Skip to main content

DataHub Lite (Experimental)

What is it?

DataHub Lite is a lightweight embeddable version of DataHub with no external dependencies. It is intended to enable local developer tooling use-cases such as simple access to metadata for scripts and other tools. DataHub Lite is compatible with the DataHub metadata format and all the ingestion connectors that DataHub supports. It was built as a reaction to recap to prove that a similar lightweight system could be built within DataHub quite easily. Currently DataHub Lite uses DuckDB under the covers as its default storage layer, but that might change in the future.

Features

  • Designed for the CLI
  • Available as a Python library or REST API
  • Ingest metadata from all DataHub ingestion sources
  • Metadata Reads
    • navigate metadata using a hierarchy
    • get metadata for an entity
    • search / query metadata across all entities
  • Forward metadata automatically to a central GMS or Kafka instance

Architecture

architecture

What is it not?

DataHub Lite is NOT meant to be a replacement for the production Java DataHub server (datahub-gms). It does not offer the full set of API-s that the DataHub GMS server does. The following features are NOT supported:

  • Full-text search with ranking and relevance features
  • Graph traversal of relationships (e.g. lineage)
  • Metadata change stream over Kafka (only forwarding of writes is supported)
  • GraphQL API

Prerequisites

To use datahub lite commands, you need to install acryl-datahub > 0.9.6 (install instructions) and the datahub-lite plugin.

pip install acryl-datahub[datahub-lite]

Importing Metadata

To ingest metadata into DataHub Lite, all you have to do is change the sink: in your recipe to be a datahub-lite instance. See the detailed sink docs here. For example, here is a simple recipe file that ingests mysql metadata into datahub-lite.

# mysql.in.dhub.yaml
source:
type: mysql
config:
host_port: localhost:3306
username: datahub
password: datahub

sink:
type: datahub-lite

By default, lite will create a local instance under ~/.datahub/lite/.

Ingesting metadata into DataHub Lite is as simple as running ingestion: datahub ingest -c mysql.in.dhub.yaml

note

DataHub Lite currently doesn't support stateful ingestion, so you'll have to turn off stateful ingestion in your recipe to use it. This will be fixed shortly.

Forwarding to a central DataHub GMS over REST or Kafka

DataHub Lite can be configured to forward all writes to a central DataHub GMS using either the REST API or the Kafka API. To configure forwarding, add a forward_to section to your DataHub Lite config that conforms to a valid DataHub Sink configuration. Here is an example:

# mysql.in.dhub.yaml with forwarding to datahub-gms over REST
source:
type: mysql
config:
host_port: localhost:3306
username: datahub
password: datahub

sink:
type: datahub-lite
forward_to:
type: datahub-rest
config:
server: "http://datahub-gms:8080"
note

Forwarding is currently best-effort, so there can be losses in metadata if the remote server is down. For a reliable sync mechanism, look at the exporting metadata section to generate a standard metadata file.

Importing from a file

As a convenient short-cut, you can import metadata from any standard DataHub metadata json file (e.g. generated via using a file sink) by issuing a datahub lite import command.

> datahub lite import --file metadata_events.json

Exploring Metadata

The datahub lite group of commands provides a set of capabilities for you to explore the metadata you just ingested.

List (ls)

Listing functions like a directory structure that is customized based on the kind of system being explored. DataHub's metadata is automatically organized into databases, tables, views, dashboards, charts, etc.

note

Using the ls command below is much more pleasant when you have tab completion enabled on your shell. Check out the Setting up Tab Completion section at the bottom of the guide.

> datahub lite ls /
databases
bi_tools
tags
# Stepping down one level
> datahub lite ls /databases
mysql
# Stepping down another level
> datahub lite ls /databases/mysql
instances
...
# At the final level
> datahub lite ls /databases/mysql/instances/default/databases/datahub/tables/
metadata_aspect_v2
metadata_index
# Listing a leaf entity functions just like the unix ls command
> datahub lite ls /databases/mysql/instances/default/databases/datahub/tables/metadata_aspect_v2
metadata_aspect_v2

Read (get)

Once you have located a path of interest, you can read metadata at that entity, by issuing a get. You can additionally filter the metadata retrieved from an entity by the aspect type of the metadata (e.g. to request the schema, filter by the schemaMetadata aspect).

Aside: If you are curious what all types of entities and aspects DataHub supports, check out the metadata model of entities like Dataset, Dashboard etc.

The general template for the get responses looks like:

{
"urn": <urn_of_the_entity>,
<aspect_name>: {
"value": <aspect_value>, # aspect value as a dictionary
"systemMetadata": <system_metadata> # present if details are requested
}
}

Here is what executing a get command looks like:

Get metadata for an entity by path
> datahub lite get --path /databases/mysql/instances/default/databases/datahub/tables/metadata_aspect_v2
{
"urn": "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)",
"container": {
"value": {
"container": "urn:li:container:21d4204e13d5b984c58acad468ecdbdd"
}
},
"status": {
"value": {
"removed": false
}
},
"datasetProperties": {
"value": {
"customProperties": {},
"name": "metadata_aspect_v2",
"tags": []
}
},
"schemaMetadata": {
"value": {
"schemaName": "datahub.metadata_aspect_v2",
"platform": "urn:li:dataPlatform:mysql",
"version": 0,
"created": {
"time": 0,
"actor": "urn:li:corpuser:unknown"
},
"lastModified": {
"time": 0,
"actor": "urn:li:corpuser:unknown"
},
"hash": "",
"platformSchema": {
"com.linkedin.schema.MySqlDDL": {
"tableSchema": ""
}
},
"fields": [
{
"fieldPath": "urn",
"nullable": false,
"type": {
"type": {
"com.linkedin.schema.StringType": {}
}
},
"nativeDataType": "VARCHAR(collation='utf8mb4_bin', length=500)",
"recursive": false,
"isPartOfKey": true
},
{
"fieldPath": "aspect",
"nullable": false,
"type": {
"type": {
"com.linkedin.schema.StringType": {}
}
},
"nativeDataType": "VARCHAR(collation='utf8mb4_bin', length=200)",
"recursive": false,
"isPartOfKey": true
},
{
"fieldPath": "version",
"nullable": false,
"type": {
"type": {
"com.linkedin.schema.NumberType": {}
}
},
"nativeDataType": "BIGINT(display_width=20)",
"recursive": false,
"isPartOfKey": true
},
{
"fieldPath": "metadata",
"nullable": false,
"type": {
"type": {
"com.linkedin.schema.StringType": {}
}
},
"nativeDataType": "LONGTEXT(collation='utf8mb4_bin')",
"recursive": false,
"isPartOfKey": false
},
{
"fieldPath": "systemmetadata",
"nullable": true,
"type": {
"type": {
"com.linkedin.schema.StringType": {}
}
},
"nativeDataType": "LONGTEXT(collation='utf8mb4_bin')",
"recursive": false,
"isPartOfKey": false
},
{
"fieldPath": "createdon",
"nullable": false,
"type": {
"type": {
"com.linkedin.schema.TimeType": {}
}
},
"nativeDataType": "DATETIME(fsp=6)",
"recursive": false,
"isPartOfKey": false
},
{
"fieldPath": "createdby",
"nullable": false,
"type": {
"type": {
"com.linkedin.schema.StringType": {}
}
},
"nativeDataType": "VARCHAR(collation='utf8mb4_bin', length=255)",
"recursive": false,
"isPartOfKey": false
},
{
"fieldPath": "createdfor",
"nullable": true,
"type": {
"type": {
"com.linkedin.schema.StringType": {}
}
},
"nativeDataType": "VARCHAR(collation='utf8mb4_bin', length=255)",
"recursive": false,
"isPartOfKey": false
}
]
}
},
"subTypes": {
"value": {
"typeNames": [
"table"
]
}
}
}

Get metadata for an entity filtered by specific aspect

> datahub lite get --path /databases/mysql/instances/default/databases/datahub/tables/metadata_aspect_v2 --aspect status
{
"urn": "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)",
"status": {
"value": {
"removed": false
}
}
}
note

Using the get command by path is much more pleasant when you have tab completion enabled on your shell. Check out the Setting up Tab Completion section at the bottom of the guide.

Get metadata using the urn of the entity

> datahub lite get --urn "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)" --aspect status
{
"urn": "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)",
"status": {
"value": {
"removed": false
}
}
}
Get metadata with additional details (systemMetadata)
> datahub lite get --path /databases/mysql/instances/default/databases/datahub/tables/metadata_aspect_v2 --aspect status --verbose
{
"urn": "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)",
"status": {
"value": {
"removed": false
},
"systemMetadata": {
"lastObserved": 1673982834666,
"runId": "mysql-2023_01_17-11_13_12",
"properties": {
"sysVersion": 1
}
}
}
}

Point-in-time Queries

DataHub Lite preserves every version of metadata ingested, just like DataHub GMS. You can also query the metadata as of a specific point in time by adding the --asof parameter to your get command.

> datahub lite get "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)" --aspect status --asof 2020-01-01
null

> datahub lite get "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)" --aspect status --asof 2023-01-16
{
"urn": "urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)",
"status": {
"removed": false
}
}

DataHub Lite also allows you to search using queries within the metadata using the datahub lite search command. You can provide a free form search query like: "customer" and DataHub Lite will attempt to find entities that match the name customer either in the id of the entity or within the name fields of aspects in the entities.

> datahub lite search pet
{"id": "urn:li:dataset:(urn:li:dataPlatform:looker,long_tail_companions.explore.long_tail_pets,PROD)", "aspect": "urn", "snippet": null}
{"id": "urn:li:dataset:(urn:li:dataPlatform:looker,long_tail_companions.explore.long_tail_pets,PROD)", "aspect": "datasetProperties", "snippet": "{\"customProperties\": {\"looker.explore.label\": \"Long Tail Pets\", \"looker.explore.file\": \"long_tail_companions.model.lkml\"}, \"externalUrl\": \"https://acryl.cloud.looker.com/explore/long_tail_companions/long_tail_pets\", \"name\": \"Long Tail Pets\", \"tags\": []}"}

You can also query the metadata precisely using DuckDB's JSON extract functions. Writing these functions requires that you understand the DataHub metadata model and how the data is laid out in DataHub Lite.

For example, to find all entities whose datasetProperties aspect includes the view_definition in its customProperties sub-field, we can issue the following command:

> datahub lite search --aspect datasetProperties --flavor exact "metadata -> '$.customProperties' ->> '$.view_definition' IS NOT NULL"
{"id": "urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_MUTEXES,PROD)", "aspect": "datasetProperties", "snippet": "{\"customProperties\": {\"view_definition\": \"CREATE TEMPORARY TABLE `INNODB_MUTEXES` (\\n  `NAME` varchar(4000) NOT NULL DEFAULT '',\\n  `CREATE_FILE` varchar(4000) NOT NULL DEFAULT '',\\n  `CREATE_LINE` int(11) unsigned NOT NULL DEFAULT 0,\\n  `OS_WAITS` bigint(21) unsigned NOT NULL DEFAULT 0\\n) ENGINE=MEMORY DEFAULT CHARSET=utf8\", \"is_view\": \"True\"}, \"name\": \"INNODB_MUTEXES\", \"tags\": []}"}
{"id": "urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.user_variables,PROD)", "aspect": "datasetProperties", "snippet": "{\"customProperties\": {\"view_definition\": \"CREATE TEMPORARY TABLE `user_variables` (\\n `VARIABLE_NAME` varchar(64) NOT NULL DEFAULT '',\\n `VARIABLE_VALUE` varchar(2048) DEFAULT NULL,\\n `VARIABLE_TYPE` varchar(64) NOT NULL DEFAULT '',\\n `CHARACTER_SET_NAME` varchar(32) DEFAULT NULL\\n) ENGINE=MEMORY DEFAULT CHARSET=utf8\", \"is_view\": \"True\"}, \"name\": \"user_variables\", \"tags\": []}"}
{"id": "urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_TABLESPACES_ENCRYPTION,PROD)", "aspect": "datasetProperties", "snippet": "{\"customProperties\": {\"view_definition\": \"CREATE TEMPORARY TABLE `INNODB_TABLESPACES_ENCRYPTION` (\\n `SPACE` int(11) unsigned NOT NULL DEFAULT 0,\\n `NAME` varchar(655) DEFAULT NULL,\\n `ENCRYPTION_SCHEME` int(11) unsigned NOT NULL DEFAULT 0,\\n `KEYSERVER_REQUESTS` int(11) unsigned NOT NULL DEFAULT 0,\\n `MIN_KEY_VERSION` int(11) unsigned NOT NULL DEFAULT 0,\\n `CURRENT_KEY_VERSION` int(11) unsigned NOT NULL DEFAULT 0,\\n `KEY_ROTATION_PAGE_NUMBER` bigint(21) unsigned DEFAULT NULL,\\n `KEY_ROTATION_MAX_PAGE_NUMBER` bigint(21) unsigned DEFAULT NULL,\\n `CURRENT_KEY_ID` int(11) unsigned NOT NULL DEFAULT 0,\\n `ROTATING_OR_FLUSHING` int(1) NOT NULL DEFAULT 0\\n) ENGINE=MEMORY DEFAULT CHARSET=utf8\", \"is_view\": \"True\"}, \"name\": \"INNODB_TABLESPACES_ENCRYPTION\", \"tags\": []}"}

Search will return results that include the id of the entity that matched along with the aspect and the content of the aspect as part of the snippet field. If you just want the id of the entity to be returned, use the --no-details flag.

> datahub lite search --aspect datasetProperties --flavor exact "metadata -> '$.customProperties' ->> '$.view_definition' IS NOT NULL" --no-details
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_SYS_FOREIGN,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_CMPMEM_RESET,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_FT_DEFAULT_STOPWORD,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_SYS_TABLES,PROD)
...
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_SYS_COLUMNS,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.INNODB_FT_CONFIG,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.USER_STATISTICS,PROD)

List Urns (list-urns)

List all the ids in the DataHub Lite instance.

> datahub lite list-urns
urn:li:container:21d4204e13d5b984c58acad468ecdbdd
urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_aspect_v2,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,datahub.metadata_index,PROD)
urn:li:container:aa82e8309ce84acc350640647a54ca3b
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.ALL_PLUGINS,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.APPLICABLE_ROLES,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.CHARACTER_SETS,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.CHECK_CONSTRAINTS,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.COLLATIONS,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.COLLATION_CHARACTER_SET_APPLICABILITY,PROD)
urn:li:dataset:(urn:li:dataPlatform:mysql,information_schema.COLUMNS,PROD)
...

HTTP Server (serve)

DataHub Lite can be run as a lightweight HTTP server, exposing an OpenAPI spec over FastAPI.

> datahub lite serve
INFO: Started server process [33364]
INFO: Waiting for application startup.
INFO: Application startup complete.
INFO: Uvicorn running on http://127.0.0.1:8979 (Press CTRL+C to quit)

OpenAPI docs are available via your browser at the same port: http://localhost:8979

The server exposes similar commands as the lite cli commands over HTTP:

  • entities: list of all entity ids and get metadata for an entity
  • browse: traverse the entity hierarchy in a path based way
  • search: execute search against the metadata

Server Configuration

Configuration for the server is picked up from the standard location for the datahub cli: ~/.datahubenv and can be created using datahub lite init.

Here is a sample config file with the lite section filled out:

gms:
server: http://localhost:8080
token: ''
lite:
config:
file: /Users/<username>/.datahub/lite/datahub.duckdb
type: duckdb
forward_to:
type: datahub-rest
server: "http://datahub-gms:8080

Admin Commands

Export Metadata (export)

The export command allows you to export the contents of DataHub Lite into a metadata events file that you can then send to another DataHub instance (e.g. over REST).

> datahub lite export --file datahub_lite_export.json
Successfully exported 1785 events to datahub_lite_export.json

Clear (nuke)

If you want to clear your DataHub lite instance, you can just issue the nuke command.

> datahub lite nuke
DataHub Lite destroyed at <path>

Use a different file (init)

By default, DataHub Lite will create and use a local duckdb instance located at ~/.datahub/lite/datahub.duckdb. If you want to use a different location, you can configure it using the datahub lite init command.

> datahub lite init --type duckdb --file my_local_datahub.duckdb
Will replace datahub lite config type='duckdb' config={'file': '/Users/<username>/.datahub/lite/datahub.duckdb', 'options': {}} with type='duckdb' config={'file': 'my_local_datahub.duckdb', 'options': {}} [y/N]: y
DataHub Lite inited at my_local_datahub.duckdb

Reindex

DataHub Lite maintains a few derived tables to make access possible via both the native id (urn) as well as the logical path of the entity. The reindex command recomputes these indexes.

Caveat Emptor!

DataHub Lite is a very new project. Do not use it for production use-cases. The API-s and storage formats are subject to change and we get feedback from early adopters. That said, we are really interested in accepting PR-s and suggestions for improvements to this fledgling project.

Advanced Options

Tab Completion

Using the datahub lite commands like ls or get is much more pleasant when you have tab completion enabled on your shell. Tab completion is supported on the command line through the Click Shell completion module. To set up shell completion for your shell, follow the instructions below based on your shell variant:

Option 1: Inline eval (easy, less performant)

Add this to ~/.zshrc:

eval "$(_DATAHUB_COMPLETE=zsh_source datahub)"

Using eval means that the command is invoked and evaluated every time a shell is started, which can delay shell responsiveness. To speed it up, write the generated script to a file, then source that.

Save the script somewhere.

# the additional sed patches completion to be path oriented and not add spaces between each completed token
_DATAHUB_COMPLETE=zsh_source datahub | sed 's;compadd;compadd -S /;' > ~/.datahub-complete.zsh

Source the file in ~/.zshrc.

. ~/.datahub-complete.zsh