Merge pull request #10797 from tangruotian/issue-10795

bugfix: Agent复用同级节点时跳过了复用锁 #10795
TencentBlueKing · Aug 13, 2024 · 7b17c53 · 7b17c53
2 parents e157675 + 072389a
commit 7b17c53
Show file tree

Hide file tree

Showing 3 changed files with 34 additions and 38 deletions.
diff --git a/src/backend/ci/core/common/common-pipeline-yaml/src/main/resources/schema/V3_0/ci.json b/src/backend/ci/core/common/common-pipeline-yaml/src/main/resources/schema/V3_0/ci.json
@@ -255,6 +255,9 @@
                         "type" : "string"
                       }
                     }
+                  },
+                  "skip-wip" : {
+                    "type" : "boolean"
                   }
                 }
               } ]
@@ -943,6 +946,9 @@
                       "type" : "string"
                     }
                   }
+                },
+                "skip-wip" : {
+                  "type" : "boolean"
                 }
               }
             } ]
@@ -2226,6 +2232,12 @@
                 "type" : "string"
               }
             },
+            "chat-id" : {
+              "type" : "array",
+              "items" : {
+                "type" : "string"
+              }
+            },
             "notify-groups" : {
               "type" : "array",
               "items" : {

diff --git a/...dispatch/src/main/kotlin/com/tencent/devops/dispatch/service/ThirdPartyDispatchService.kt b/...dispatch/src/main/kotlin/com/tencent/devops/dispatch/service/ThirdPartyDispatchService.kt
@@ -74,7 +74,6 @@ import com.tencent.devops.environment.pojo.thirdpartyagent.EnvNodeAgent
 import com.tencent.devops.environment.pojo.thirdpartyagent.ThirdPartyAgent
 import com.tencent.devops.process.api.service.ServiceBuildResource
 import com.tencent.devops.process.api.service.ServiceVarResource
-import com.tencent.devops.process.constant.ProcessMessageCode
 import com.tencent.devops.process.engine.common.VMUtils
 import com.tencent.devops.process.pojo.SetContextVarData
 import com.tencent.devops.process.pojo.VmInfo
@@ -435,22 +434,12 @@ class ThirdPartyDispatchService @Autowired constructor(
                 )
             }/console/pipeline/${dispatchMessage.event.projectId}/$pipelineId/detail/$lockedBuildId"
             if (lockedBuildId != dispatchMessage.event.buildId) {
-                I18nUtil.getCodeLanMessage(
-                    messageCode = ProcessMessageCode.BK_LOCKED,
-                    language = I18nUtil.getDefaultLocaleLanguage()
-                ) + ": $linkTip<a target='_blank' href='$link'>" +
-                        I18nUtil.getCodeLanMessage(
-                            messageCode = ProcessMessageCode.BK_CLICK,
-                            language = I18nUtil.getDefaultLocaleLanguage()
-                        ) + "</a>"
+                "$linkTip<a target='_blank' href='$link'>$lockedBuildId</a>"
             } else {
-                I18nUtil.getCodeLanMessage(
-                    messageCode = ProcessMessageCode.BK_CURRENT,
-                    language = I18nUtil.getDefaultLocaleLanguage()
-                ) + ": $linkTip"
+                linkTip
             }
         } ?: ""
-        log(
+        logWarn(
             dispatchMessage.event,
             I18nUtil.getCodeLanMessage(
                 messageCode = AGENT_REUSE_MUTEX_REDISPATCH,

diff --git a/...in/com/tencent/devops/process/engine/control/command/container/impl/AgentReuseMutexCmd.kt b/...in/com/tencent/devops/process/engine/control/command/container/impl/AgentReuseMutexCmd.kt
@@ -44,8 +44,8 @@ class AgentReuseMutexCmd @Autowired constructor(
 ) : ContainerCmd {
     override fun canExecute(commandContext: ContainerContext): Boolean {
         return commandContext.cmdFlowState == CmdFlowState.CONTINUE &&
-            !commandContext.buildStatus.isFinish() &&
-            commandContext.container.controlOption.agentReuseMutex != null
+                !commandContext.buildStatus.isFinish() &&
+                commandContext.container.controlOption.agentReuseMutex != null
     }
 
     override fun execute(commandContext: ContainerContext) {
@@ -59,22 +59,15 @@ class AgentReuseMutexCmd @Autowired constructor(
      * 互斥情况存在三种
      * 1、依赖某个AgentId，直接往下执行即可
      * 2、Root节点，即没有复用节点的节点，根据类型先拿取锁
-     * 3、Reuse节点，但没有复用节点，可能是因为和复用节点同级，或存在和复用节点先后顺序不明确的，
-     * 这种先拿取复用的JobId看有没有，没有就按root节点的逻辑走
+     * 3、Reuse节点，但没有复用节点，可能是因为和复用节点同级，或存在和复用节点先后顺序不明确的，就按root节点的逻辑走
      */
     private fun doExecute(commandContext: ContainerContext) {
         var mutex = commandContext.container.controlOption.agentReuseMutex!!
+        // 复用的不用参与锁逻辑
         if (!mutex.type.needEngineLock()) {
             commandContext.cmdFlowState = CmdFlowState.CONTINUE
             return
         }
-        // 如果有依赖Job且不是依赖类型的可以先拿一下上下文看看有没有已经写入了，如果写入了可以直接跳过
-        if (mutex.reUseJobId != null &&
-            commandContext.variables.containsKey(AgentReuseMutex.genAgentContextKey(mutex.reUseJobId!!))
-        ) {
-            commandContext.cmdFlowState = CmdFlowState.CONTINUE
-            return
-        }
         // 极端情况上下文没有写入，且agentId还是空，理论上不会有，逻辑上出现了就失败
         if (mutex.agentOrEnvId.isNullOrBlank()) {
             return agentIdNullError(commandContext, mutex, null)
@@ -85,7 +78,7 @@ class AgentReuseMutexCmd @Autowired constructor(
             AgentReuseMutexType.AGENT_ID, AgentReuseMutexType.AGENT_NAME -> {
                 acquireMutex(commandContext, mutex)
             }
-
+            // 环境的因为无法确定节点不参与锁
             else -> {
                 commandContext.cmdFlowState = CmdFlowState.CONTINUE
             }
@@ -165,16 +158,18 @@ class AgentReuseMutexCmd @Autowired constructor(
                     commandContext.cmdFlowState = CmdFlowState.LOOP
                 }
 
-                else -> { // 正常运行
-                    commandContext.cmdFlowState = CmdFlowState.CONTINUE // 检查通过，继续向下执行
+                // 正常运行
+                else -> {
+                    // 检查通过，继续向下执行
+                    commandContext.cmdFlowState = CmdFlowState.CONTINUE
                 }
             }
         } else if (commandContext.container.status.isFinish()) { // 对于存在重放的结束消息做闭环
             val event = commandContext.event
 
             LOG.info(
                 "AGENT_REUSE|ENGINE|${event.buildId}|${event.source}|status=${commandContext.container.status}" +
-                    "|concurrent_container_event"
+                        "|concurrent_container_event"
             )
 
             releaseContainerMutex(
@@ -310,7 +305,7 @@ class AgentReuseMutexCmd @Autowired constructor(
             containerVar = emptyMap(), buildStatus = null,
             timestamps = mapOf(
                 BuildTimestampType.JOB_AGENT_REUSE_MUTEX_QUEUE to
-                    BuildRecordTimeStamp(null, LocalDateTime.now().timestampmilli())
+                        BuildRecordTimeStamp(null, LocalDateTime.now().timestampmilli())
             )
         )
     }
@@ -347,9 +342,9 @@ class AgentReuseMutexCmd @Autowired constructor(
             // 排队等待时间为0的时候，立即超时, 退出队列，并失败, 没有就继续在队列中,timeOut时间为分钟
             if (mutex.timeout == 0 || timeDiff > TimeUnit.MINUTES.toSeconds(mutex.timeout.toLong())) {
                 val desc = "${
-                if (mutex.timeoutVar.isNullOrBlank()) {
-                    "[${mutex.timeout} minutes]"
-                } else " timeoutVar[${mutex.timeoutVar}] setup to [${mutex.timeout} minutes]"
+                    if (mutex.timeoutVar.isNullOrBlank()) {
+                        "[${mutex.timeout} minutes]"
+                    } else " timeoutVar[${mutex.timeoutVar}] setup to [${mutex.timeout} minutes]"
                 } "
                 logAgentMutex(
                     container, mutex, lockedBuildId,
@@ -443,7 +438,7 @@ class AgentReuseMutexCmd @Autowired constructor(
             containerVar = emptyMap(), buildStatus = null,
             timestamps = mapOf(
                 BuildTimestampType.JOB_MUTEX_QUEUE to
-                    BuildRecordTimeStamp(LocalDateTime.now().timestampmilli(), null)
+                        BuildRecordTimeStamp(LocalDateTime.now().timestampmilli(), null)
             )
         )
     }
@@ -514,10 +509,10 @@ class AgentReuseMutexCmd @Autowired constructor(
                         messageCode = ProcessMessageCode.BK_LOCKED,
                         language = I18nUtil.getDefaultLocaleLanguage()
                     ) + ": $linkTip<a target='_blank' href='$link'>" +
-                        I18nUtil.getCodeLanMessage(
-                            messageCode = ProcessMessageCode.BK_CLICK,
-                            language = I18nUtil.getDefaultLocaleLanguage()
-                        ) + "</a> | $msg"
+                            I18nUtil.getCodeLanMessage(
+                                messageCode = ProcessMessageCode.BK_CLICK,
+                                language = I18nUtil.getDefaultLocaleLanguage()
+                            ) + "</a> | $msg"
                 } else {
                     I18nUtil.getCodeLanMessage(
                         messageCode = ProcessMessageCode.BK_CURRENT,