Merge branch 'master' of https://github.com/didi/nightingale

changelog

@@ -139,4 +139,15 @@
   - 使用etc/nginx.conf替换原来的nginx.conf
   - n9e-prober替换旧的n9e-prober
   - n9e-agentd替换n9e-agent
-  - n9e-server替换n9e-rdb、n9e-ams、n9e-job、n9e-monapi、n9e-transfer、n9e-judge
+  - n9e-server替换n9e-rdb、n9e-ams、n9e-job、n9e-monapi、n9e-transfer、n9e-judge
+
+4.0.1
+升级内容：
+  - 修复消息通知的问题
+
+重新安装：见 https://n9e.didiyun.com/docs/install/binary/
+
+升级方法：
+  - 将 *.tpl 文件放到 etc/tpl 下
+  - 替换etc/server.yml
+  - 替换n9e-server

control

@@ -1,7 +1,7 @@
 #!/bin/bash
 
 # release version
-version=4.0.0
+version=4.0.1
 
 CWD=$(cd $(dirname $0)/; pwd)
 cd $CWD
@@ -206,9 +206,10 @@ exec()
   params=${@:2}
 
   if [ ${#2} -gt 0 ]; then
-    for param in $params
+    for mod in $params
     do
-      $1 $param
+      mod=${mod#n9e-}
+      $1 $mod
       if [ "x${mod}" = "xall" ]; then
         break
       fi

etc/agentd.yml

@@ -43,8 +43,8 @@ report:
 
   fields:
     cpu: cat /proc/cpuinfo | grep processor | wc -l
-    mem: cat /proc/meminfo | grep MemTotal | awk '{printf "%dGi", $2/1024/1024}'
-    disk: df -m | grep '/dev/' | grep -v '/var/lib' | grep -v tmpfs | awk '{sum += $2};END{printf "%dGi", sum/1024}'
+    mem: cat /proc/meminfo | grep MemTotal | awk '{printf "%.1fGi", $2/1024/1024}'
+    disk: df -m | grep '/dev/' | grep -v '/var/lib' | grep -v tmpfs | awk '{sum += $2};END{printf "%.1fGi", sum/1024}'
 
 sys:
   # timeout in ms

etc/dict.json

@@ -1 +0,0 @@
-{}

etc/nginx.conf

@@ -61,33 +61,13 @@ http {
     gzip_vary off;
     gzip_disable "MSIE [1-6]\.";
 
-    upstream n9e.rdb {
-        server 127.0.0.1:8000;
-        keepalive 60;
-    }
-
-    upstream n9e.ams {
-        server 127.0.0.1:8000;
-        keepalive 60;
-    }
-
-    upstream n9e.job {
-        server 127.0.0.1:8000;
-        keepalive 60;
-    }
-
-    upstream n9e.monapi {
-        server 127.0.0.1:8000;
-        keepalive 60;
-    }
-
-    upstream n9e.transfer {
+    upstream n9e.server {
         server 127.0.0.1:8000;
         keepalive 60;
     }
 
     upstream n9e.index {
-        server 127.0.0.1:8000;
+        server 127.0.0.1:8012;
         keepalive 60;
     }
 
@@ -112,27 +92,27 @@ http {
         }
 
         location /api/rdb {
-            proxy_pass http://n9e.rdb;
+            proxy_pass http://n9e.server;
         }
 
         location /api/ams {
-            proxy_pass http://n9e.ams;
+            proxy_pass http://n9e.server;
         }
 
         location /api/job {
-            proxy_pass http://n9e.job;
+            proxy_pass http://n9e.server;
         }
 
         location /api/mon {
-            proxy_pass http://n9e.monapi;
+            proxy_pass http://n9e.server;
         }
 
         location /api/index {
-            proxy_pass http://n9e.index;
+            proxy_pass http://n9e.server;
         }
 
         location /api/transfer {
-            proxy_pass http://n9e.transfer;
+            proxy_pass http://n9e.server;
         }
     }
 

etc/server.yml

@@ -19,6 +19,7 @@ http:
 
 tokens:
   - rdb-builtin-token
+  - ams-builtin-token
 
 redis:
   # as queue 
@@ -120,9 +121,13 @@ transfer:
       seriesLimit: 0
       docsLimit: 0
       daysLimit: 7                               # max query time
+
       # https://m3db.github.io/m3/m3db/architecture/consistencylevels/
       writeConsistencyLevel: "majority"          # one|majority|all
       readConsistencyLevel: "unstrict_majority"  # one|unstrict_majority|majority|all
+      writeTimeout: 5s
+      fetchTimeout: 5s
+      connectTimeout: 5s
       config:
         service:
           # KV environment, zone, and service from which to write/read KV data (placement
@@ -195,4 +200,4 @@ judge:
 wechat:
   corp_id: "xxxxxxxxxxxxx"
   agent_id: 1000000
-  secret: "xxxxxxxxxxxxxxxxx"
+  secret: "xxxxxxxxxxxxxxxxx"

etc/service/server.service

@@ -17,5 +17,8 @@ Restart=always
 RestartSec=1
 StartLimitInterval=0
 
+# LimitNOFILE=60000
+# LimitNPROC=10000
+
 [Install]
 WantedBy=multi-user.target

src/modules/server/http/router_auth.go

@@ -58,7 +58,7 @@ func getConfigFile(name, ext string) (string, error) {
 }
 
 func init() {
-	filename, err := getConfigFile("login-code-sms", "tpl")
+	filename, err := getConfigFile("tpl/login-code-sms", "tpl")
 	if err != nil {
 		log.Fatal(err)
 	}
@@ -68,7 +68,7 @@ func init() {
 		log.Fatalf("open %s err: %s", filename, err)
 	}
 
-	filename, err = getConfigFile("login-code-email", "tpl")
+	filename, err = getConfigFile("tpl/login-code-email", "tpl")
 	if err != nil {
 		log.Fatal(err)
 	}
@@ -77,7 +77,7 @@ func init() {
 		log.Fatalf("open %s err: %s", filename, err)
 	}
 
-	filename, err = getConfigFile("password-changed-email", "tpl")
+	filename, err = getConfigFile("tpl/password-changed-email", "tpl")
 	if err != nil {
 		log.Fatal(err)
 	}

src/modules/server/notify/notify.go

@@ -54,6 +54,7 @@ func DoNotify(isUpgrade bool, events ...*models.Event) {
 
 	notifyTypes := config.Config.Monapi.Notify[prio]
 	for i := 0; i < len(notifyTypes); i++ {
+		var err error
 		switch notifyTypes[i] {
 		case "voice":
 			if events[0].EventType == models.ALERT {
@@ -62,34 +63,36 @@ func DoNotify(isUpgrade bool, events ...*models.Event) {
 					tos = append(tos, users[j].Phone)
 				}
 
-				send(slice.Set(tos), events[0].Sname, "", "voice")
+				err = send(slice.Set(tos), events[0].Sname, "", "voice")
 			}
+
 		case "sms":
 			tos := []string{}
 			for j := 0; j < len(users); j++ {
 				tos = append(tos, users[j].Phone)
 			}
 
-			send(slice.Set(tos), content, "", "sms")
+			err = send(slice.Set(tos), content, "", "sms")
 		case "mail":
 			tos := []string{}
 			for j := 0; j < len(users); j++ {
 				tos = append(tos, users[j].Email)
 			}
 
-			if err := send(slice.Set(tos), mailContent, subject, "mail"); err == nil {
-				logger.Infof("sendMail: %+v", events[0])
-			}
+			err = send(slice.Set(tos), mailContent, subject, "mail")
 		case "im":
 			tos := []string{}
 			for j := 0; j < len(users); j++ {
 				tos = append(tos, users[j].Im)
 			}
 
-			send(slice.Set(tos), content, "", "im")
+			err = send(slice.Set(tos), content, "", "im")
 		default:
 			logger.Errorf("not support %s to send notify, events: %+v", notifyTypes[i], events)
 		}
+		if err != nil {
+			logger.Errorf("send %s users:%+v content:%s err:%v", notifyTypes[i], users, content, err)
+		}
 	}
 }
 
@@ -179,7 +182,7 @@ func genContent(isUpgrade bool, events []*models.Event) (string, string) {
 	}
 
 	// 生成告警邮件
-	fp := path.Join(file.SelfDir(), "etc", "mail.tpl")
+	fp := path.Join(file.SelfDir(), "etc", "tpl", "mail.tpl")
 	t, err := template.ParseFiles(fp)
 	if err != nil {
 		logger.Errorf("InternalServerError: cannot parse %s %v", fp, err)
@@ -196,7 +199,7 @@ func genContent(isUpgrade bool, events []*models.Event) (string, string) {
 	}
 
 	// 生成告警短信，短信和IM复用一个内容模板
-	fp = path.Join(file.SelfDir(), "etc", "sms.tpl")
+	fp = path.Join(file.SelfDir(), "etc", "tpl", "sms.tpl")
 	t, err = template.New("sms.tpl").Funcs(template.FuncMap{
 		"unescaped":  func(str string) interface{} { return template.HTML(str) },
 		"urlconvert": func(str string) interface{} { return template.URL(str) },
@@ -411,13 +414,14 @@ func send(tos []string, content, subject, notifyType string) error {
 		return fmt.Errorf("tos is empty")
 	}
 
+	message.Tos = tos
 	message.Content = strings.TrimSpace(content)
 	if message.Content == "" {
 		return fmt.Errorf("content is blank")
 	}
 
-	if notifyType == "email" {
-		message.Subject = strings.TrimSpace(message.Subject)
+	if notifyType == "mail" {
+		message.Subject = strings.TrimSpace(subject)
 		if message.Subject == "" {
 			return fmt.Errorf("subject is blank")
 		}
@@ -428,12 +432,14 @@ func send(tos []string, content, subject, notifyType string) error {
 		redisc.Write(&message, cron.VOICE_QUEUE_NAME)
 	case "sms":
 		redisc.Write(&message, cron.SMS_QUEUE_NAME)
-	case "email":
+	case "mail":
 		redisc.Write(&message, cron.MAIL_QUEUE_NAME)
 	case "im":
 		redisc.Write(&message, cron.IM_QUEUE_NAME)
 	}
 
+	logger.Infof("write %s message:%+v", message, notifyType)
+
 	return nil
 }