feat: taskControl的loop循环太久导致Consumer线程池被占满问题 #11352

TencentBlueKing · Jan 7, 2025 · 1fe5cd4 · 1fe5cd4
1 parent 4eb97e9
commit 1fe5cd4
Show file tree

Hide file tree

Showing 3 changed files with 12 additions and 5 deletions.
diff --git a/...-base/src/main/kotlin/com/tencent/devops/process/engine/control/VmOperateTaskGenerator.kt b/...-base/src/main/kotlin/com/tencent/devops/process/engine/control/VmOperateTaskGenerator.kt
@@ -92,26 +92,30 @@ class VmOperateTaskGenerator {
         val taskName: String
         val taskAtom: String
         var timeout: Long? = null
+        var timeoutVar: String? = null
         if (container is VMBuildContainer) {
             val buildType = container.dispatchType?.buildType()?.name ?: BuildType.DOCKER.name
             val baseOS = container.baseOS.name
             atomCode = "$START_VM_TASK_ATOM-$buildType-$baseOS"
             taskType = EnvControlTaskType.VM.name
             taskName = "Prepare_Job#${container.id!!}"
             taskAtom = START_VM_TASK_ATOM
-            timeout = container.jobControlOption?.prepareTimeout?.toLong()
+            timeout = container.jobControlOption?.timeout?.toLong()
+            timeoutVar = container.jobControlOption?.timeoutVar
         } else {
             atomCode = START_NORMAL_TASK_ATOM
             taskType = EnvControlTaskType.NORMAL.name
             taskName = "Prepare_Job#${container.id!!}(N)"
             taskAtom = START_NORMAL_TASK_ATOM
             if (container is NormalContainer) {
-                timeout = container.jobControlOption?.prepareTimeout?.toLong()
+                timeout = container.jobControlOption?.timeout?.toLong()
+                timeoutVar = container.jobControlOption?.timeoutVar
             }
         }
         val additionalOptions = ElementAdditionalOptions(
             runCondition = RunCondition.PRE_TASK_FAILED_BUT_CANCEL,
-            timeout = timeout
+            timeout = timeout,
+            timeoutVar = timeoutVar
         )
         return PipelineBuildTask(
             projectId = projectId,

diff --git a/...ne/src/main/kotlin/com/tencent/devops/process/engine/atom/vm/DispatchVMStartupTaskAtom.kt b/...ne/src/main/kotlin/com/tencent/devops/process/engine/atom/vm/DispatchVMStartupTaskAtom.kt
@@ -230,7 +230,7 @@ class DispatchVMStartupTaskAtom @Autowired constructor(
         )
         dispatch(task, pipelineInfo!!, param, vmNames, container!!, ignoreEnvAgentIds, pipelineAuthorizer)
         logger.info("[$buildId]|STARTUP_VM|VM=${param.baseOS}-$vmNames($vmSeqId)|Dispatch startup")
-        return AtomResponse(BuildStatus.CALL_WAITING)
+        return AtomResponse(BuildStatus.RUNNING)
     }
 
     private fun dispatch(

diff --git a/...ocess/biz-engine/src/main/kotlin/com/tencent/devops/process/engine/control/TaskControl.kt b/...ocess/biz-engine/src/main/kotlin/com/tencent/devops/process/engine/control/TaskControl.kt
@@ -40,6 +40,7 @@ import com.tencent.devops.common.event.dispatcher.pipeline.PipelineEventDispatch
 import com.tencent.devops.process.engine.atom.AtomResponse
 import com.tencent.devops.process.engine.atom.TaskAtomService
 import com.tencent.devops.process.engine.common.BS_ATOM_STATUS_REFRESH_DELAY_MILLS
+import com.tencent.devops.process.engine.common.VMUtils.getStartVmLabel
 import com.tencent.devops.process.engine.control.lock.ContainerIdLock
 import com.tencent.devops.process.engine.pojo.PipelineBuildTask
 import com.tencent.devops.process.engine.pojo.event.PipelineBuildAtomTaskEvent
@@ -175,7 +176,9 @@ class TaskControl @Autowired constructor(
 
             if (buildStatus.isRunning()) { // 仍然在运行中--没有结束的
                 // 如果是要轮循的才需要定时消息轮循
-                loopDispatch(buildTask = buildTask)
+                if (!taskId.startsWith(getStartVmLabel())) {
+                    loopDispatch(buildTask = buildTask)
+                }
             } else {
                 pipelineBuildTaskService.finishTask(
                     buildTask = buildTask,