feat: task失败可以立即重试 TencentBlueKing#6400

fitzcao · Mar 28, 2022 · e5006d2 · e5006d2
1 parent 7062938
commit e5006d2
Show file tree

Hide file tree

Showing 5 changed files with 13 additions and 27 deletions.
diff --git a/...on/common-pipeline/src/main/kotlin/com/tencent/devops/common/pipeline/utils/ModelUtils.kt b/...on/common-pipeline/src/main/kotlin/com/tencent/devops/common/pipeline/utils/ModelUtils.kt
@@ -117,10 +117,11 @@ object ModelUtils {
                     val jobStatus = BuildStatus.parse(c.status)
                     c.canRetry = jobStatus.isFailure() || jobStatus.isCancel()
                 }
-
-                val failElements = mutableListOf<Element>()
-                c.elements.forEach { e ->
-                    refreshElement(element = e, failElements = failElements)
+                if (c.canRetry == true) {
+                    val failElements = mutableListOf<Element>()
+                    c.elements.forEach { e ->
+                        refreshElement(element = e, failElements = failElements)
+                    }
                 }
             }
         }

diff --git a/...ess/api-process/src/main/kotlin/com/tencent/devops/process/constant/ProcessMessageCode.kt b/...ess/api-process/src/main/kotlin/com/tencent/devops/process/constant/ProcessMessageCode.kt
@@ -143,7 +143,7 @@ object ProcessMessageCode {
     const val ERROR_TIMEOUT_IN_BUILD_QUEUE = "2101904" //  排队超时，取消运行! [{0}]
     const val ERROR_PARUS_PIEPLINE_IS_RUNNINT = "2101905" // 暂停的流水线已开始运行
     const val ERROR_ELEMENT_TOO_LONG = "2101906" // {0} element大小越界
-    const val ERROR_STOP_VM_UN_FINISH = "2101907" // stopVm还未完成，请稍后重试
+    const val ERROR_JOB_RUNNING = "2101907" // job非完成态，不能进行重试
 
     const val ERROR_NO_BUILD_EXISTS_BY_ID = "2101100" // 流水线构建[{0}]不存在
     const val ERROR_NO_PIPELINE_EXISTS_BY_ID = "2101101" // 流水线[{0}]不存在

diff --git a/...process/biz-base/src/main/kotlin/com/tencent/devops/process/engine/dao/PipelineInfoDao.kt b/...process/biz-base/src/main/kotlin/com/tencent/devops/process/engine/dao/PipelineInfoDao.kt
@@ -555,7 +555,7 @@ class PipelineInfoDao {
     fun listByProject(dslContext: DSLContext, projectCode: String): Result<Record2<String, Long>> {
         return with(T_PIPELINE_INFO) {
             dslContext.select(PIPELINE_ID.`as`("pipelineId"), ID.`as`("id")).from(this)
-                .where(PROJECT_ID.eq(projectCode)).fetch()
+                .where(PROJECT_ID.eq(projectCode).and(DELETE.eq(false))).fetch()
         }
     }
 

diff --git a/...s/src/main/kotlin/com/tencent/devops/process/service/builds/PipelineRetryFacadeService.kt b/...s/src/main/kotlin/com/tencent/devops/process/service/builds/PipelineRetryFacadeService.kt
@@ -33,6 +33,7 @@ import com.tencent.devops.common.event.dispatcher.pipeline.PipelineEventDispatch
 import com.tencent.devops.common.event.enums.ActionType
 import com.tencent.devops.common.log.utils.BuildLogPrinter
 import com.tencent.devops.common.pipeline.enums.BuildStatus
+import com.tencent.devops.common.service.utils.MessageCodeUtil
 import com.tencent.devops.process.constant.ProcessMessageCode
 import com.tencent.devops.process.engine.common.VMUtils
 import com.tencent.devops.process.engine.pojo.PipelineBuildContainer
@@ -122,33 +123,17 @@ class PipelineRetryFacadeService @Autowired constructor(
         )
     }
 
-    // 若当前job的关机任务还未完成，报错让用户稍后再重试。否则会因为开机比关机早到引发引擎调度问题
+    // 若当前job状态还未完成，报错让用户稍后再重试。否则会因为开机比关机早到引发引擎调度问题
     private fun checkStopTask(
         projectId: String,
         buildId: String,
         containerInfo: PipelineBuildContainer
     ) {
-        var jobFinish = true
         if (!containerInfo.status.isFinish()) {
-            val runningTasks = pipelineTaskService.listContainerBuildTasks(
-                projectId = projectId,
-                buildId = buildId,
-                containerSeqId = containerInfo.containerId,
-                buildStatusSet = setOf(BuildStatus.RUNNING)
-            )
-            runningTasks.forEach {
-                if (it.taskId.startsWith(VMUtils.getStopVmLabel()) || it.taskId.startsWith(VMUtils.getEndLabel())) {
-                    jobFinish = false
-                    return@forEach
-                }
-            }
-        }
-
-        if (!jobFinish) {
-            logger.warn("retry runningJob: $projectId|$buildId stopVm not finish")
+            logger.warn("retry runningJob: $projectId|$buildId｜${containerInfo.containerId} is running")
             throw ErrorCodeException(
-                errorCode = ProcessMessageCode.ERROR_STOP_VM_UN_FINISH,
-                defaultMessage = "$buildId｜${containerInfo.containerId} stopVm未完成,请稍后重试"
+                errorCode = ProcessMessageCode.ERROR_JOB_RUNNING,
+                defaultMessage = MessageCodeUtil.getCodeLanMessage(ProcessMessageCode.ERROR_JOB_RUNNING)
             )
         }
     }

diff --git a/support-files/sql/5001_ci_project-init_dml_mysql.sql b/support-files/sql/5001_ci_project-init_dml_mysql.sql
@@ -171,7 +171,7 @@ REPLACE INTO `T_MESSAGE_CODE_DETAIL` (`ID`, `MESSAGE_CODE`, `MODULE_CODE`, `MESS
 REPLACE INTO `T_MESSAGE_CODE_DETAIL` (`ID`, `MESSAGE_CODE`, `MODULE_CODE`, `MESSAGE_DETAIL_ZH_CN`, `MESSAGE_DETAIL_EN`) VALUES  ('b6d9420993954802884f474828cc0db1', '2101904', '01', '流水线：排队超时，取消运行! [{0}]', 'Pipeline:  Queue timed out, canceled running! [{0}]');
 REPLACE INTO `T_MESSAGE_CODE_DETAIL` (`ID`, `MESSAGE_CODE`, `MODULE_CODE`, `MESSAGE_DETAIL_ZH_CN`, `MESSAGE_DETAIL_EN`) VALUES  ('7d8f2d20b939448f91550967d5ff63b6', '2101905', '01', '流水线：暂停的流水线已开始运行', 'Pipeline:  Pipeline: puase pipeline running now');
 REPLACE INTO `T_MESSAGE_CODE_DETAIL` (`ID`, `MESSAGE_CODE`, `MODULE_CODE`, `MESSAGE_DETAIL_ZH_CN`, `MESSAGE_DETAIL_EN`) VALUES  ('4e6d9a9c3c874c16b2fcff32b6a731fd', '2101906', '01', '流水线：{0} element大小越界', 'Pipeline:  Pipeline:  {0} element size too long');
-REPLACE INTO `T_MESSAGE_CODE_DETAIL` (`ID`, `MESSAGE_CODE`, `MODULE_CODE`, `MESSAGE_DETAIL_ZH_CN`, `MESSAGE_DETAIL_EN`) VALUES  ('894102da867c431baf9fca6bfd6bc640', '2101907', '01', '流水线: stopVm还未完成，请稍后重试', 'Pipeline:  stopVm running, please wait...');
+REPLACE INTO `T_MESSAGE_CODE_DETAIL` (`ID`, `MESSAGE_CODE`, `MODULE_CODE`, `MESSAGE_DETAIL_ZH_CN`, `MESSAGE_DETAIL_EN`) VALUES  ('894102da867c431baf9fca6bfd6bc640', '2101907', '01', '流水线: job非完成态，不能进行重试', 'Pipeline:  job is running, forbid retry');
 
 
 REPLACE INTO `T_MESSAGE_CODE_DETAIL` (`ID`, `MESSAGE_CODE`, `MODULE_CODE`, `MESSAGE_DETAIL_ZH_CN`, `MESSAGE_DETAIL_EN`) VALUES  ('a1d8bb8bc3e54b0daaf6d22daef17650', '2101908', '01', '流水线：质量红线(准入)配置有误：Fail to find quality gate intercept element', 'Pipeline: The quality red line (access) is incorrectly configured: Fail to find quality gate intercept element');