BIGTOP-4146: Reorganize task log files

apache · Jun 20, 2024 · 7500084 · 7500084
1 parent 355e305
commit 7500084
Show file tree

Hide file tree

Showing 34 changed files with 121 additions and 470 deletions.
diff --git a/.gitignore b/.gitignore
@@ -40,6 +40,7 @@ build/
 
 ### Log ###
 logs
+tasklogs
 *.log
 npm-debug.log*
 yarn-debug.log*

diff --git a/...agent/src/main/java/org/apache/bigtop/manager/agent/executor/AbstractCommandExecutor.java b/...agent/src/main/java/org/apache/bigtop/manager/agent/executor/AbstractCommandExecutor.java
@@ -20,7 +20,6 @@
 
 import org.apache.bigtop.manager.agent.holder.SpringContextHolder;
 import org.apache.bigtop.manager.common.constants.MessageConstants;
-import org.apache.bigtop.manager.common.message.entity.command.CommandLogMessage;
 import org.apache.bigtop.manager.common.message.entity.command.CommandRequestMessage;
 import org.apache.bigtop.manager.common.message.entity.command.CommandResponseMessage;
 import org.apache.bigtop.manager.common.shell.ShellResult;
@@ -68,15 +67,4 @@ protected void doExecuteOnDevMode() {
     }
 
     protected abstract void doExecute();
-
-    protected void writeBackCommandLog(String log) {
-        CommandLogMessage logMessage = new CommandLogMessage();
-        logMessage.setLog(log);
-        logMessage.setMessageId(commandRequestMessage.getMessageId());
-        logMessage.setHostname(commandRequestMessage.getHostname());
-        logMessage.setTaskId(commandRequestMessage.getTaskId());
-        logMessage.setStageId(commandRequestMessage.getStageId());
-        logMessage.setJobId(commandRequestMessage.getJobId());
-        SpringContextHolder.getAgentWebSocket().sendMessage(logMessage);
-    }
 }
diff --git a/...gent/src/main/java/org/apache/bigtop/manager/agent/executor/ComponentCommandExecutor.java b/...gent/src/main/java/org/apache/bigtop/manager/agent/executor/ComponentCommandExecutor.java
@@ -49,7 +49,7 @@ public void doExecute() {
         CommandPayload commandPayload =
                 JsonUtils.readFromString(commandRequestMessage.getMessagePayload(), CommandPayload.class);
         log.info("[agent executeTask] taskEvent is: {}", commandRequestMessage);
-        ShellResult shellResult = StackExecutor.execute(commandPayload, this::writeBackCommandLog);
+        ShellResult shellResult = StackExecutor.execute(commandPayload);
 
         commandResponseMessage.setCode(shellResult.getExitCode());
         commandResponseMessage.setResult(shellResult.getResult());

diff --git a/...gent/src/main/java/org/apache/bigtop/manager/agent/scheduler/DefaultCommandScheduler.java b/...gent/src/main/java/org/apache/bigtop/manager/agent/scheduler/DefaultCommandScheduler.java
@@ -22,6 +22,7 @@
 import org.apache.bigtop.manager.agent.executor.CommandExecutors;
 import org.apache.bigtop.manager.common.message.entity.command.CommandRequestMessage;
 
+import org.slf4j.MDC;
 import org.springframework.stereotype.Component;
 
 import lombok.extern.slf4j.Slf4j;
@@ -45,10 +46,13 @@ public class DefaultCommandScheduler implements CommandScheduler {
     public void submit(CommandRequestMessage message) {
         queue.offer(() -> {
             try {
+                MDC.put("taskId", message.getTaskId().toString());
                 CommandExecutor commandExecutor = CommandExecutors.getCommandExecutor(message.getCommandMessageType());
                 commandExecutor.execute(message);
             } catch (Exception e) {
                 log.error("Error when running command", e);
+            } finally {
+                MDC.clear();
             }
         });
     }

diff --git a/bigtop-manager-agent/src/main/resources/logback-spring.xml b/bigtop-manager-agent/src/main/resources/logback-spring.xml
@@ -19,76 +19,58 @@
 -->
 
 <configuration scan="true" scanPeriod="120 seconds">
-    <property name="log.base" value="logs"/>
-    <property scope="context" name="log.base.ctx" value="${log.base}"/>
+    <property name="log.base" value="logs" />
+    <property scope="context" name="log.base.ctx" value="${log.base}" />
 
     <appender name="STDOUT" class="ch.qos.logback.core.ConsoleAppender">
         <encoder>
-            <pattern>[%-5level] %date{yyyy-MM-dd HH:mm:ss.SSS Z} [%thread] %logger{96}:[%line] - %msg%n</pattern>
+            <pattern>[%-5level] %date{yyyy-MM-dd HH:mm:ss.SSS Z} [%thread] %logger{1}:[%line] - %m%n</pattern>
             <charset>UTF-8</charset>
         </encoder>
     </appender>
 
-    <appender name="WSLOGFILE" class="ch.qos.logback.core.rolling.RollingFileAppender">
-        <file>${log.base}/ws.log</file>
-        <rollingPolicy class="ch.qos.logback.core.rolling.SizeAndTimeBasedRollingPolicy">
-            <fileNamePattern>${log.base}/ws.%d{yyyy-MM-dd}.%i.log</fileNamePattern>
-            <maxHistory>30</maxHistory>
-            <maxFileSize>200MB</maxFileSize>
-            <totalSizeCap>10GB</totalSizeCap>
-            <cleanHistoryOnStart>true</cleanHistoryOnStart>
-        </rollingPolicy>
-        <encoder>
-            <pattern>[%-5level] %date{yyyy-MM-dd HH:mm:ss.SSS Z} [%thread] %logger{96}:[%line] - %msg%n</pattern>
-            <charset>UTF-8</charset>
-        </encoder>
-    </appender>
-    <appender name="STACKLOGFILE" class="ch.qos.logback.core.rolling.RollingFileAppender">
-        <file>${log.base}/bigtop-manager-stack.log</file>
+    <appender name="AGENT_LOG_FILE" class="ch.qos.logback.core.rolling.RollingFileAppender">
+        <file>${log.base}/bigtop-manager-agent.log</file>
         <rollingPolicy class="ch.qos.logback.core.rolling.SizeAndTimeBasedRollingPolicy">
-            <fileNamePattern>${log.base}/bigtop-manager-stack.%d{yyyy-MM-dd}.%i.log</fileNamePattern>
+            <fileNamePattern>${log.base}/bigtop-manager-agent.%d{yyyy-MM-dd}.%i.log</fileNamePattern>
             <maxHistory>30</maxHistory>
             <maxFileSize>200MB</maxFileSize>
             <totalSizeCap>10GB</totalSizeCap>
             <cleanHistoryOnStart>true</cleanHistoryOnStart>
         </rollingPolicy>
         <encoder>
-            <pattern>[%-5level] %date{yyyy-MM-dd HH:mm:ss.SSS Z} [%thread] %logger{96}:[%line] - %msg%n</pattern>
+            <pattern>[%-5level] %date{yyyy-MM-dd HH:mm:ss.SSS Z} [%thread] %logger{1}:[%line] - %m%n</pattern>
             <charset>UTF-8</charset>
         </encoder>
     </appender>
-    <appender name="AGENTLOGFILE" class="ch.qos.logback.core.rolling.RollingFileAppender">
-        <file>${log.base}/bigtop-manager-agent.log</file>
+
+    <appender name="TASK_LOG_FILE" class="ch.qos.logback.classic.sift.SiftingAppender">
         <filter class="ch.qos.logback.core.filter.EvaluatorFilter">
             <evaluator>
-                <expression>logger.contains("org.apache.bigtop.manager.stack")</expression>
+                <expression>mdc.get("taskId") != null</expression>
             </evaluator>
-            <onMatch>DENY</onMatch>
-            <onMismatch>ACCEPT</onMismatch>
+            <onMatch>ACCEPT</onMatch>
+            <onMismatch>DENY</onMismatch>
         </filter>
-        <rollingPolicy class="ch.qos.logback.core.rolling.SizeAndTimeBasedRollingPolicy">
-            <fileNamePattern>${log.base}/bigtop-manager-agent.%d{yyyy-MM-dd}.%i.log</fileNamePattern>
-            <maxHistory>30</maxHistory>
-            <maxFileSize>200MB</maxFileSize>
-            <totalSizeCap>10GB</totalSizeCap>
-            <cleanHistoryOnStart>true</cleanHistoryOnStart>
-        </rollingPolicy>
-        <encoder>
-            <pattern>[%-5level] %date{yyyy-MM-dd HH:mm:ss.SSS Z} [%thread] %logger{96}:[%line] - %msg%n</pattern>
-            <charset>UTF-8</charset>
-        </encoder>
+        <discriminator class="ch.qos.logback.classic.sift.MDCBasedDiscriminator">
+            <key>taskId</key>
+            <defaultValue>0</defaultValue>
+        </discriminator>
+        <sift>
+            <appender name="task-${taskId}" class="ch.qos.logback.core.FileAppender">
+                <file>tasklogs/task-${taskId}.log</file>
+                <encoder>
+                    <pattern>[%-5level] %date{yyyy-MM-dd HH:mm:ss.SSS Z} - %m%n</pattern>
+                    <charset>UTF-8</charset>
+                </encoder>
+                <append>true</append>
+            </appender>
+        </sift>
     </appender>
 
-    <logger name="org.apache.bigtop.manager.agent.ws.AgentWebSocketHandler" level="INFO" additivity="true">
-        <appender-ref ref="WSLOGFILE"/>
-    </logger>
-
-    <logger name="org.apache.bigtop.manager.stack" level="INFO" additivity="true">
-        <appender-ref ref="STACKLOGFILE"/>
-    </logger>
-
     <root level="INFO">
-        <appender-ref ref="STDOUT"/>
-        <appender-ref ref="AGENTLOGFILE"/>
+        <appender-ref ref="STDOUT" />
+        <appender-ref ref="AGENT_LOG_FILE" />
+        <appender-ref ref="TASK_LOG_FILE" />
     </root>
 </configuration>
diff --git a/.../main/java/org/apache/bigtop/manager/common/message/entity/command/CommandLogMessage.java b/.../main/java/org/apache/bigtop/manager/common/message/entity/command/CommandLogMessage.java
diff --git a/...mon/src/main/java/org/apache/bigtop/manager/common/message/serializer/KryoPoolHolder.java b/...mon/src/main/java/org/apache/bigtop/manager/common/message/serializer/KryoPoolHolder.java
@@ -23,7 +23,6 @@
 import org.apache.bigtop.manager.common.message.entity.BaseRequestMessage;
 import org.apache.bigtop.manager.common.message.entity.BaseResponseMessage;
 import org.apache.bigtop.manager.common.message.entity.HeartbeatMessage;
-import org.apache.bigtop.manager.common.message.entity.command.CommandLogMessage;
 import org.apache.bigtop.manager.common.message.entity.command.CommandMessageType;
 import org.apache.bigtop.manager.common.message.entity.command.CommandRequestMessage;
 import org.apache.bigtop.manager.common.message.entity.command.CommandResponseMessage;
@@ -62,7 +61,6 @@ protected Kryo create() {
             kryo.register(HeartbeatMessage.class);
             kryo.register(CommandResponseMessage.class);
             kryo.register(CommandRequestMessage.class);
-            kryo.register(CommandLogMessage.class);
 
             // message pojo
             kryo.register(HostInfo.class);

diff --git a/bigtop-manager-dao/src/main/java/org/apache/bigtop/manager/dao/entity/CommandLog.java b/bigtop-manager-dao/src/main/java/org/apache/bigtop/manager/dao/entity/CommandLog.java
diff --git a/...ager-dao/src/main/java/org/apache/bigtop/manager/dao/repository/CommandLogRepository.java b/...ager-dao/src/main/java/org/apache/bigtop/manager/dao/repository/CommandLogRepository.java
diff --git a/.../main/java/org/apache/bigtop/manager/server/command/stage/runner/AbstractStageRunner.java b/.../main/java/org/apache/bigtop/manager/server/command/stage/runner/AbstractStageRunner.java
@@ -89,8 +89,10 @@ public void run() {
                 boolean taskSuccess = res != null && res.getCode() == MessageConstants.SUCCESS_CODE;
 
                 if (taskSuccess) {
+                    commandLogService.onLogReceived(task.getId(), task.getHostname(), "Success!");
                     onTaskSuccess(task);
                 } else {
+                    commandLogService.onLogReceived(task.getId(), task.getHostname(), "Failed!");
                     onTaskFailure(task);
                 }