Skip to main content
Version: 0.13.0

Metadata Events

DataHub makes use a few important Kafka events for operation. The most notable of these include

  1. Metadata Change Proposal
  2. Metadata Change Log (Versioned + Timeseries)
  3. Platform Event

Each event is originally authored using PDL, a modeling language developed by LinkedIn, and then converted into their Avro equivalents, which are used when writing and reading the events to Kafka.

In the document, we'll describe each of these events in detail - including notes about their structure & semantics.

Metadata Change Proposal (MCP)

A Metadata Change Proposal represents a request to change to a specific aspect on an enterprise's Metadata Graph. Each MCP provides a new value for a given aspect. For example, a single MCP can be emitted to change ownership or documentation or domains or deprecation status for a data asset.

Emission

MCPs may be emitted by clients of DataHub's low-level ingestion APIs (e.g. ingestion sources) during the process of metadata ingestion. The DataHub Python API exposes an interface for easily sending MCPs into DataHub.

The default Kafka topic name for MCPs is MetadataChangeProposal_v1.

Consumption

DataHub's storage layer actively listens for new Metadata Change Proposals, attempts to apply the requested change to the Metadata Graph.

Schema

NameTypeDescriptionOptional
entityUrnStringThe unique identifier for the Entity being changed. For example, a Dataset's urn.False
entityTypeStringThe type of the entity the new aspect is associated with. This corresponds to the entity name in the DataHub Entity Registry, for example 'dataset'.False
entityKeyAspectObjectThe key struct of the entity that was changed. Only present if the Metadata Change Proposal contained the raw key struct.True
changeTypeStringThe change type. CREATE, UPSERT and DELETE are currently supported.False
aspectNameStringThe entity aspect which was changed.False
aspectObjectThe new aspect value. Null if the aspect was deleted.True
aspect.contentTypeStringThe serialization type of the aspect itself. The only supported value is application/json.False
aspect.valueStringThe serialized aspect. This is a JSON-serialized representing the aspect document originally defined in PDL. See https://github.com/datahub-project/datahub/tree/master/metadata-models/src/main/pegasus/com/linkedin for more.False
systemMetadataObjectThe new system metadata. This includes the the ingestion run-id, model registry and more. For the full structure, see https://github.com/datahub-project/datahub/blob/master/metadata-models/src/main/pegasus/com/linkedin/mxe/SystemMetadata.pdlTrue

The PDL schema can be found here.

Examples

An MCP representing a request to update the 'ownership' aspect for a particular Dataset:

{
"entityType": "dataset",
"entityUrn": "urn:li:dataset:(urn:li:dataPlatform:hdfs,SampleHdfsDataset,PROD)",
"changeType": "UPSERT",
"aspectName": "ownership",
"aspect": {
"value": "{\"owners\":[{\"type\":\"DATAOWNER\",\"owner\":\"urn:li:corpuser:datahub\"}],\"lastModified\":{\"actor\":\"urn:li:corpuser:datahub\",\"time\":1651516640488}}",
"contentType": "application/json"
},
"systemMetadata": {
"lastObserved": 1651516640493,
"runId": "no-run-id-provided",
"registryName": "unknownRegistry",
"registryVersion": "0.0.0.0-dev",
"properties": null
}
}

Note how the aspect payload is serialized as JSON inside the "value" field. The exact structure of the aspect is determined by its PDL schema. (For example, the ownership schema)

Metadata Change Log (MCL)

A Metadata Change Log represents any change which has been made to the Metadata Graph. Metadata Change Log events are emitted to Kafka immediately after writing the change the durable storage.

There are 2 flavors of Metadata Change Log: versioned and timeseries. These correspond to the type of aspects which were updated for a given change. Versioned aspects are those which represent the "latest" state of some attributes, for example the most recent owners of an asset or its documentation. Timeseries aspects are those which represent events related to an asset that occurred at a particular time, for example profiling of a Dataset.

Emission

MCLs are emitted when any change is made to an entity on the DataHub Metadata Graph, this includes writing to any aspect of an entity.

Two distinct topics are maintained for Metadata Change Log. The default Kafka topic name for versioned aspects is MetadataChangeLog_Versioned_v1 and for timeseries aspects is MetadataChangeLog_Timeseries_v1.

Consumption

DataHub ships with a Kafka Consumer Job (mae-consumer-job) which listens for MCLs and uses them to update DataHub's search and graph indices, as well as to generate derived Platform Events (described below).

In addition, the Actions Framework consumes Metadata Change Logs to power its Metadata Change Log event API.

Schema

NameTypeDescriptionOptional
entityUrnStringThe unique identifier for the Entity being changed. For example, a Dataset's urn.False
entityTypeStringThe type of the entity the new aspect is associated with. This corresponds to the entity name in the DataHub Entity Registry, for example 'dataset'.False
entityKeyAspectObjectThe key struct of the entity that was changed. Only present if the Metadata Change Proposal contained the raw key struct.True
changeTypeStringThe change type. CREATE, UPSERT and DELETE are currently supported.False
aspectNameStringThe entity aspect which was changed.False
aspectObjectThe new aspect value. Null if the aspect was deleted.True
aspect.contentTypeStringThe serialization type of the aspect itself. The only supported value is application/json.False
aspect.valueStringThe serialized aspect. This is a JSON-serialized representing the aspect document originally defined in PDL. See https://github.com/datahub-project/datahub/tree/master/metadata-models/src/main/pegasus/com/linkedin for more.False
previousAspectValueObjectThe previous aspect value. Null if the aspect did not exist previously.True
previousAspectValue.contentTypeStringThe serialization type of the aspect itself. The only supported value is application/jsonFalse
previousAspectValue.valueStringThe serialized aspect. This is a JSON-serialized representing the aspect document originally defined in PDL. See https://github.com/datahub-project/datahub/tree/master/metadata-models/src/main/pegasus/com/linkedin for more.False
systemMetadataObjectThe new system metadata. This includes the the ingestion run-id, model registry and more. For the full structure, see https://github.com/datahub-project/datahub/blob/master/metadata-models/src/main/pegasus/com/linkedin/mxe/SystemMetadata.pdlTrue
previousSystemMetadataObjectThe previous system metadata. This includes the the ingestion run-id, model registry and more. For the full structure, see https://github.com/datahub-project/datahub/blob/master/metadata-models/src/main/pegasus/com/linkedin/mxe/SystemMetadata.pdlTrue
createdObjectAudit stamp about who triggered the Metadata Change and when.False
created.timeNumberThe timestamp in milliseconds when the aspect change occurred.False
created.actorStringThe URN of the actor (e.g. corpuser) that triggered the change.

The PDL schema for can be found here.

Examples

An MCL corresponding to a change in the 'ownership' aspect for a particular Dataset:

{
"entityType": "dataset",
"entityUrn": "urn:li:dataset:(urn:li:dataPlatform:hdfs,SampleHdfsDataset,PROD)",
"changeType": "UPSERT",
"aspectName": "ownership",
"aspect": {
"value": "{\"owners\":[{\"type\":\"DATAOWNER\",\"owner\":\"urn:li:corpuser:datahub\"}],\"lastModified\":{\"actor\":\"urn:li:corpuser:datahub\",\"time\":1651516640488}}",
"contentType": "application/json"
},
"previousAspectValue": {
"value": "{\"owners\":[{\"owner\":\"urn:li:corpuser:jdoe\",\"type\":\"DATAOWNER\"},{\"owner\":\"urn:li:corpuser:datahub\",\"type\":\"DATAOWNER\"}],\"lastModified\":{\"actor\":\"urn:li:corpuser:jdoe\",\"time\":1581407189000}}",
"contentType": "application/json"
},
"systemMetadata": {
"lastObserved": 1651516640493,
"runId": "no-run-id-provided",
"registryName": "unknownRegistry",
"registryVersion": "0.0.0.0-dev",
"properties": null
},
"previousSystemMetadata": {
"lastObserved": 1651516415088,
"runId": "file-2022_05_02-11_33_35",
"registryName": null,
"registryVersion": null,
"properties": null
},
"created": {
"time": 1651516640490,
"actor": "urn:li:corpuser:datahub",
"impersonator": null
}
}

Note how the aspect payload is serialized as JSON inside the "value" field. The exact structure of the aspect is determined by its PDL schema. (For example, the ownership schema)

Platform Event (PE)

A Platform Event represents an arbitrary business-logic event emitted by DataHub. Each Platform Event has a name which determines its contents.

Types

  • Entity Change Event (entityChangeEvent): The most important Platform Event is named Entity Change Event, and represents a log of semantic changes (tag addition, removal, deprecation change, etc) that have occurred on DataHub. It is used an important component of the DataHub Actions Framework.

All registered Platform Event types are declared inside the DataHub Entity Registry (entity-registry.yml).

Emission

All Platform Events are generated by DataHub itself during normal operation.

PEs are extremely dynamic - they can contain arbitrary payloads depending on the name. Thus, can be emitted in a variety of circumstances.

The default Kafka topic name for all Platform Events is PlatformEvent_v1.

Consumption

The Actions Framework consumes Platform Events to power its Entity Change Event API.

Schema

NameTypeDescriptionOptional
headerObjectHeader fieldsFalse
header.timestampMillisLongThe time at which the event was generated.False
nameStringThe name / type of the event.False
payloadObjectThe event itself.False
payload.contentTypeStringThe serialization type of the event payload. The only supported value is application/json.False
payload.valueStringThe serialized payload. This is a JSON-serialized representing the payload document originally defined in PDL. See https://github.com/datahub-project/datahub/tree/master/metadata-models/src/main/pegasus/com/linkedin for more.False

The full PDL schema can be found here.

Examples

An example of an 'Entity Change Event' Platform Event that is emitted when a new owner is added to a Dataset:

{
"header": {
"timestampMillis": 1655390732551
},
"name": "entityChangeEvent",
"payload": {
"value": "{\"entityUrn\":\"urn:li:dataset:abc\",\"entityType\":\"dataset\",\"category\":\"OWNER\",\"operation\":\"ADD\",\"modifier\":\"urn:li:corpuser:jdoe\",\"parameters\":{\"ownerUrn\":\"urn:li:corpuser:jdoe\",\"ownerType\":\"BUSINESS_OWNER\"},\"auditStamp\":{\"actor\":\"urn:li:corpuser:jdoe\",\"time\":1649953100653}}",
"contentType": "application/json"
}

Note how the actual payload for the event is serialized as JSON inside the 'payload' field. The exact structure of the Platform Event is determined by its PDL schema. (For example, the Entity Change Event schema)

Failed Metadata Change Proposal (FMCP)

When a Metadata Change Proposal cannot be processed successfully, the event is written to a dead letter queue in an event called Failed Metadata Change Proposal (FMCP).

The event simply wraps the original Metadata Change Proposal and an error message, which contains the reason for rejection. This event can be used for debugging any potential ingestion issues, as well as for re-playing any previous rejected proposal if necessary.

Emission

FMCEs are emitted when MCEs cannot be successfully committed to DataHub's storage layer.

The default Kafka topic name for FMCPs is FailedMetadataChangeProposal_v1.

Consumption

No active consumers.

Schema

The PDL schema can be found here.

Deprecated Events

DataHub ships with a set of deprecated events, which were historically used for proposing and logging changes to the Metadata Graph.

Each event in this category was deprecated due to its inflexibility - namely the fact that the schemas had to be updated when a new aspect was introduced. These events have since been replaced by the more flexible events described above (Metadata Change Proposal, Metadata Change Log).

It is not recommended to build dependencies on deprecated events.

Metadata Change Event (MCE)

A Metadata Change Event represents a request to change multiple aspects for the same entity. It leverages a deprecated concept of Snapshot, which is a strongly-typed list of aspects for the same entity.

A MCE is a "proposal" for a set of metadata changes, as opposed to MAE, which is conveying a committed change. Consequently, only successfully accepted and processed MCEs will lead to the emission of a corresponding MAE / MCLs.

Emission

MCEs may be emitted by clients of DataHub's low-level ingestion APIs (e.g. ingestion sources) during the process of metadata ingestion.

The default Kafka topic name for MCEs is MetadataChangeEvent_v4.

Consumption

DataHub's storage layer actively listens for new Metadata Change Events, attempts to apply the requested changes to the Metadata Graph.

Schema

The PDL schema can be found here.

Examples

An example of an MCE emitted to change the 'ownership' aspect for an Entity:

{
"proposedSnapshot": {
"com.linkedin.pegasus2avro.metadata.snapshot.DatasetSnapshot": {
"urn": "urn:li:dataset:(urn:li:dataPlatform:hive,SampleHiveDataset,PROD)",
"aspects": [
{
"com.linkedin.pegasus2avro.common.Ownership": {
"owners": [
{
"owner": "urn:li:corpuser:jdoe",
"type": "DATAOWNER",
"source": null
},
{
"owner": "urn:li:corpuser:datahub",
"type": "DATAOWNER",
"source": null
}
],
"lastModified": {
"time": 1581407189000,
"actor": "urn:li:corpuser:jdoe",
"impersonator": null
}
}
}
]
}
}
}

Metadata Audit Event (MAE)

A Metadata Audit Event captures changes made to one or multiple metadata aspects associated with a particular entity, in the form of a metadata snapshot (deprecated) before the change, and a metadata snapshot after the change.

Every source-of-truth for a particular metadata aspect is expected to emit a MAE whenever a change is committed to that aspect. By ensuring that, any listener of MAE will be able to construct a complete view of the latest state for all aspects. Furthermore, because each MAE contains the "after image", any mistake made in emitting the MAE can be easily mitigated by emitting a follow-up MAE with the correction. By the same token, the initial bootstrap problem for any newly added entity can also be solved by emitting a MAE containing all the latest metadata aspects associated with that entity.

Emission

Note: In recent versions of DataHub (mid 2022), MAEs are no longer actively emitted, and will soon be completely removed from DataHub. Use Metadata Change Log instead.

MAEs are emitted once any metadata change has been successfully committed into DataHub's storage layer.

The default Kafka topic name for MAEs is MetadataAuditEvent_v4.

Consumption

No active consumers.

Schema

The PDL schema can be found here.

Examples

An example of an MAE emitted representing a change made to the 'ownership' aspect for an Entity (owner removed):

{
"oldSnapshot": {
"com.linkedin.pegasus2avro.metadata.snapshot.DatasetSnapshot": {
"urn": "urn:li:dataset:(urn:li:dataPlatform:hive,SampleHiveDataset,PROD)",
"aspects": [
{
"com.linkedin.pegasus2avro.common.Ownership": {
"owners": [
{
"owner": "urn:li:corpuser:jdoe",
"type": "DATAOWNER",
"source": null
},
{
"owner": "urn:li:corpuser:datahub",
"type": "DATAOWNER",
"source": null
}
],
"lastModified": {
"time": 1581407189000,
"actor": "urn:li:corpuser:jdoe",
"impersonator": null
}
}
}
]
}
},
"newSnapshot": {
"com.linkedin.pegasus2avro.metadata.snapshot.DatasetSnapshot": {
"urn": "urn:li:dataset:(urn:li:dataPlatform:hive,SampleHiveDataset,PROD)",
"aspects": [
{
"com.linkedin.pegasus2avro.common.Ownership": {
"owners": [
{
"owner": "urn:li:corpuser:datahub",
"type": "DATAOWNER",
"source": null
}
],
"lastModified": {
"time": 1581407189000,
"actor": "urn:li:corpuser:jdoe",
"impersonator": null
}
}
}
]
}
}
}

Failed Metadata Change Event (FMCE)

When a Metadata Change Event cannot be processed successfully, the event is written to a dead letter queue in an event called Failed Metadata Change Event (FMCE).

The event simply wraps the original Metadata Change Event and an error message, which contains the reason for rejection. This event can be used for debugging any potential ingestion issues, as well as for re-playing any previous rejected proposal if necessary.

Emission

FMCEs are emitted when MCEs cannot be successfully committed to DataHub's storage layer.

Consumption

No active consumers.

Schema

The PDL schema can be found here.

The default Kafka topic name for FMCEs is FailedMetadataChangeEvent_v4.