[[["易于理解","easyToUnderstand","thumb-up"],["解决了我的问题","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["很难理解","hardToUnderstand","thumb-down"],["信息或示例代码不正确","incorrectInformationOrSampleCode","thumb-down"],["没有我需要的信息/示例","missingTheInformationSamplesINeed","thumb-down"],["翻译问题","translationIssue","thumb-down"],["其他","otherDown","thumb-down"]],["最后更新时间 (UTC):2025-08-18。"],[[["\u003cp\u003eThe \u003cstrong\u003eExecution details\u003c/strong\u003e tab in Dataflow's monitoring UI helps optimize job performance and diagnose issues like slow or stuck pipelines without impacting VM performance.\u003c/p\u003e\n"],["\u003cp\u003eThis feature is useful for troubleshooting stuck or slow pipelines, optimizing performance, or gaining insight into a job's execution, providing insights through four views: \u003cstrong\u003eStage progress\u003c/strong\u003e, \u003cstrong\u003eStage info panel\u003c/strong\u003e, \u003cstrong\u003eStage workflow\u003c/strong\u003e, and \u003cstrong\u003eWorker progress\u003c/strong\u003e.\u003c/p\u003e\n"],["\u003cp\u003e\u003cstrong\u003eStage progress\u003c/strong\u003e view visually displays job execution stages over time for both batch and streaming jobs, including critical path analysis and data freshness insights, the longer a stage bar, the longer it took to complete.\u003c/p\u003e\n"],["\u003cp\u003e\u003cstrong\u003eStage info panel\u003c/strong\u003e details the steps within a fused stage, ranked by wall time, helping identify bottlenecks, and can be accessed by selecting a bar within the \u003cstrong\u003eStage progress\u003c/strong\u003e view.\u003c/p\u003e\n"],["\u003cp\u003eThe \u003cstrong\u003eWorker progress\u003c/strong\u003e view, exclusive to batch jobs, displays individual work items and CPU utilization for each worker, offering insights into resource usage and potential underutilization.\u003c/p\u003e\n"]]],[],null,["# Execution details\n\nThis page describes how to use the **Execution details** tab in the\nDataflow monitoring interface.\n\nOverview\n--------\n\nWhen Dataflow runs a job, it converts the *steps* of the pipeline\ninto *stages* . Whereas each step represents an individual transform, a stage\nrepresents a single unit of work that is performed by Dataflow.\nTo optimize the pipeline, Dataflow might\n[fuse](/dataflow/docs/pipeline-lifecycle#fusion_optimization) multiple steps\ninto one stage.\n\nThe **Execution details** tab in the Dataflow monitoring interface\ndisplays information about the stages of a job. You can use the\n**Execution details** tab to troubleshoot performance issues, such as:\n\n- Slow stages that cause performance bottlenecks\n- Stuck stages that are not advancing\n- Worker VMs that are lagging behind other workers\n\nView execution details\n----------------------\n\nTo view the execution details for a job, perform the following steps:\n\n1. In the Google Cloud console, go to the **Dataflow**\n \\\u003e **Jobs** page.\n\n [Go to Jobs](https://console.cloud.google.com/dataflow/jobs)\n2. Select a job.\n\n3. Click the **Execution details** tab.\n\n4. Select one of the following views:\n\n - **Stage progress**\n - **Stage workflow**\n - **Worker progress** (batch jobs only)\n\nThe following sections describe each of these views.\n\nStage progress view\n-------------------\n\nThe **Stage progress** view lets you observe the overall progress of the job and\ncompare relative progress between stages. The layout of the **Stage progress**\nview differs between batch jobs and streaming jobs.\n\n### Stage progress for batch jobs\n\nFor batch jobs, the **Stage progress** view shows the job stages in order of\ntheir start times. For each stage, it displays the following elements:\n\n- A bar that shows the stop and end times.\n- A line chart that shows the progress of the stage over time as a percentage of the stage's total work.\n- The total time spent in the stage.\n\nTo filter which stages are displayed, click **Filter stages** . To view the\ncritical path, toggle **Critical path**. The critical path is the sequence of\nstages that contribute to the overall job runtime. For example, it excludes\nbranches that finished earlier than the overall job, and inputs that did not\ndelay downstream processing.\n\nThe **Stage Info** panel shows more detailed information about a stage. To view\nthe details for a stage, click the progress bar for that stage. The **Stage\nInfo** panel shows the following information about a stage:\n\n- Status\n- Progress as a percentage\n- Start and end times\n- The pipeline steps that this stage encompasses\n- The slowest steps by wall time\n- Details about any [stragglers](/dataflow/docs/guides/troubleshoot-batch-stragglers)\n\nIf the panel is not visible, click last_page\n**Toggle panel \"Stage info\"**.\n\n### Stage progress for streaming jobs\n\nFor streaming jobs, the **Stage progress** view has two visualizations of\ndata freshness.\n[Data freshness](/dataflow/docs/guides/using-monitoring-intf#data_freshness_streaming)\nis the difference between a data element's timestamp and the time when the\nelement is processed. Larger values mean the pipeline is taking longer to\nprocess the input data.\n\nThe first visualization shows data freshness per stage as a line graph. To see\nthe data freshness at a specific instant of time, hold the pointer over the\ngraph. To select the time range, use the time picker or click the graph and drag\nto select the range. To filter which stages are displayed, click **Filter\nstages**.\n\nThe graph also highlights anomalies in the data:\n\n- Potential slowness: Data freshness exceeds the 95th percentile for the selected time window.\n- Potential stuckness: Data freshness exceeds the 99th percentile for the selected time window.\n\nThe second visualization shows the stages as a series of bars. The stages are\narranged in topological order. Stages with no descendants are shown first,\nfollowed by their descendants. The lengths of the bars represent data freshness.\nTo see the data freshness values at a specific point, click the graph. The\nbars update to show data freshness at the selected time.\n\nThe following image shows a job with four stages. At the selected timestamp,\nthe data freshness ranges from 9 seconds to 13 seconds.\n\nThe next image shows the same job with a different timestamp selected. At this\npoint, the data freshness for all stages exceeds 4 minutes, signalling that the\npipeline might be stuck.\n\nThe **Stage Info** panel shows more detailed information about a stage. To view\nthe details for a stage, click the progress bar for that stage. The **Stage\nInfo** panel shows the following information about a stage:\n\n- Status\n- [System lag](/monitoring/api/metrics_gcp_d_h#dataflow/job/per_stage_system_lag): The maximum time that an item of data has been awaiting processing\n- [Data watermark](/monitoring/api/metrics_gcp_d_h#dataflow/job/per_stage_data_wat:ermark_age): The estimated completion time of data input for this stage\n- Details about any [stragglers](/dataflow/docs/guides/troubleshoot-streaming-stragglers)\n- The pipeline steps that this stage encompasses\n\nIf the panel is not visible, click last_page\n**Toggle panel \"Stage info\"**.\n\nStage workflow\n--------------\n\n**Stage workflow** view shows the job stages as a workflow graph. To view\nthe details for a stage, click the box for that stage.\n\nFor batch jobs, click **Critical path** to view only the stages that directly\ncontribute to the job's overall runtime.\n\nWorker progress\n---------------\n\nFor batch jobs, the **Worker progress** view shows the workers for a particular\nstage. This view is not available for streaming jobs. To access this view,\nselect **Worker progress** and select the stage in **Filter workers by stage** .\nAlternatively, you can activate this view from the **Stage progress** view as\nfollows:\n\n1. In the **Stage progress** view, identify the stage that you want to view.\n2. Hold the pointer over the bar for that stage.\n3. In the **Stage** card, click **View workers** . The **Worker progress** view is shown with the stage pre-selected.\n\nEach bar maps to a work item scheduled to a worker. A sparkline that tracks CPU\nutilization on a worker is located with each worker, making it easier to spot\nunderutilization issues.\n\nWhat's next\n-----------\n\n- Learn more about [troubleshooting Dataflow pipelines](/dataflow/docs/guides/troubleshooting-your-pipeline).\n- Read about the different components of [Dataflow's\n web-based monitoring user interface](/dataflow/docs/guides/monitoring-overview)."]]