[#5748][refactor]get more reasonable monitoring indexes of the zk cluster (#5768)

* [#5748][refactor]get more reasonable monitoring indexes of the cluster

add warning state on the front-end
each zk node shows monitoring indexes correctly

Co-authored-by: unknown <jun.qi1@YTX.COM>

dolphinscheduler-api/src/main/java/org/apache/dolphinscheduler/api/utils/FourLetterWordMain.java

@@ -25,10 +25,12 @@ import java.io.BufferedReader;
 import java.io.IOException;
 import java.io.InputStreamReader;
 import java.io.OutputStream;
+
 import java.net.InetAddress;
 import java.net.InetSocketAddress;
 import java.net.Socket;
 import java.net.SocketTimeoutException;
+import java.net.ConnectException;
 import java.util.Objects;
 
 public class FourLetterWordMain {
@@ -88,7 +90,7 @@ public class FourLetterWordMain {
                 }
                 return sb.toString();
             }
-        } catch (SocketTimeoutException e) {
+        } catch (SocketTimeoutException | ConnectException e ) {
             throw new IOException("Exception while executing four letter word: " + cmd, e);
         }
     }

dolphinscheduler-api/src/main/java/org/apache/dolphinscheduler/api/utils/ZooKeeperState.java

@@ -16,6 +16,7 @@
  */
 package org.apache.dolphinscheduler.api.utils;
 
+import org.apache.dolphinscheduler.common.Constants;
 import org.apache.dolphinscheduler.common.utils.StringUtils;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
@@ -43,6 +44,8 @@ public class ZooKeeperState {
 	private int watches = -1;
 	private int connections = -1;
 
+	private boolean healthFlag = false;
+
 	public ZooKeeperState(String connectionString) {
 		String host = connectionString.substring(0,
 				connectionString.indexOf(':'));
@@ -55,55 +58,68 @@ public class ZooKeeperState {
 	public void getZookeeperInfo() {
 		String content = cmd("srvr");
 		if (StringUtils.isNotBlank(content)) {
-			try (Scanner scannerForStat = new Scanner(content))  {
+			if(Constants.STRING_FALSE.equals(content)){
-				while (scannerForStat.hasNext()) {
+				healthFlag = true;
-					String line = scannerForStat.nextLine();
+			}else {
-					if (line.startsWith("Latency min/avg/max:")) {
+				try (Scanner scannerForStat = new Scanner(content))  {
-						String[] latencys = getStringValueFromLine(line).split("/");
+					while (scannerForStat.hasNext()) {
-						minLatency = Float.parseFloat(latencys[0]);
+						String line = scannerForStat.nextLine();
-						avgLatency = Float.parseFloat(latencys[1]);
+						if (line.startsWith("Latency min/avg/max:")) {
-						maxLatency = Float.parseFloat(latencys[2]);
+							String[] latencys = getStringValueFromLine(line).split("/");
-					} else if (line.startsWith("Received:")) {
+							minLatency = Float.parseFloat(latencys[0]);
-						received = Long.parseLong(getStringValueFromLine(line));
+							avgLatency = Float.parseFloat(latencys[1]);
-					} else if (line.startsWith("Sent:")) {
+							maxLatency = Float.parseFloat(latencys[2]);
-						sent = Long.parseLong(getStringValueFromLine(line));
+						} else if (line.startsWith("Received:")) {
-					} else if (line.startsWith("Outstanding:")) {
+							received = Long.parseLong(getStringValueFromLine(line));
-						outStanding = Integer.parseInt(getStringValueFromLine(line));
+						} else if (line.startsWith("Sent:")) {
-					} else if (line.startsWith("Zxid:")) {
+							sent = Long.parseLong(getStringValueFromLine(line));
-						zxid = Long.parseLong(getStringValueFromLine(line).substring(2), 16);
+						} else if (line.startsWith("Outstanding:")) {
-					} else if (line.startsWith("Mode:")) {
+							outStanding = Integer.parseInt(getStringValueFromLine(line));
-						mode = getStringValueFromLine(line);
+						} else if (line.startsWith("Zxid:")) {
-					} else if (line.startsWith("Node count:")) {
+							zxid = Long.parseLong(getStringValueFromLine(line).substring(2), 16);
-						nodeCount = Integer.parseInt(getStringValueFromLine(line));
+						} else if (line.startsWith("Mode:")) {
+							mode = getStringValueFromLine(line);
+						} else if (line.startsWith("Node count:")) {
+							nodeCount = Integer.parseInt(getStringValueFromLine(line));
+						}
 					}
 				}
-			} 	
+			}
+
 		}
 
 		String wchsText = cmd("wchs");
 		if (StringUtils.isNotBlank(wchsText)) {
-			try (Scanner scannerForWchs = new Scanner(wchsText)) {
+			if(Constants.STRING_FALSE.equals(wchsText)){
-				while (scannerForWchs.hasNext()) {
+				healthFlag = true;
-					String line = scannerForWchs.nextLine();
+			}else {
-					if (line.startsWith("Total watches:")) {
+				try (Scanner scannerForWchs = new Scanner(wchsText)) {
-						watches = Integer.parseInt(getStringValueFromLine(line));
+					while (scannerForWchs.hasNext()) {
+						String line = scannerForWchs.nextLine();
+						if (line.startsWith("Total watches:")) {
+							watches = Integer.parseInt(getStringValueFromLine(line));
+						}
 					}
 				}
-			}	
+			}
 		}
 
 		String consText = cmd("cons");
 		if (StringUtils.isNotBlank(consText)) {
-			Scanner scannerForCons = new Scanner(consText);
+			if(Constants.STRING_FALSE.equals(consText)){
-			if (StringUtils.isNotBlank(consText)) {
+				healthFlag = true;
-				connections = 0;
+			}else {
-			}
+				Scanner scannerForCons = new Scanner(consText);
-			while (scannerForCons.hasNext()) {
+				if (StringUtils.isNotBlank(consText)) {
-				@SuppressWarnings("unused")
+					connections = 0;
-				String line = scannerForCons.nextLine();
+				}
-				++connections;
+				while (scannerForCons.hasNext()) {
+					@SuppressWarnings("unused")
+					String line = scannerForCons.nextLine();
+					++connections;
+				}
+				scannerForCons.close();
 			}
-			scannerForCons.close();
 		}
 	}
 
@@ -121,7 +137,7 @@ public class ZooKeeperState {
 	private class SendThread extends Thread {
 		private String cmd;
 
-		private String ret = "";
+		private String ret = Constants.STRING_FALSE;
 
 		public SendThread(String cmd) {
 			this.cmd = cmd;
@@ -150,7 +166,7 @@ public class ZooKeeperState {
 		} catch (InterruptedException e) {
 			logger.error("send " + cmd + " to server " + host + ":" + port + " failed!", e);
 		}
-		return "";
+		return Constants.STRING_FALSE;
 	}
 
 	public Logger getLogger() {
@@ -209,6 +225,10 @@ public class ZooKeeperState {
 		return connections;
 	}
 
+	public boolean isHealthFlag() {
+		return healthFlag;
+	}
+
 	@Override
 	public String toString() {
 		return "ZooKeeperState [host=" + host + ", port=" + port
@@ -217,7 +237,7 @@ public class ZooKeeperState {
 				+ ", sent=" + sent + ", outStanding=" + outStanding + ", zxid="
 				+ zxid + ", mode=" + mode + ", nodeCount=" + nodeCount
 				+ ", watches=" + watches + ", connections="
-				+ connections + "]";
+				+ connections + ",healthFlag=" + healthFlag + "]";
 	}
 
 

dolphinscheduler-api/src/main/java/org/apache/dolphinscheduler/api/utils/ZookeeperMonitor.java

@@ -78,9 +78,7 @@ public class ZookeeperMonitor extends AbstractZKClient {
             for (String zookeeperServer : zookeeperServersArray) {
                 ZooKeeperState state = new ZooKeeperState(zookeeperServer);
                 boolean ok = state.ruok();
-                if (ok) {
+                state.getZookeeperInfo();
-                    state.getZookeeperInfo();
-                }
 
                 int connections = state.getConnections();
                 int watches = state.getWatches();
@@ -91,7 +89,7 @@ public class ZookeeperMonitor extends AbstractZKClient {
                 float avgLatency = state.getAvgLatency();
                 float maxLatency = state.getMaxLatency();
                 int nodeCount = state.getNodeCount();
-                int status = ok ? 1 : 0;
+                int status = state.isHealthFlag() ? -1: (ok ? 1 : 0);
                 Date date = new Date();
 
                 ZookeeperRecord zookeeperRecord = new ZookeeperRecord(zookeeperServer,connections,watches,sent,received,mode,minLatency,avgLatency,maxLatency,nodeCount,status,date);

dolphinscheduler-ui/src/js/conf/home/pages/monitor/pages/servers/_source/zookeeperList.vue

@@ -59,7 +59,7 @@
             <span>{{$t('Node self-test status')}}</span>
           </th>
         </tr>
-        <tr v-for="(item, $index) in list" :key="$index">
+        <tr v-for="(item, $index) in list" :key="$index" @mouseenter="_showErrorMessage(item)">
           <td>
             <span>{{$index + 1}}</span>
           </td>
@@ -97,7 +97,8 @@
           </td>
           <td>
             <span class="state">
-              <em class="ans-icon-success-solid success" v-if="item.state"></em>
+              <em class="ans-icon-success-solid success" v-if="item.state === 1"></em>
+              <em class="ans-icon-warn-solid warn" v-else-if="item.state === 0"></em>
               <em class="ans-icon-fail-solid error" v-else></em>
             </span>
           </td>
@@ -107,6 +108,10 @@
   </div>
 </template>
 <script>
+
+  import _ from 'lodash'
+  import i18n from '@/module/i18n'
+
   export default {
     name: 'zookeeper-list',
     data () {
@@ -116,6 +121,15 @@
     },
     props: {
       list: Array
+    },
+    methods:{
+      _showErrorMessage:_.debounce(function (item){
+        const hostname = item.hostname
+        const state = item.state
+        if(state === -1){
+          this.$message.error(`${i18n.$t('Can not connect to zookeeper server:')}`+hostname)
+        }
+      },600)
     }
   }
 </script>
@@ -131,6 +145,9 @@
       .success {
         color: #33cc00;
       }
+      .warn {
+              color: #fabc05;
+      }
       .error {
         color: #ff0000;
       }

dolphinscheduler-ui/src/js/module/i18n/locale/en_US.js

@@ -506,6 +506,7 @@ export default {
   'Query time': 'Query time',
   'Node self-test status': 'Node self-test status',
   'Health status': 'Health status',
+  'Can not connect to zookeeper server:': 'Can not connect to zookeeper server:',
   'Max connections': 'Max connections',
   'Threads connections': 'Threads connections',
   'Max used connections': 'Max used connections',

dolphinscheduler-ui/src/js/module/i18n/locale/zh_CN.js

@@ -506,6 +506,7 @@ export default {
   'Query time': '当前查询时间',
   'Node self-test status': '节点自检状态',
   'Health status': '健康状态',
+  'Can not connect to zookeeper server:': 'zk节点连接失败:',
   'Max connections': '最大连接数',
   'Threads connections': '当前连接数',
   'Max used connections': '同时使用连接最大数',