Skip to content

Latest commit

 

History

History
401 lines (320 loc) · 25.4 KB

cluster.md

File metadata and controls

401 lines (320 loc) · 25.4 KB
Raw
subcategory
Compute

databricks_cluster resource

This resource allows you to create, update, and delete clusters.

data "databricks_node_type" "smallest" {
  local_disk = true
}

data "databricks_spark_version" "latest_lts" {
  long_term_support = true
}

resource "databricks_cluster" "shared_autoscaling" {
  cluster_name            = "Shared Autoscaling"
  spark_version           = data.databricks_spark_version.latest_lts.id
  node_type_id            = data.databricks_node_type.smallest.id
  autotermination_minutes = 20
  autoscale {
    min_workers = 1
    max_workers = 50
  }
}

Argument Reference

  • cluster_name - (Optional) Cluster name, which doesn’t have to be unique. If not specified at creation, the cluster name will be an empty string.
  • spark_version - (Required) Runtime version of the cluster. Any supported databricks_spark_version id. We advise using Cluster Policies to restrict the list of versions for simplicity while maintaining enough control.
  • driver_node_type_id - (Optional) The node type of the Spark driver. This field is optional; if unset, API will set the driver node type to the same value as node_type_id defined above.
  • node_type_id - (Required - optional if instance_pool_id is given) Any supported databricks_node_type id. If instance_pool_id is specified, this field is not needed.
  • instance_pool_id (Optional - required if node_type_id is not given) - To reduce cluster start time, you can attach a cluster to a predefined pool of idle instances. When attached to a pool, a cluster allocates its driver and worker nodes from the pool. If the pool does not have sufficient idle resources to accommodate the cluster’s request, it expands by allocating new instances from the instance provider. When an attached cluster changes its state to TERMINATED, the instances it used are returned to the pool and reused by a different cluster.
  • driver_instance_pool_id (Optional) - similar to instance_pool_id, but for driver node. If omitted, and instance_pool_id is specified, then driver will be allocated from that pool.
  • policy_id - (Optional) Identifier of Cluster Policy to validate cluster and preset certain defaults. The primary use for cluster policies is to allow users to create policy-scoped clusters via UI rather than sharing configuration for API-created clusters. For example, when you specify policy_id of external metastore policy, you still have to fill in relevant keys for spark_conf.
  • autotermination_minutes - (Optional) Automatically terminate the cluster after being inactive for this time in minutes. If not set, Databricks won't automatically terminate an inactive cluster. If specified, the threshold must be between 10 and 10000 minutes. You can also set this value to 0 to explicitly disable automatic termination. We highly recommend having this setting present for Interactive/BI clusters.
  • enable_elastic_disk - (Optional) If you don’t want to allocate a fixed number of EBS volumes at cluster creation time, use autoscaling local storage. With autoscaling local storage, Databricks monitors the amount of free disk space available on your cluster’s Spark workers. If a worker begins to run too low on disk, Databricks automatically attaches a new EBS volume to the worker before it runs out of disk space. EBS volumes are attached up to a limit of 5 TB of total disk space per instance (including the instance’s local storage). To scale down EBS usage, make sure you have autotermination_minutes and autoscale attributes set. More documentation available at cluster configuration page.
  • enable_local_disk_encryption - (Optional) Some instance types you use to run clusters may have locally attached disks. Databricks may store shuffle data or temporary data on these locally attached disks. To ensure that all data at rest is encrypted for all storage types, including shuffle data stored temporarily on your cluster’s local disks, you can enable local disk encryption. When local disk encryption is enabled, Databricks generates an encryption key locally unique to each cluster node and encrypting all data stored on local disks. The scope of the key is local to each cluster node and is destroyed along with the cluster node itself. During its lifetime, the key resides in memory for encryption and decryption and is stored encrypted on the disk. Your workloads may run more slowly because of the performance impact of reading and writing encrypted data to and from local volumes. This feature is not available for all Azure Databricks subscriptions. Contact your Microsoft or Databricks account representative to request access.
  • single_user_name - (Optional) The optional user name of the user to assign to an interactive cluster. This field is required when using standard AAD Passthrough for Azure Data Lake Storage (ADLS) with a single-user cluster (i.e., not high-concurrency clusters).
  • idempotency_token - (Optional) An optional token to guarantee the idempotency of cluster creation requests. If an active cluster with the provided token already exists, the request will not create a new cluster, but it will return the existing running cluster's ID instead. If you specify the idempotency token, upon failure, you can retry until the request succeeds. Databricks platform guarantees to launch exactly one cluster with that idempotency token. This token should have at most 64 characters.
  • ssh_public_keys - (Optional) SSH public key contents that will be added to each Spark node in this cluster. The corresponding private keys can be used to login with the user name ubuntu on port 2200. You can specify up to 10 keys.
  • spark_env_vars - (Optional) Map with environment variable key-value pairs to fine-tune Spark clusters. Key-value pairs of the form (X,Y) are exported (i.e., X='Y') while launching the driver and workers.
  • custom_tags - (Optional) Additional tags for cluster resources. Databricks will tag all cluster resources (e.g., AWS EC2 instances and EBS volumes) with these tags in addition to default_tags.
  • spark_conf - (Optional) Map with key-value pairs to fine-tune Spark clusters, where you can provide custom Spark configuration properties in a cluster configuration.
  • is_pinned - (Optional) boolean value specifying if cluster is pinned (not pinned by default). You must be a Databricks administrator to use this. The pinned clusters' maximum number is limited to 20, so apply may fail if you have more than that.

The following example demonstrates how to create an autoscaling cluster with Delta Cache enabled:

data "databricks_node_type" "smallest" {
  local_disk = true
}

data "databricks_spark_version" "latest_lts" {
  long_term_support = true
}

resource "databricks_cluster" "shared_autoscaling" {
  cluster_name            = "Shared Autoscaling"
  spark_version           = data.databricks_spark_version.latest_lts.id
  node_type_id            = data.databricks_node_type.smallest.id
  autotermination_minutes = 20
  autoscale {
    min_workers = 1
    max_workers = 50
  }
  spark_conf = {
    "spark.databricks.io.cache.enabled": true,
    "spark.databricks.io.cache.maxDiskUsage": "50g",
    "spark.databricks.io.cache.maxMetaDataCache": "1g"
  }
}

Fixed size or autoscaling cluster

When you create a Databricks cluster, you can either provide a num_workers for the fixed-size cluster or provide min_workers and/or max_workers for the cluster within the autoscale group. When you give a fixed-sized cluster, Databricks ensures that your cluster has a specified number of workers. When you provide a range for the number of workers, Databricks chooses the appropriate number of workers required to run your job - also known as "autoscaling." With autoscaling, Databricks dynamically reallocates workers to account for the characteristics of your job. Certain parts of your pipeline may be more computationally demanding than others, and Databricks automatically adds additional workers during these phases of your job (and removes them when they’re no longer needed).

autoscale optional configuration block supports the following:

  • min_workers - (Optional) The minimum number of workers to which the cluster can scale down when underutilized. It is also the initial number of workers the cluster will have after creation.
  • max_workers - (Optional) The maximum number of workers to which the cluster can scale up when overloaded. max_workers must be strictly greater than min_workers.

When using a Single Node cluster, num_workers needs to be 0. It can be set to 0 explicitly, or simply not specified, as it defaults to 0. When num_workers is 0, provider checks for presence of the required Spark configurations:

  • spark.master must has prefix local, like local[*]
  • spark.databricks.cluster.profile must have value singleNode

and also custom_tag entry:

  • "ResourceClass" = "SingleNode"

The following example demonstrates how to create an single node cluster:

data "databricks_node_type" "smallest" {
  local_disk = true
}

data "databricks_spark_version" "latest_lts" {
  long_term_support = true
}

resource "databricks_cluster" "single_node" {
  cluster_name            = "Single Node"
  spark_version           = data.databricks_spark_version.latest_lts.id
  node_type_id            = data.databricks_node_type.smallest.id
  autotermination_minutes = 20

  spark_conf = {
    # Single-node
    "spark.databricks.cluster.profile" : "singleNode"
    "spark.master" : "local[*]"
  }

  custom_tags = {
    "ResourceClass" = "SingleNode"
  }
}

High-Concurrency clusters

To create High-Concurrency cluster, following settings should be provided:

  • spark_conf should have following items:
    • spark.databricks.repl.allowedLanguages set to a list of supported languages, for example: python,sql, or python,sql,r. Scala is not supported!
    • spark.databricks.cluster.profile set to serverless
  • custom_tags should have tag ResourceClass set to value Serverless

For example:

esource "databricks_cluster" "cluster_with_table_access_control" {
  cluster_name            = "Shared High-Concurrency"
  spark_version           = data.databricks_spark_version.latest_lts.id
  node_type_id            = data.databricks_node_type.smallest.id
  autotermination_minutes = 20

  spark_conf = {
    "spark.databricks.repl.allowedLanguages": "python,sql",
    "spark.databricks.cluster.profile": "serverless"
  }

  custom_tags = {
    "ResourceClass" = "Serverless"
  }
}

library Configuration Block

To install libraries, one must specify each library in a separate configuration block. Each different type of library has a slightly different syntax. It's possible to set only one type of library within one config block. Otherwise, the plan will fail with an error.

Installing JAR artifacts on a cluster. Location can be anything, that is DBFS or mounted object store (s3, adls, ...)

library {
  jar = "dbfs:/FileStore/app-0.0.1.jar"
}

Installing Python EGG artifacts. Location can be anything, that is DBFS or mounted object store (s3, adls, ...)

library {
  egg = "dbfs:/FileStore/foo.egg"
}

Installing Python Wheel artifacts. Location can be anything, that is DBFS or mounted object store (s3, adls, ...)

library {
  whl = "dbfs:/FileStore/baz.whl"
}

Installing Python PyPI artifacts. You can optionally also specify the repo parameter for custom PyPI mirror, which should be accessible without any authentication for the network that cluster runs in.

library {
  pypi {
    package = "fbprophet==0.6"
    // repo can also be specified here
  }
}

Installing artifacts from Maven repository. You can also optionally specify a repo parameter for custom Maven-style repository, that should be accessible without any authentication for the network that cluster runs in. It can even be properly configured maven s3 wagon, AWS CodeArtifact or Azure Artifacts.

library {
  maven {
    coordinates = "com.amazon.deequ:deequ:1.0.4"
    // exlusions block is optional
    exclusions = ["org.apache.avro:avro"]
  }
}

Installing artifacts from CRan. You can also optionally specify a repo parameter for a custom cran mirror.

library {
  cran {
    package = "rkeops"
  }
}

cluster_log_conf

Example of pushing all cluster logs to DBFS:

cluster_log_conf {
  dbfs {
    destination = "dbfs:/cluster-logs"
  }
}

Example of pushing all cluster logs to S3:

cluster_log_conf {
  s3 {
    destination = "s3a://acmecorp-main/cluster-logs"
    region = "us-east-1"
  }
}

There are a few more advanced attributes for S3 log delivery:

  • destination - S3 destination, e.g., s3://my-bucket/some-prefix You must configure the cluster with an instance profile, and the instance profile must have write access to the destination. You cannot use AWS keys.
  • region - (Optional) S3 region, e.g. us-west-2. Either region or endpoint must be set. If both are set, the endpoint is used.
  • endpoint - (Optional) S3 endpoint, e.g. https://s3-us-west-2.amazonaws.com. Either region or endpoint needs to be set. If both are set, the endpoint is used.
  • enable_encryption - (Optional) Enable server-side encryption, false by default.
  • encryption_type - (Optional) The encryption type, it could be sse-s3 or sse-kms. It is used only when encryption is enabled, and the default type is sse-s3.
  • kms_key - (Optional) KMS key used if encryption is enabled and encryption type is set to sse-kms.
  • canned_acl - (Optional) Set canned access control list, e.g. bucket-owner-full-control. If canned_cal is set, the cluster instance profile must have s3:PutObjectAcl permission on the destination bucket and prefix. The full list of possible canned ACLs can be found here. By default, only the object owner gets full control. If you are using a cross-account role for writing data, you may want to set bucket-owner-full-control to make bucket owners able to read the logs.

init_scripts

You can specify up to 10 different init scripts for the specific cluster. If you want a shell script to run on all clusters and jobs within the same workspace, you should consider databricks_global_init_script.

Example of taking init script from DBFS:

init_scripts {
  dbfs {
    destination = "dbfs:/init-scripts/install-elk.sh"
  }
}

Example of taking init script from S3:

init_scripts {
  s3 {
    destination = "s3a://acmecorp-main/init-scripts/install-elk.sh"
    region = "us-east-1"
  }
}

Like the cluster_log_conf configuration block, init scripts support S3 and DBFS locations. In addition, you can also specify a local file as follows:

init_scripts {
  file {
    destination = "file:/my/local/file.sh"
  }
}

Take note that this can only be specified for clusters with custom Docker containers.

aws_attributes

aws_attributes optional configuration block contains attributes related to clusters running on Amazon Web Services.

Here is the example of shared autoscaling cluster with some of AWS options set:

resource "databricks_cluster" "this" {
  cluster_name            = "Shared Autoscaling"
  spark_version           = "6.6.x-scala2.11"
  node_type_id            = "i3.xlarge"
  autotermination_minutes = 20
  autoscale {
    min_workers = 1
    max_workers = 50
  }
  aws_attributes {
    availability            = "SPOT"
    zone_id                 = "us-east-1"
    first_on_demand         = 1
    spot_bid_price_percent  = 100
  }
}

The following options are available:

  • zone_id - (Required) Identifier for the availability zone/datacenter in which the cluster resides. This string will be of a form like “us-west-2a”. The provided availability zone must be in the same region as the Databricks deployment. For example, “us-west-2a” is not a valid zone ID if the Databricks deployment resides in the “us-east-1” region.
  • availability - (Optional) Availability type used for all subsequent nodes past the first_on_demand ones. Valid values are SPOT, SPOT_WITH_FALLBACK and ON_DEMAND. Note: If first_on_demand is zero, this availability type will be used for the entire cluster.
  • first_on_demand - (Optional) The first first_on_demand nodes of the cluster will be placed on on-demand instances. If this value is greater than 0, the cluster driver node will be placed on an on-demand instance. If this value is greater than or equal to the current cluster size, all nodes will be placed on on-demand instances. If this value is less than the current cluster size, first_on_demand nodes will be placed on on-demand instances, and the remainder will be placed on availability instances. This value does not affect cluster size and cannot be mutated over the lifetime of a cluster.
  • spot_bid_price_percent - (Optional) The max price for AWS spot instances, as a percentage of the corresponding instance type’s on-demand price. For example, if this field is set to 50, and the cluster needs a new i3.xlarge spot instance, then the max price is half of the price of on-demand i3.xlarge instances. Similarly, if this field is set to 200, the max price is twice the price of on-demand i3.xlarge instances. If not specified, the default value is 100. When spot instances are requested for this cluster, only spot instances whose max price percentage matches this field will be considered. For safety, we enforce this field to be no more than 10000.
  • instance_profile_arn - (Optional) Nodes for this cluster will only be placed on AWS instances with this instance profile. Please see databricks_instance_profile resource documentation for extended examples on adding a valid instance profile using Terraform.
  • ebs_volume_type - (Optional) The type of EBS volumes that will be launched with this cluster. Valid values are GENERAL_PURPOSE_SSD or THROUGHPUT_OPTIMIZED_HDD. Use this option only if you're not picking Delta Optimized i3.* node types.
  • ebs_volume_count - (Optional) The number of volumes launched for each instance. You can choose up to 10 volumes. This feature is only enabled for supported node types. Legacy node types cannot specify custom EBS volumes. For node types with no instance store, at least one EBS volume needs to be specified; otherwise, cluster creation will fail. These EBS volumes will be mounted at /ebs0, /ebs1, and etc. Instance store volumes will be mounted at /local_disk0, /local_disk1, and etc. If EBS volumes are attached, Databricks will configure Spark to use only the EBS volumes for scratch storage because heterogeneously sized scratch devices can lead to inefficient disk utilization. If no EBS volumes are attached, Databricks will configure Spark to use instance store volumes. If EBS volumes are specified, then the Spark configuration spark.local.dir will be overridden.
  • ebs_volume_size - (Optional) The size of each EBS volume (in GiB) launched for each instance. For general purpose SSD, this value must be within the range 100 - 4096. For throughput optimized HDD, this value must be within the range 500 - 4096. Custom EBS volumes cannot be specified for the legacy node types (memory-optimized and compute-optimized).

azure_attributes

azure_attributes optional configuration block contains attributes related to clusters running on Azure.

Here is the example of shared autoscaling cluster with some of AWS options set:

resource "databricks_cluster" "this" {
  cluster_name            = "Shared Autoscaling"
  spark_version           = "6.6.x-scala2.11"
  node_type_id            = "Standard_DS3_v2"
  autotermination_minutes = 20
  autoscale {
    min_workers = 1
    max_workers = 50
  }
  azure_attributes {
    availability       = "SPOT_AZURE"
    first_on_demand    = 1
    spot_bid_max_price = 100
  }
}

The following options are available:

  • availability - (Optional) Availability type used for all subsequent nodes past the first_on_demand ones. Valid values are SPOT_AZURE, SPOT_WITH_FALLBACK_AZURE, and ON_DEMAND_AZURE. Note: If first_on_demand is zero, this availability type will be used for the entire cluster.
  • first_on_demand - (Optional) The first first_on_demand nodes of the cluster will be placed on on-demand instances. If this value is greater than 0, the cluster driver node will be placed on an on-demand instance. If this value is greater than or equal to the current cluster size, all nodes will be placed on on-demand instances. If this value is less than the current cluster size, first_on_demand nodes will be placed on on-demand instances, and the remainder will be placed on availability instances. This value does not affect cluster size and cannot be mutated over the lifetime of a cluster.
  • spot_bid_max_price - (Optional) The max price for Azure spot instances. Use -1 to specify lowest price.

gcp_attributes

gcp_attributes optional configuration block contains attributes related to clusters running on GCP.

The following options are available:

  • use_preemptible_executors - (Optional, bool) if we should use preemptible executors (GCP documentation)
  • google_service_account - (Optional, string) Google Service Account email address that the cluster uses to authenticate with Google Identity. This field is used for authentication with the GCS and BigQuery data sources.

docker_image

Databricks Container Services lets you specify a Docker image when you create a cluster. You need to enable Container Services in Admin Console / Advanced page in the user interface. By enabling this feature, you acknowledge and agree that your usage of this feature is subject to the applicable additional terms.

docker_image configuration block has the following attributes:

  • url - URL for the Docker image
  • basic_auth - (Optional) basic_auth.username and basic_auth.password for Docker repository. Docker registry credentials are encrypted when they are stored in Databricks internal storage and when they are passed to a registry upon fetching Docker images at cluster launch. However, other authenticated and authorized API users of this workspace can access the username and password.

Example usage with azurerm_container_registry and docker_registry_image, that you can adapt to your specific use-case:

resource "docker_registry_image" "this" {
  name = "${azurerm_container_registry.this.login_server}/sample:latest"
  build {
    # ...
  }
}

resource "databricks_cluster" "this" {
  # ...
  docker_image {
    url = docker_registry_image.this.name
    basic_auth {
      username = azurerm_container_registry.this.admin_username
      password = azurerm_container_registry.this.admin_password
    }
  }
}

Attribute Reference

In addition to all arguments above, the following attributes are exported:

  • id - Canonical unique identifier for the cluster.
  • default_tags - (map) Tags that are added by Databricks by default, regardless of any custom_tags that may have been added. These include: Vendor: Databricks, Creator: <username_of_creator>, ClusterName: <name_of_cluster>, ClusterId: <id_of_cluster>, Name:
  • state - (string) State of the cluster.

Access Control

  • databricks_group and databricks_user can control which groups or individual users can create clusters.
  • databricks_cluster_policy can control which kinds of clusters users can create.
  • Users, who have access to Cluster Policy, but do not have an allow_cluster_create argument set would still be able to create clusters, but within the boundary of the policy.
  • databricks_permissions can control which groups or individual users can Manage, Restart or Attach to individual clusters.
  • instance_profile_arn (AWS only) can control which data a given cluster can access through cloud-native controls.

Import

The resource cluster can be imported using cluster id.

$ terraform import databricks_cluster.this <cluster-id>