dataproc

Google Cloud Dataproc

This module Manages a Google Cloud Dataproc cluster resource, including IAM.

• TODO
• Examples
  • Simple
  • Cluster configuration on GCE
  • Cluster configuration on GCE with CMEK encryption
  • Cluster configuration on GKE
• IAM
  • Authoritative IAM
  • Additive IAM
• Variables
• Outputs
• Fixtures

TODO

• Add support for Cloud Dataproc autoscaling policy.

Examples

Simple

module "dataproc-cluster" {
  source     = "./fabric/modules/dataproc"
  project_id = var.project_id
  name       = "my-cluster"
  region     = var.region
}
# tftest modules=1 resources=1

Cluster configuration on GCE

To set cluster configuration use the 'dataproc_config.cluster_config' variable. If you don't want to use dedicated service account, remember to grant roles/dataproc.worker to Compute Default Service Account.

module "dataproc-service-account" {
  source     = "./fabric/modules/iam-service-account"
  project_id = var.project_id
  name       = "dataproc-worker"
  iam_project_roles = {
    (var.project_id) = ["roles/dataproc.worker"]
  }
}

module "firewall" {
  source     = "./fabric/modules/net-vpc-firewall"
  project_id = var.project_id
  network    = var.vpc.name
  ingress_rules = {
    allow-ingress-dataproc = {
      description = "Allow all traffic between Dataproc nodes."
      targets     = ["dataproc"]
      sources     = ["dataproc"]
    }
  }
}

module "processing-dp-cluster" {
  source     = "./fabric/modules/dataproc"
  project_id = var.project_id
  name       = "my-cluster"
  region     = var.region
  dataproc_config = {
    cluster_config = {
      gce_cluster_config = {
        internal_ip_only       = true
        service_account        = module.dataproc-service-account.email
        service_account_scopes = ["cloud-platform"]
        subnetwork             = var.subnet.self_link
        tags                   = ["dataproc"]
        zone                   = "${var.region}-b"
      }
    }
  }
  depends_on = [
    module.dataproc-service-account, # ensure all grants are done before creating the cluster
  ]
}
# tftest modules=3 resources=7 e2e

Cluster configuration on GCE with CMEK encryption

To set cluster configuration use the Customer Managed Encryption key, set dataproc_config.encryption_config. variable. The Compute Engine service agent and the Cloud Storage service agent need to have CryptoKey Encrypter/Decrypter role on they configured KMS key (Documentation).

module "dataproc-service-account" {
  source     = "./fabric/modules/iam-service-account"
  project_id = var.project_id
  name       = "dataproc-worker"
  iam_project_roles = {
    (var.project_id) = ["roles/dataproc.worker", "roles/cloudkms.cryptoKeyEncrypterDecrypter"]
  }
}

module "firewall" {
  source     = "./fabric/modules/net-vpc-firewall"
  project_id = var.project_id
  network    = var.vpc.name
  ingress_rules = {
    allow-ingress-dataproc = {
      description = "Allow all traffic between Dataproc nodes."
      targets     = ["dataproc"]
      sources     = ["dataproc"]
    }
  }
}

module "processing-dp-cluster" {
  source     = "./fabric/modules/dataproc"
  project_id = var.project_id
  name       = "my-cluster"
  region     = var.region
  dataproc_config = {
    cluster_config = {
      gce_cluster_config = {
        internal_ip_only       = true
        service_account        = module.dataproc-service-account.email
        service_account_scopes = ["cloud-platform"]
        subnetwork             = var.subnet.self_link
        tags                   = ["dataproc"]
        zone                   = "${var.region}-b"
      }
    }
    encryption_config = {
      kms_key_name = var.kms_key.id
    }
  }
  depends_on = [
    module.dataproc-service-account, # ensure all grants are done before creating the cluster
  ]
}
# tftest modules=3 resources=8 e2e

Cluster configuration on GKE

To set cluster configuration GKE use the 'dataproc_config.virtual_cluster_config' variable. This example shows usage of dedicated Service Account.

locals {
  dataproc_namespace = "foobar"
}

module "dataproc-service-account" {
  source     = "./fabric/modules/iam-service-account"
  project_id = var.project_id
  name       = "dataproc-worker"
  iam = {
    "roles/iam.workloadIdentityUser" = [
      "serviceAccount:${var.project_id}.svc.id.goog[${local.dataproc_namespace}/agent]",
      "serviceAccount:${var.project_id}.svc.id.goog[${local.dataproc_namespace}/spark-driver]",
      "serviceAccount:${var.project_id}.svc.id.goog[${local.dataproc_namespace}/spark-executor]"
    ]
  }
  iam_project_roles = {
    (var.project_id) = ["roles/dataproc.worker"]
  }
  depends_on = [
    module.gke-cluster-standard, # granting workloadIdentityUser requires cluster/pool to be created first
  ]
}

module "processing-dp-cluster" {
  source     = "./fabric/modules/dataproc"
  project_id = var.project_id
  name       = "my-dataproc-cluster"
  region     = var.region
  dataproc_config = {
    virtual_cluster_config = {
      kubernetes_cluster_config = {
        kubernetes_namespace = local.dataproc_namespace
        kubernetes_software_config = {
          component_version = {
            "SPARK" : "3.1-dataproc-14"
          }
          properties = {
            "dataproc:dataproc.gke.agent.google-service-account"          = module.dataproc-service-account.email
            "dataproc:dataproc.gke.spark.driver.google-service-account"   = module.dataproc-service-account.email
            "dataproc:dataproc.gke.spark.executor.google-service-account" = module.dataproc-service-account.email
          }
        }
        gke_cluster_config = {
          gke_cluster_target = module.gke-cluster-standard.id
          node_pool_target = {
            node_pool = "node-pool-name"
            roles     = ["DEFAULT"]
          }
        }
      }
    }
  }
}
# tftest modules=4 resources=6 fixtures=fixtures/gke-cluster-standard.tf e2e

IAM

IAM is managed via several variables that implement different features and levels of control:

• iam and iam_by_principals configure authoritative bindings that manage individual roles exclusively, and are internally merged
• iam_bindings configure authoritative bindings with optional support for conditions, and are not internally merged with the previous two variables
• iam_bindings_additive configure additive bindings via individual role/member pairs with optional support conditions

The authoritative and additive approaches can be used together, provided different roles are managed by each. Some care must also be taken with the iam_by_principals variable to ensure that variable keys are static values, so that Terraform is able to compute the dependency graph.

Refer to the project module for examples of the IAM interface.

Authoritative IAM

module "processing-dp-cluster" {
  source     = "./fabric/modules/dataproc"
  project_id = var.project_id
  name       = "my-cluster"
  region     = var.region
  iam_by_principals = {
    "group:gcp-data-engineers@example.net" = [
      "roles/dataproc.viewer"
    ]
  }
  iam = {
    "roles/dataproc.viewer" = [
      "serviceAccount:service-account@PROJECT_ID.iam.gserviceaccount.com"
    ]
  }
}
# tftest modules=1 resources=2

Additive IAM

module "processing-dp-cluster" {
  source     = "./fabric/modules/dataproc"
  project_id = var.project_id
  name       = "my-cluster"
  region     = var.region
  iam_bindings_additive = {
    am1-viewer = {
      member = "user:am1@example.com"
      role   = "roles/dataproc.viewer"
    }
  }
}
# tftest modules=1 resources=2

Variables

name	description	type	required	default
name	Cluster name.	string	✓
project_id	Project ID.	string	✓
region	Dataproc region.	string	✓
dataproc_config	Dataproc cluster config.	object({…})		{}
iam	IAM bindings in {ROLE => [MEMBERS]} format.	map(list(string))		{}
iam_bindings	Authoritative IAM bindings in {KEY => {role = ROLE, members = [], condition = {}}}. Keys are arbitrary.	map(object({…}))		{}
iam_bindings_additive	Individual additive IAM bindings. Keys are arbitrary.	map(object({…}))		{}
iam_by_principals	Authoritative IAM binding in {PRINCIPAL => [ROLES]} format. Principals need to be statically defined to avoid cycle errors. Merged internally with the iam variable.	map(list(string))		{}
labels	The resource labels for instance to use to annotate any related underlying resources, such as Compute Engine VMs.	map(string)		{}

Outputs

name	description	sensitive
id	Fully qualified cluster id.
name	The name of the cluster.

Fixtures

• gke-cluster-standard.tf