Configuring the Hive Metastore Service in Kubernetes#

The starburst-hive Helm chart configures the Hive Metastore Service, and optionally the backend database in the cluster detailed in the following sections. The following Dell Data Analytics Engine, powered by Starburst Enterprise platform (SEP) connectors and features require a HMS:

Use your registry credentials, and follow best practices by creating an override file for changes to default values as desired.

In addition to the configuration properties described in this document, you can also use the base Hive connector’s metastore configuration properties, the Thrift metastore configuration properties, and the AWS Glue metastore configuration properties as needed, depending on your environment.

Before you begin#

Get the latest starburst-hive Helm chart as described in our installation guide with the configured registry access.

Configure the Hive Metastore#

There are several top-level nodes in the HMS Helm chart that you must modify for a minimum HMS configuration:

  • serviceAccountName

  • resources

  • database

  • expose

  • hiveMetastoreWarehouseDir

  • hdfs

  • objectStorage

As with SEP, we strongly suggest that you initially deploy HMS with the minimum configuration described in this topic and ensure that it deploys and is accessible before making any additional customizations described in our reference documentation.

Note

Store customizations in a separate file containing only changed values as recommended in our best practices guide. In this topic for example, customizations are stored in a file named hms-values.yaml that is used in the Helm upgrade command.

Configure resources and service account#

Ensure that the following top-level nodes of the Helm chart have the correct values to reflect your environment:

  • serviceAccountName: we strongly recommend using a service account for the pod.

  • resources: ensure that the CPU and memory sizes are appropriate for your instance type.

The default values for the resources node are as follows:

heapSizePercentage: 85

resources:
  requests:
    memory: "1Gi"
    cpu: 1
  limits:
    memory: "1Gi"
    cpu: 1

Caution

We strongly suggest leaving the heapSizePercentage at the default value.

Configure the PostgreSQL backend database#

The configuration properties for the internal PostgreSQL backend database are found in the database top-level node. As a minimum configuration, you must ensure that the following are set correctly for your environment:

database:
  internal:
    databaseName: hive
    databasePassword: HivePass1234
    databaseUser: hive
    driver: org.postgresql.Driver
    port: 5432
  type: internal

Note

Alternatively, you can use an external backend database for production usage that you must manage yourself.

The following table lists the available backend database configuration properties:

Internal HMS backend database configuration#

Node name

Description

database.type

Set to internal to use a database in the k8s cluster, managed by the chart

database.internal.image

Docker container images used for the PostgreSQL server

database.internal.volume

Storage volume to persist the database. The default configuration requests a new persistent volume (PV).

database.internal.volume.persistentVolumeClaim

The default configuration, which requests a new persistent volume (PV).

database.internal.volume.existingVolumeClaim

Alternative volume configuration, which use existing volume claim by referencing the name as the value in quotes, e.g., "my_claim".

database.internal.volume.emptyDir

Alternative volume configuration, which configures an empty directory on the pod, keeping in mind that a pod replacement loses the database content.

database.internal.resources

database.internal.databaseName

Name of the internal database

database.internal.databaseUser

User to connect to the internal database

database.internal.databasePassword

Password to connect to internal database

database.internal.envFrom

YAML sequence of mappings to define Secret or Configmap as a source of environment variables for the PostgreSQL container.

database.internal.env

YAML sequence of mappings to define two keys environment variables for the PostgreSQL container.

Note

The database.resources node is separate from the top-level resources node. It defines the resources available to the backing database itself, not the HMS server.

Examples#

OpenShift deployments often do not have access to pull from the default Docker registry library/postgres. You can replace it with an image from the Red Hat registry, which requires additional environment variables set with the parameter database.internal.env:

database:
  type: internal
  internal:
    image:
       repository: "registry.redhat.io/rhscl/postgresql-96-rhel7"
       tag: "latest"
    env:
      - name: POSTGRESQL_DATABASE
        value: "hive"
      - name: POSTGRESQL_USER
        value: "hive"
      - name: POSTGRESQL_PASSWORD
        value: "HivePass1234"

Another option is to create a Secret (ex. postgresql-secret) containing variables needed by postgresql which are mentioned in previous code block, and pass it to the container with envFrom parameter:

database:
  type: internal
  internal:
    image:
       repository: "registry.redhat.io/rhscl/postgresql-96-rhel7"
       tag: "latest"
    envFrom:
      - secretRef:
          name: postgresql-secret

External backend database for HMS#

This section shows the setup for using of an external PostgreSQL, MySQL, Oracle, or Microsoft SQL Server database. You must provide the necessary details for the external server, and ensure that it can be reached from the k8s cluster pod. Set the database.type to external and configure the connection properties:

database:
  type: external
  external:
    driver:
    jdbcUrl:
    password:
    setPasswordsViaEnvFrom: false
    user:
External HMS backend database configuration#

Node name

Description

database.type

Set to external to use an existing PostgreSQL, MySQL, Oracle, or SQL Server database outside the cluster.

database.external.jdbcUrl

JDBC URL to connect to the external database as required by the database and used driver, including hostname and port. Ensure you use a valid JDBC URL as required by the PostgreSQL, MySQL, Oracle, or SQL Server driver. Typically, the syntax requires the host, port, and database name as follows:

  • For PostgreSQL: jdbc:postgresql://host:port/database.

  • For MySQL: jdbc:mysql://host:port/database.

  • For Oracle, using the thin or OCI driver: jdbc:oracle:<driver>:@//host:port/<service_name>.

  • For SQL Server: The database name must be passed as an environment variable HIVE_METASTORE_DB_NAME and the JDBC connection string syntax jdbc:sqlserver://host:port.

database.external.driver

Valid values are as follows:

  • For PostgreSQL: org.postgresql.Driver.

  • For an external MySQL or a compatible database: com.mysql.jdbc.Driver.

  • For Oracle: oracle.jdbc.OracleDriver.

  • For SQL Server: com.microsoft.sqlserver.jdbc.SQLServerDriver.

database.external.user

Database user name to access the external database using JDBC.

database.external.password

Password for the user configured to access the external database using JDBC.

Expose the pod to the outside network#

The expose section of the starburst-hive Helm chart works similarly to the SEP server expose section. Differences are isolated to the configured default values. Additionally, ingress is not supported, as the HMS service uses the TCP protocol and not HTTP.

By default, the HMS is available at the hostname hive and port 9083. As a result, the default Thrift URL for the cluster is thrift://hive:9083. Ensure to adapt your configured catalogs to use the correct Thrift URL. You can use the URL for any catalog:

catalog:
  datalake: |
    connector.name=hive
    hive.metastore.uri=thrift://hive:9083

The default type is clusterIp:

expose:
  type: "clusterIp"
  clusterIp:
    name: "hive"
    ports:
      http:
        port: 9083

You can also use nodePort:

expose:
  type: "nodePort"
  nodePort:
    name: "hive"
    ports:
      http:
        port: 9083
        nodePort: 30083

Additionally, you can use loadBalancer:

expose:
  type: "loadBalancer"
  loadBalancer:
  name: "hive"
  IP: ""
  ports:
    http:
      port: 9083
  annotations: {}
  sourceRanges: []

Configure storage#

Configure a connection to the Hive Metastore Service (HMS), the Hadoop Distributed File System (HDFS), or an object store using the following three top-level nodes:

  • hiveMetastoreWarehouseDir

  • hdfs

  • objectStorage

Note

You must also configure the catalog with the appropriate credentials.

The default configuration for each of these properties is empty.

Add the location of your Hive metastore’s warehouse directory to the hiveMetastoreWarehouseDir node to enable the HMS to store metadata and gather statistics. In the hdfs node, add the hadoopUserName you use to connect to the warehouse directory.

There are several templates for configuring object storage in the objectStorage node. For example, you can define how to connect to S3:

objectStorage:
awsS3:
  accessKey:
  endpoint:
  pathStyleAccess: false
  region:
  secretKey:

The Helm chart also includes templates for Azure, Azure Data Lake, and Google object storage.

Note

For AWS, you can provide access to S3 using secrets, or by using IAM credentials attached to the metastore pod.

The following table lists the available storage configuration properties:

HMS storage configuration#

Node name

Description

hiveMetastoreWarehouseDir

The location of your Hive Metastore’s warehouse directory. For example, s3://example/hive-default-warehouse/.

hdfs.hadoopUserName

User name for Hadoop HDFS access

objectStorage.awsS3.*

Configuration for AWS S3 access

objectStorage.awsS3.region

AWS region name

objectStorage.awsS3.endpoint

AWS S3 endpoint, for example http[s]://<bucket>.s3-<AWS-region>.amazonaws.com.

objectStorage.awsS3.accessKey

Name of the access key for AWS S3

objectStorage.awsS3.secretKey

Name of the secret key for AWS S3

objectStorage.awsS3.pathStyleAccess

objectStorage.gs.*

Configuration for Google Storage access

objectStorage.gs.cloudKeyFileSecret

Name of the secret with the file containing the access key to the cloud storage. The key of the secret must be named key.json.

objectStorage.azure.*:

Configuration for Microsoft Azure storage systems

objectStorage.azure.abfs.*

Configuration for Azure Blob Filesystem (ABFS)

objectStorage.azure.abfs.authType

Authentication to access ABFS, Valid values areaccessKey or oauth, configuration in the following properties

objectStorage.azure.abfs.accessKey.*

Configuration for access key authentication to ABFS

objectStorage.azure.abfs.accessKey.storageAccount

Name of the ABFS account to access

objectStorage.azure.abfs.accessKey.accessKey

Actual access key to use for ABFS access

objectStorage.azure.abfs.oauth.*

Configuration for OAuth authentication to ABFS

objectStorage.azure.abfs.oauth.clientId

Client identifier for OAuth authentication

objectStorage.azure.abfs.oauth.secret

Secret for OAuth

objectStorage.azure.abfs.oauth.endpoint

Endpoint URL for OAuth

objectStorage.azure.wasb.*

Configuration for Windows Azure Storage Blob (WASB)

objectStorage.azure.wasb.storageAccount

Name of the storage account to use for WASB

objectStorage.azure.wasb.accessKey

Key to access WASB

objectStorage.azure.adl

Configuration for Azure Data Lake (ADL)

objectStorage.azure.adl.oauth2.*

Configuration for OAuth authentication to ADL

objectStorage.azure.adl.oauth2.clientId

Client identifier for OAuth access to ADL

objectStorage.azure.adl.oauth2.credential

Credential for OAuth access to ADL

objectStorage.azure.adl.oauth2.refreshUrl:

Refresh URL for the OAuth access to ADL

More information about the configuration options is available in the following resources:

Metastore configuration for Avro#

In order to enable Avro tables when using Hive 3.x, you need to add the following property definition to the Hive metastore configuration file hive-site.xml:

<property>
     <name>metastore.storage.schema.reader.impl</name>
     <value>org.apache.hadoop.hive.metastore.SerDeStorageSchemaReader</value>
 </property>

For more information about additional files, see Adding files.

Configure TLS (optional)#

Note

This is separate from configuring TLS on SEP itself.

If your organization uses TLS, you can enable and configure your HMS to work with it. The most straightforward way to handle TLS is to terminate TLS at the load balancer or ingress, using a signed certificate. We strongly suggest this method, which requires no additional configuration in the HMS.

If you choose not to handle TLS using that method, you can instead configure it in the expose top-level node of the HMS Helm chart:

expose:
  type: "[clusterIp|nodePort|loadBalancer|ingress]"

The default type is clusterIp. For details on configuring each of these types, see exposing the pod to the outside network.

Additional settings#

Server start up configuration#

You can configure a startup shell script for the HMS using the following variables:

  • initFile

  • extraArguments

initFile#

Use initfile to pass a shell script to run before HMS is launched. The content of the script must be an inline string. The original startup command is passed as the first argument at the end of the script as exec "$@". Use exec "$1" to any additional arguments.

extraArguments#

Use extraArguments to pass a list of additional arguments to the initFile script.

The following example shows how you can use initFile and extraArguments to run a custom startup script. The initFile script must end with exec "$@":

initFile: |
  #!/bin/bash
  echo "Custom init for $2"
  exec "$@"
extraArguments:
  - TEST_ARG

Docker image and registry#

The Helm chart for the HMS uses a similar configuration for its Docker image and registry section as the Helm chart for SEP.

image:
  pullPolicy: "IfNotPresent"
  repository: "harbor.starburstdata.net/starburstdata/hive"
  tag: "3.1.3-e.3"

registryCredentials:
  enabled: false
  password:
  registry:
  username:

imagePullSecrets:

Additional volumes#

Additional volumes may be necessary for persisting files. These can be defined in additionalVolumes. None are defined by default:

additionalVolumes: []

You can add one or more volumes supported by k8s to all nodes in the cluster.

Specify a path to create a directory to store the keys for your ConfigMap or Secret.

You may also specify an optional subPath parameter which takes an optional key in the ConfigMap or Secret volume you create. If you specify subPath, a key named subPath from your ConfigMap or Secret is mounted as a file within the directory specified in path.

additionalVolumes:
  - path: /mnt/InContainer
    volume:
      emptyDir: {}
  - path: /etc/hive/conf/test_config.txt
    subPath: test_config.txt
    volume:
      configMap:
        name: "configmap-in-volume"

Node assignment#

You can configure your cluster to use a specific node and pod for the HMS:

nodeSelector: {}
tolerations: []
affinity: {}

Our SEP configuration documentation contains examples and resources to help you configure these YAML nodes.

Annotations#

You can annotate your deployment or pods using the following variables:

  • deploymentAnnotations

  • podAnnotations

Security context#

You can optionally configure security contexts to specify privileges and access control settings for your HMS pods.

securityContext:

If you do not want to set the serviceContext for the default service account, you can restrict it by configuring the service account for the HMS pod.

Environment variables#

You can pass environment variables to your HMS container using the same variables as the internal database:

envFrom: []
env: []

Both variables are specified as mapping sequences. For example:

envFrom:
  - secretRef:
      name: my-secret-with-vars
env:
  - name: MY_VARIABLE
    value: some-value