@@ -62,6 +62,7 @@ For example, to run without auth, use the following config:
 http://APISERVER_IP:APISERVER_PORT?inClusterConfig=false
 ```
 Refer [heapster docs](https://github.com/kubernetes/heapster/blob/1e40b0f4b5eeb3f02e11ee22c2b6fda36b6e6ea1/docs/source-configuration.md#kubernetes) for a complete list of available options.
+* `-hostname-override`: A customized node name used for node-problem-detector to update conditions and emit events. node-problem-detector gets node name first from `hostname-override`, then `NODE_NAME` environment variable and finally fall back to `os.Hostname`.
 
 ## Build Image
 Run `make` in the top directory. It will:

cmd/node_problem_detector.go

@@ -26,13 +26,15 @@ import (
 	"k8s.io/node-problem-detector/pkg/version"
 
 	"github.com/golang/glog"
+	"fmt"
 )
 
 // TODO: Move flags to options directory.
 var (
 	kernelMonitorConfigPath = flag.String("kernel-monitor", "/config/kernel-monitor.json", "The path to the kernel monitor config file")
 	apiServerOverride       = flag.String("apiserver-override", "", "Custom URI used to connect to Kubernetes ApiServer")
 	printVersion            = flag.Bool("version", false, "Print version information and quit")
+	hostnameOverride        = flag.String("hostname-override", "", "Custom node name used to override hostname")
 )
 
 func validateCmdParams() {
@@ -41,6 +43,36 @@ func validateCmdParams() {
 	}
 }
 
+func getNodeNameOrDie() string {
+	var nodeName string
+
+	// Check hostname override first for customized node name.
+	if *hostnameOverride != "" {
+		return *hostnameOverride
+	}
+
+	// Get node name from environment variable NODE_NAME
+	// By default, assume that the NODE_NAME env should have been set with
+	// downward api or user defined exported environment variable. We prefer it because sometimes
+	// the hostname returned by os.Hostname is not right because:
+	// 1. User may override the hostname.
+	// 2. For some cloud providers, os.Hostname is different from the real hostname.
+	nodeName = os.Getenv("NODE_NAME")
+	if nodeName != "" {
+		return nodeName
+	}
+
+	// For backward compatibility. If the env is not set, get the hostname
+	// from os.Hostname(). This may not work for all configurations and
+	// environments.
+	nodeName, err := os.Hostname()
+	if err != nil {
+		panic(fmt.Sprintf("Failed to get host name: %v", err))
+	}
+
+	return nodeName
+}
+
 func main() {
 	flag.Parse()
 	validateCmdParams()
@@ -50,7 +82,9 @@ func main() {
 		os.Exit(0)
 	}
 
+	nodeName := getNodeNameOrDie()
+
 	k := kernelmonitor.NewKernelMonitorOrDie(*kernelMonitorConfigPath)
-	p := problemdetector.NewProblemDetector(k, *apiServerOverride)
+	p := problemdetector.NewProblemDetector(k, *apiServerOverride, nodeName)
 	p.Run()
 }

pkg/problemclient/problem_client.go

@@ -20,7 +20,6 @@ import (
 	"encoding/json"
 	"fmt"
 	"net/url"
-	"os"
 
 	"k8s.io/kubernetes/pkg/api"
 	"k8s.io/kubernetes/pkg/api/unversioned"
@@ -51,7 +50,7 @@ type nodeProblemClient struct {
 }
 
 // NewClientOrDie creates a new problem client, panics if error occurs.
-func NewClientOrDie(apiServerOverride string) Client {
+func NewClientOrDie(apiServerOverride, nodeName string) Client {
 	c := &nodeProblemClient{clock: clock.RealClock{}}
 
 	// we have checked it is a valid URI after command line argument is parsed.:)
@@ -64,23 +63,7 @@ func NewClientOrDie(apiServerOverride string) Client {
 
 	// TODO(random-liu): Set QPS Limit
 	c.client = client.NewOrDie(cfg)
-	// Get node name from environment variable NODE_NAME
-	// By default, assume that the NODE_NAME env should have been set with
-	// downward api. We prefer it because sometimes the hostname returned
-	// by os.Hostname is not right because:
-	// 1. User may override the hostname.
-	// 2. For some cloud providers, os.Hostname is different from the real hostname.
-	c.nodeName = os.Getenv("NODE_NAME")
-	if c.nodeName == "" {
-		// For backward compatibility. If the env is not set, get the hostname
-		// from os.Hostname(). This may not work for all configurations and
-		// environments.
-		var err error
-		c.nodeName, err = os.Hostname()
-		if err != nil {
-			panic("empty node name")
-		}
-	}
+	c.nodeName = nodeName
 	c.nodeRef = getNodeRef(c.nodeName)
 	c.recorders = make(map[string]record.EventRecorder)
 	return c

pkg/problemdetector/problem_detector.go

@@ -41,8 +41,8 @@ type problemDetector struct {
 
 // NewProblemDetector creates the problem detector. Currently we just directly passed in the problem daemons, but
 // in the future we may want to let the problem daemons register themselves.
-func NewProblemDetector(monitor kernelmonitor.KernelMonitor, apiServerOverride string) ProblemDetector {
-	client := problemclient.NewClientOrDie(apiServerOverride)
+func NewProblemDetector(monitor kernelmonitor.KernelMonitor, apiServerOverride, nodeName string) ProblemDetector {
+	client := problemclient.NewClientOrDie(apiServerOverride, nodeName)
 	return &problemDetector{
 		client:           client,
 		conditionManager: condition.NewConditionManager(client, clock.RealClock{}),