Skip to main content
Version: Next

Structured Properties

Why Would You Use Structured Properties?​

Structured properties are a structured, named set of properties that can be attached to logical entities like Datasets, DataJobs, etc. Structured properties have values that are types. Conceptually, they are like “field definitions”.

Learn more about structured properties in the Structured Properties Feature Guide.

Goal Of This Guide​

This guide will show you how to execute the following actions with structured properties.

  • Create structured properties
  • Read structured properties
  • Delete structured properties
  • Add structured properties to a dataset
  • Patch structured properties (add / remove / update a single property)
  • Update structured property with breaking schema changes
  • Search & aggregations using structured properties

Prerequisites​

For this tutorial, you need to deploy DataHub Quickstart and ingest sample data. For detailed information, please refer to Datahub Quickstart Guide.

Additionally, you need to have the following tools installed according to the method you choose to interact with DataHub:

Install the relevant CLI version. Forms are available as of CLI version 0.13.1. The corresponding DataHub Cloud release version is v0.2.16.5 Connect to your instance via init:

  • Run datahub init to update the instance you want to load into.
  • Set the server to your sandbox instance, https://{your-instance-address}/gms.
  • Set the token to your access token.

Create Structured Properties​

The following code will create a structured property io.acryl.privacy.retentionTime. 

mutation createStructuredProperty {
  createStructuredProperty(
    input: {
      id: "retentionTime",
      qualifiedName:"retentionTime",
      displayName: "Retention Time",
      description: "Retention Time is used to figure out how long to retain records in a dataset",
      valueType: "urn:li:dataType:datahub.number",
      allowedValues: [
        {numberValue: 30, description: "30 days, usually reserved for datasets that are ephemeral and contain pii"},
        {numberValue: 90, description:"description: Use this for datasets that drive monthly reporting but contain pii"},
        {numberValue: 365, description:"Use this for non-sensitive data that can be retained for longer"}
      ],
      cardinality: SINGLE,
      entityTypes: ["urn:li:entityType:datahub.dataset", "urn:li:entityType:datahub.dataFlow"],
    }
  ) {
    urn
  }
}

Read Structured Properties​

You can see the properties you created by running the following command:

datahub properties get --urn {urn}

For example, you can run datahub properties get --urn urn:li:structuredProperty:io.acryl.privacy.retentionTime. If successful, you should see metadata about your properties returned.

{
  "urn": "urn:li:structuredProperty:io.acryl.privacy.retentionTime",
  "qualified_name": "io.acryl.privacy.retentionTime",
  "type": "urn:li:dataType:datahub.number",
  "description": "Retention Time is used to figure out how long to retain records in a dataset",
  "display_name": "Retention Time",
  "entity_types": [
    "urn:li:entityType:datahub.dataset",
    "urn:li:entityType:datahub.dataFlow"
  ],
  "cardinality": "MULTIPLE",
  "allowed_values": [
    {
      "value": "30",
      "description": "30 days, usually reserved for datasets that are ephemeral and contain pii"
    },
    {
      "value": "90",
      "description": "Use this for datasets that drive monthly reporting but contain pii"
    },
    {
      "value": "365",
      "description": "Use this for non-sensitive data that can be retained for longer"
    }
  ]
}

Set Structured Property To a Dataset​

This action will set/replace all structured properties on the entity. See PATCH operations to add/remove a single property.

mutation upsertStructuredProperties {
  upsertStructuredProperties(
    input: {
      assetUrn: "urn:li:mydataset1",
      structuredPropertyInputParams: [
        {
          structuredPropertyUrn: "urn:li:structuredProperty:mystructuredproperty",
          values: [
            {
              stringValue: "123"
            }
          ]
        }
      ]
    }
  ) {
    properties {
      structuredProperty {
        urn
      }
    }
  }
}

Expected Outcomes​

Once your datasets are uploaded, you can view them in the UI and view the properties associated with them under the Properties tab.

Or you can run the following command to view the properties associated with the dataset:

datahub dataset get --urn {urn}

Read Structured Properties From a Dataset​

For reading all structured properties from a dataset:

query getDataset {
  dataset(urn: "urn:li:dataset:(urn:li:dataPlatform:snowflake,long_tail_companions.ecommerce.customer,PROD)") {
    structuredProperties {
      properties {
        structuredProperty {
          urn
          type
          definition {
            displayName
            description
            allowedValues {
              description
            }
          }
        }
        values {
          ... on StringValue {
            stringValue
          }
          ... on NumberValue {
            numberValue
          }
        }
        valueEntities {
          urn
          type
        }
      }
    }
  }
}

Remove Structured Properties From a Dataset​

For removing a structured property or list of structured properties from a dataset:

mutation removeStructuredProperties {
  removeStructuredProperties(
    input: {
      assetUrn: "urn:li:mydataset1",
      structuredPropertyUrns: ["urn:li:structuredProperty:mystructuredproperty"]
    }
  ) {
    properties {
            structuredProperty {urn}
        }
  }
}

Patch Structured Property Value​

This section will show you how to patch a structured property value - either by removing, adding, or upserting a single property.

Add Structured Property Value​

For this example, we'll extend create a second structured property and apply both properties to the same dataset used previously. After this your system should include both io.acryl.privacy.retentionTime and io.acryl.privacy.retentionTime02.

Let's start by creating the second structured property.

curl -X 'POST' -v \
  'http://localhost:8080/openapi/v2/entity/structuredProperty/urn%3Ali%3AstructuredProperty%3Aio.acryl.privacy.retentionTime02/propertyDefinition' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "qualifiedName": "io.acryl.privacy.retentionTime02",
    "displayName": "Retention Time 02",
    "valueType": "urn:li:dataType:datahub.string",
    "allowedValues": [
        {
            "value": {"string": "foo2"},
            "description": "test foo2 value"
        },
        {
            "value": {"string": "bar2"},
            "description": "test bar2 value"
        }
    ],
    "cardinality": "SINGLE",
    "entityTypes": [
        "urn:li:entityType:datahub.dataset"
    ]
}' | jq

This command will attach one of each of the two properties to our test dataset urn:li:dataset:(urn:li:dataPlatform:hive,SampleHiveDataset,PROD) Specifically, this will set io.acryl.privacy.retentionTime as 60.0 and io.acryl.privacy.retentionTime02 as bar2.

curl -X 'POST' -v \
  'http://localhost:8080/openapi/v2/entity/dataset/urn%3Ali%3Adataset%3A%28urn%3Ali%3AdataPlatform%3Ahive%2CSampleHiveDataset%2CPROD%29/structuredProperties' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "properties": [
    {
      "propertyUrn": "urn:li:structuredProperty:io.acryl.privacy.retentionTime",
      "values": [
        {"double": 60.0}
      ]
    },
    {
      "propertyUrn": "urn:li:structuredProperty:io.acryl.privacy.retentionTime02",
      "values": [
        {"string": "bar2"}
      ]
    }
  ]
}' | jq

Expected Outcomes​

You can see that the dataset now has two structured properties attached to it.

Remove Structured Property Value​

The expected state of our test dataset include 2 structured properties. We'd like to remove the first one (io.acryl.privacy.retentionTime) and preserve the second property. (io.acryl.privacy.retentionTime02).

curl -X 'PATCH' -v \
  'http://localhost:8080/openapi/v2/entity/dataset/urn%3Ali%3Adataset%3A%28urn%3Ali%3AdataPlatform%3Ahive%2CSampleHiveDataset%2CPROD%29/structuredProperties' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json-patch+json' \
  -d '{
        "patch": [
            {
                "op": "remove",
                "path": "/properties/urn:li:structuredProperty:io.acryl.privacy.retentionTime"
            }
        ],
        "arrayPrimaryKeys": {
            "properties": [
                "propertyUrn"
            ]
        }
      }' | jq

The response will show that the expected property has been removed.

{
  "urn": "urn:li:dataset:(urn:li:dataPlatform:hive,SampleHiveDataset,PROD)",
  "aspects": {
    "structuredProperties": {
      "value": {
        "properties": [
          {
            "values": [
              {
                "string": "bar2"
              }
            ],
            "propertyUrn": "urn:li:structuredProperty:io.acryl.privacy.retentionTime02"
          }
        ]
      }
    }
  }
}

Expected Outcomes​

You can see that the first property has been removed and the second property is still present.

Upsert Structured Property Value​

In this example, we'll add the property back with a different value, preserving the existing property.

mutation updateStructuredProperty {
  updateStructuredProperty(
    input: {
      urn: "urn:li:structuredProperty:retentionTime",
      displayName: "Retention Time",
      description: "Retention Time is used to figure out how long to retain records in a dataset",
      newAllowedValues: [
        {
          numberValue: 30,
          description: "30 days, usually reserved for datasets that are ephemeral and contain pii"
        },
        {
          numberValue: 90,
          description: "Use this for datasets that drive monthly reporting but contain pii"
        },
        {
          numberValue: 365,
          description: "Use this for non-sensitive data that can be retained for longer"
        }
      ]
    }
  ) {
    urn
  }
}

Expected Outcomes​

You can see that the first property has been added back with a new value and the second property is still present.

Delete Structured Properties​

There are two types of deletion present in DataHub: hard and soft delete.

SOFT DELETE

A soft deleted Structured Property does not remove any underlying data on the Structured Property entity or the Structured Property's values written to other entities. The soft delete is 100% reversible with zero data loss. When a Structured Property is soft deleted, a few operations are not available.

Structured Property Soft Delete Effects:

  • Entities with a soft deleted Structured Property value will not return the soft deleted properties
  • Updates to a soft deleted Structured Property's definition are denied
  • Adding a soft deleted Structured Property's value to an entity is denied
  • Search filters using a soft deleted Structured Property will be denied
HARD DELETE

A hard deleted Structured Property REMOVES all underlying data for the Structured Property entity and the Structured Property's values written to other entities. The hard delete is NOT reversible.

Structured Property Hard Delete Effects:

  • Structured Property entity is removed
  • Structured Property values are removed via PATCH MCPs on their respective entities
  • Rollback is not possible
  • Elasticsearch index mappings will continue to contain references to the hard deleted property until reindex

Soft Delete​

The following command will soft delete the test property.

datahub delete --urn {urn}

Hard Delete​

The following command will hard delete the test property.

datahub delete --urn {urn} --hard

Index Mappings Cleanup​

After the asynchronous delete of all Structured Property values have been processed, triggered by the above hard delete, it is possible to remove the remaining index mappings. Note that if even 1 Structured Property value remains the mapping will not be removed for a given entity index.

Run the DataHub system-update job (automatically run with every helm upgrade or install and quickstart) with the following environment variables enabled.

This will trigger an ES index which will take time to complete. During the process the entire index is recreated.

ELASTICSEARCH_INDEX_BUILDER_MAPPINGS_REINDEX=true
ENABLE_STRUCTURED_PROPERTIES_SYSTEM_UPDATE=true

Update Structured Property With Breaking Schema Changes​

This section will demonstrate how to make backwards incompatible schema changes. Making backwards incompatible schema changes will remove previously written data.

Breaking schema changes are implemented by setting a version string within the Structured Property definition. This version must be in the following format: yyyyMMddhhmmss, i.e. 20240614080000

IMPORTANT NOTES

Old values will not be retrieve-able after the new Structured Property definition is applied.

The old values will be subject to deletion asynchronously (future work).

In the following example, we'll revisit the retentionTime structured property and apply a breaking change by changing the cardinality from MULTIPLE to SINGLE. Normally this change would be rejected as a backwards incompatible change since values that were previously written may have multiple values written which would no longer be valid.

Edit the previously created definition yaml: Change the cardinality to SINGLE and add a version.

- id: io.acryl.privacy.retentionTime
  # - urn: urn:li:structuredProperty:io.acryl.privacy.retentionTime # optional if id is provided
  qualified_name: io.acryl.privacy.retentionTime # required if urn is provided
  type: number
  cardinality: SINGLE
  version: '20240614080000'
  display_name: Retention Time
  entity_types:
    - dataset # or urn:li:entityType:datahub.dataset
    - dataFlow
  description: "Retention Time is used to figure out how long to retain records in a dataset"
  allowed_values:
    - value: 30
      description: 30 days, usually reserved for datasets that are ephemeral and contain pii
    - value: 90
      description: Use this for datasets that drive monthly reporting but contain pii
    - value: 365
      description: Use this for non-sensitive data that can be retained for longer

Use the CLI to create your properties:

datahub properties upsert -f {properties_yaml}

If successful, you should see Created structured property urn:li:structuredProperty:...

Structured Properties - Search & Aggregation​

Currently Structured Properties can be used to filter search results. This currently excludes fulltext search.

The following examples re-use the two previously defined Structured Properties.

io.acryl.privacy.retentionTime - An example numeric property.

io.acryl.privacy.retentionTime02 - An example string property.

Range Query:

Document should be returned based on the previously assigned value of 60.

query {
    scrollAcrossEntities(
        input: {
            types: DATASET,
            count: 10,
            query: "*",
            orFilters: {
                and: [
                    {
                        field: "structuredProperties.io.acryl.privacy.retentionTime",
                        condition: GREATER_THAN,
                        values: [
                            "45.0"
                        ]
                    }
                ]
            }
        }
    ) {
        searchResults {
            entity {
                urn,
                type
            }
        }
    }
}

Exists Query:

Document should be returned based on the previously assigned value.

query {
  scrollAcrossEntities(
    input: {
      types: DATASET,
      count: 10,
      query: "*",
      orFilters: {
        and: [
          {
            field: "structuredProperties.io.acryl.privacy.retentionTime",
            condition: EXISTS
          }
        ]
      }
    }
  ) {
    searchResults {
      entity {
        urn,
        type
      }
    }
  }
}

Equality Query:

Document should be returned based on the previously assigned value of 'bar2'.

query {
  scrollAcrossEntities(
    input: {
      types: DATASET,
      count: 10,
      query: "*",
      orFilters: {
        and: [
          {
            field: "structuredProperties.io.acryl.privacy.retentionTime02",
            condition: EQUAL
            values: [
              "bar2"
            ]
          }
        ]
      }
    }
  ) {
    searchResults {
      entity {
        urn,
        type
      }
    }
  }
}

Structured Property Aggregations​

Structured properties can also be used in GraphQL's aggregation queries using the same naming convention outlined above for search filter field names. There are currently no aggregation endpoints for OpenAPI.

Aggregation Query:

query {
  aggregateAcrossEntities(
    input: {
      types: [], 
      facets: [
        "structuredProperties.io.acryl.privacy.retentionTime02",
        "structuredProperties.io.acryl.privacy.retentionTime"], 
      query: "*", 
      orFilters: [], 
      searchFlags: {maxAggValues: 100}
    }) {
  facets {
    field
      aggregations {
        value
        count
      }
    }
  }
}

Example Response:

{
  "data": {
    "aggregateAcrossEntities": {
      "facets": [
        {
          "field": "structuredProperties.io.acryl.privacy.retentionTime02",
          "aggregations": [
            {
              "value": "bar2",
              "count": 1
            }
          ]
        },
        {
          "field": "structuredProperties.io.acryl.privacy.retentionTime",
          "aggregations": [
            {
              "value": "60.0",
              "count": 1
            }
          ]
        }
      ]
    }
  },
  "extensions": {}
}