[[["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\u003eCompute Engine reservations can be used to ensure VM resources are available for Dataflow jobs.\u003c/p\u003e\n"],["\u003cp\u003eTo utilize reservations, create a Compute Engine reservation and pass the appropriate service option when submitting a Dataflow job, dependent on the Beam SDK version used.\u003c/p\u003e\n"],["\u003cp\u003eSetting the reservation affinity to \u003ccode\u003enone\u003c/code\u003e for low-priority workloads prevents competition for reservations with Dataflow jobs.\u003c/p\u003e\n"],["\u003cp\u003eDataflow worker configurations must match the reservation's configuration to successfully consume the reserved resources, which may require adjustments to the worker machine type.\u003c/p\u003e\n"],["\u003cp\u003eCompute Engine reservations used with Dataflow are billed by Dataflow while the job runs and by Compute Engine when idle, and they are not eligible for Compute Engine committed use discounts.\u003c/p\u003e\n"]]],[],null,["To ensure that VM resources are available when your Dataflow jobs need\nthem, you can use Compute Engine reservations. Reservations provide a high\nlevel of assurance in obtaining capacity for Compute Engine zonal\nresources.\n\nTo use Compute Engine reservations with Dataflow, perform the\nfollowing steps:\n\n1. Create a Compute Engine reservation. It can be a single-project\n reservation or a shared reservation. For more information, see the following\n documents:\n\n - [Create a reservation for a single project](/compute/docs/instances/reservations-single-project)\n - [Create a shared reservation](/compute/docs/instances/reservations-shared)\n\n The reservation can include GPU or TPU accelerators.\n2. When you submit your Dataflow job, pass one of the following\n service options, depending on which version of the Beam SDK you are using:\n\n - Beam version \\\u003c 2.29: `--experiments=skip_gce_quota_verification`\n - Beam version \\\u003e= 2.29: `--dataflow_service_options=automatically_use_created_reservation`\n\nTo prevent low-priority workloads in the same project from competing for\nreservations with Dataflow, set the reservation affinity to\n`none` when you create VMs for those workloads. For more information, see\n[Consuming reserved instances](/compute/docs/instances/reserving-zonal-resources#consuming_reserved_instances).\n\nIn order to use the reservation, the Dataflow workers must match\nthe reservation configuration. You might need to set the worker machine type for\nthe job. For more information, see\n[Workers](/dataflow/docs/request-quotas#workers).\n\nLimitations\n\n- All limitations of Compute Engine reservations apply when\n Dataflow workers consume reservations. See\n [How reservations work](/compute/docs/instances/reservations-overview#how-reservations-work).\n\n- Dataflow relies on the\n [default consumption order](/compute/docs/instances/reservations-overview#consumption-order)\n in Compute Engine. As a result, the following limitations apply:\n\n - Other workloads in the same project or Organization that don't specify the `--reservation` flag might compete with Dataflow workloads for project-specific or shared reservations.\n- Dataflow Prime jobs don't consume Compute Engine reservations.\n\nReservations and accelerators\n\nDataflow supports [*specifically targeted*\nreservations](/compute/docs/instances/reservations-consume#consuming_instances_from_a_specific_reservation)\nfor pipelines using accelerators (GPUs or TPUs). This functionality is generally\navailable with an allowlist. For instructions on using Dataflow\naccelerators with specific reservations, contact your account team.\n\nPricing\n\nDataflow bills you for VMs from *automatically consumed*\nreservations while your Dataflow job runs. When\nDataflow isn't using the VMs, Compute Engine bills you.\n\nCompute Engine pricing model\n\nIf your Dataflow usage includes VMs from [*specifically targeted*\nreservations](/compute/docs/instances/reservations-overview#consumption-type)\nthat have GPUs or TPUs, then compute resources from those reserved VMs are\nbilled according to [Compute Engine\nPricing](/compute/all-pricing). If your *specifically targeted* reservations are\nattached to a [Compute Engine resource-based\ncommitment](/compute/docs/instances/signing-up-committed-use-discounts), then\nyou also receive applicable resource-based committed use discounts (CUDs) for\nyour usage. You're also billed a management premium for compute resources\nconsumed in Dataflow. For more pricing details, see [Dataflow Pricing](/dataflow/pricing).\n\nDataflow pricing model\n\nFor any other type of Compute Engine reservations that you use with\nDataflow, your usage is billed by using the\n[Dataflow pricing model](/dataflow/pricing). Dataflow\nusage from those reservations isn't eligible for resource-based CUDs, even if\nthose reservations are attached to a resource-based commitment. This applies to\nthe following Compute Engine reservations:\n\n- *Specifically targeted* reservations that don't have GPUs or TPUs\n- All *automatically consumed* reservations\n\nWhat's next\n\nTo learn more about Compute Engine reservations, see\n[Reservations of Compute Engine zonal resources](/compute/docs/instances/reservations-overview)."]]