Apache Spark バッチ ワークロードとインタラクティブ セッション用の Serverless は、内部 IP アドレスのみを持つ VM と、サブネットで 限定公開の Google アクセス（PGA）が自動的に有効になっているリージョン サブネットで実行されます。
サブネットを指定しない場合、Serverless for Apache Spark は、バッチ ワークロードまたはセッションのリージョンにある default サブネットをバッチ ワークロードまたはセッションのサブネットとして選択します。
ワークロードで外部ネットワークまたはインターネット アクセスが必要な場合（たとえば、PyTorch Hub や Hugging Face から ML モデルなどのリソースをダウンロードする場合）、Cloud NAT を設定して、VPC ネットワークで内部 IP を使用してアウトバウンド トラフィックを許可できます。
オープン サブネット接続
Serverless for Apache Spark バッチ ワークロードまたはインタラクティブ セッション用に選択されたリージョンの VPC サブネットでは、VM インスタンス間のすべてのポートで内部サブネット通信を許可する必要があります。
次の Google Cloud CLI コマンドは、すべてのポートですべてのプロトコルを使用して VM 間の内部上り（内向き）通信を許可するサブネットにネットワーク ファイアウォールを接続します。
[[["わかりやすい","easyToUnderstand","thumb-up"],["問題の解決に役立った","solvedMyProblem","thumb-up"],["その他","otherUp","thumb-up"]],[["わかりにくい","hardToUnderstand","thumb-down"],["情報またはサンプルコードが不正確","incorrectInformationOrSampleCode","thumb-down"],["必要な情報 / サンプルがない","missingTheInformationSamplesINeed","thumb-down"],["翻訳に関する問題","translationIssue","thumb-down"],["その他","otherDown","thumb-down"]],["最終更新日 2025-09-04 UTC。"],[[["\u003cp\u003eDataproc Serverless for Spark workloads and interactive sessions require a Virtual Private Cloud (VPC) subnetwork.\u003c/p\u003e\n"],["\u003cp\u003eThe selected VPC subnet must have Private Google Access enabled to ensure internal IP address functionality, and it must allow internal subnet communication on all ports between VM instances.\u003c/p\u003e\n"],["\u003cp\u003eFor workloads needing external access, you can use Cloud NAT to enable outbound traffic via internal IPs on your VPC network.\u003c/p\u003e\n"],["\u003cp\u003eWhen using VPC Service Controls (VPC-SC), you should set up private connectivity and consider using a custom container image for pre-installing dependencies outside the VPC-SC perimeter.\u003c/p\u003e\n"]]],[],null,["# Serverless for Apache Spark network configuration\n\nThis document describes the requirements needed for Google Cloud Serverless for Apache Spark\nnetwork configuration.\n\nVirtual Private Cloud subnetwork requirements\n---------------------------------------------\n\nThis document explains the Virtual Private Cloud network requirements for\nGoogle Cloud Serverless for Apache Spark batch workloads and interactive sessions.\n\n### Private Google Access\n\nServerless for Apache Spark batch workloads and interactive sessions\nrun on VMs with internal IP addresses only and on a regional subnet with\n[Private Google Access (PGA)](/vpc/docs/configure-private-google-access)\nautomatically enabled on the subnet.\n\nIf you don't specify a subnet, Serverless for Apache Spark selects the\n`default` subnet in the batch workload or session region as the subnet for a\nbatch workload or session.\n\nIf your workload requires external network or internet\naccess, for example to download resources such as ML models from\n[PyTorch Hub](https://pytorch.org/hub/) or [Hugging Face](https://huggingface.co/),\nyou can set up [Cloud NAT](/nat/docs/overview) to allow outbound traffic\nusing internal IPs on your VPC network.\n\n### Open subnet connectivity\n\nThe VPC subnet for the region selected for the\nServerless for Apache Spark batch workload or interactive session must\nallow internal subnet communication on all ports between VM instances.\n| **Note:** To prevent malicious scripts in one workload from affecting other workloads, Serverless for Apache Spark deploys [default security measures](/dataproc-serverless/docs/concepts/security).\n\nThe following Google Cloud CLI command attaches a network firewall to a\nsubnet that allows internal ingress communications among VMs using all protocols\non all ports: \n\n```\ngcloud compute firewall-rules create allow-internal-ingress \\\n --network=NETWORK_NAME \\\n --source-ranges=SUBNET_RANGES \\\n --destination-ranges=SUBNET_RANGES \\\n --direction=ingress \\\n --action=allow \\\n --rules=all\n```\n\nNotes:\n\n- \u003cvar translate=\"no\"\u003eSUBNET_RANGES:\u003c/var\u003e See\n [Allow internal ingress connections between VMs](/firewall/docs/using-firewalls#common-use-cases-allow-internal).\n The `default` VPC network in a project with the\n `default-allow-internal` firewall rule, which allows ingress communication on\n all ports (`tcp:0-65535`, `udp:0-65535`, and `icmp protocols:ports`),\n meets the open-subnet-connectivity requirement. However, this rule also allows\n ingress by any VM instance on the network.\n\n | **Use network tags to limit connectivity**. In production, the recommended practice is to limit firewall rules to the IP addresses used by your Spark workloads.\n\nServerless for Apache Spark and VPC-SC networks\n-----------------------------------------------\n\nWith [VPC Service Controls](/vpc-service-controls/docs),\nnetwork administrators can define a security perimeter around resources of\nGoogle-managed services to control communication to and between those services.\n\nNote the following strategies when using VPC-SC\nnetworks with Serverless for Apache Spark:\n\n- [Set up private connectivity](/vpc-service-controls/docs/set-up-private-connectivity).\n\n- Create a [custom container image](/dataproc-serverless/docs/guides/custom-containers)\n that pre-installs dependencies outside the VPC-SC perimeter,\n and then [submit a Spark batch workload](/dataproc-serverless/docs/guides/custom-containers#submit_a_spark_batch_workload_using_a_custom_container_image)\n that uses your custom container image.\n\nFor more information, see\n[VPC Service Controls---Serverless for Apache Spark](/vpc-service-controls/docs/supported-products#table_dataproc_serverless)."]]
