Skip to main content
Version: Next

Protobuf Schemas

The datahub-protobuf module is designed to be used with the Java Emitter, the input is a compiled protobuf binary *.protoc files and optionally the corresponding *.proto source code. You can supply a file with multiple nested messages to be processed. If you have a file with multiple non-nested messages, you will need to separate them out into different files or supply the root message, as otherwise we will only process the first one.

Supported Features

The following protobuf features are supported and are translated into descriptions, tags, properties and terms on a dataset.

* C++/C style code comments on Messages and Fields
* Nested Types
* Scalar Values
* Well Known Type Wrappers (i.e. DoubleValue, FloatValue, StringValue)
* Enumerations
* Oneof
* Maps
* Extensions
* Web links
* Parsing of GitHub team names and slack channel references

Usage

Protobuf Compile Options

In order to support parsing comments the following option flags should be used during protoc compilation.

protoc --include_imports --include_source_info --descriptor_set_out=MyProto.protoc MyProto.proto

Code Example

Given an input stream of the protoc binary and the emitter the minimal code is shown below.

import com.linkedin.common.FabricType;
import com.linkedin.common.AuditStamp;
import com.linkedin.common.urn.CorpuserUrn;
import datahub.client.rest.RestEmitter;
import datahub.protobuf.ProtobufDataset;

RestEmitter emitter;
InputStream protocInputStream;

AuditStamp auditStamp = new AuditStamp()
.setTime(System.currentTimeMillis())
.setActor(new CorpuserUrn("datahub"));

ProtobufDataset dataset = ProtobufDataset.builder()
.setDataPlatformUrn(new DataPlatformUrn("kafka"))
.setProtocIn(protocInputStream)
.setAuditStamp(auditStamp)
.setFabricType(FabricType.DEV)
.build();

dataset.getAllMetadataChangeProposals().flatMap(Collection::stream).forEach(mcpw -> emitter.emit(mcpw, null).get());

Additionally, the raw protobuf source can be included as well as information to allow parsing of additional references to GitHub and Slack in the source code comments.

ProtobufDataset dataset = ProtobufDataset.builder()
.setDataPlatformUrn(new DataPlatformUrn("kafka"))
.setSchema(" my raw protobuf schema ")
.setProtocIn(protocInputStream)
.setAuditStamp(auditStamp)
.setFabricType(FabricType.DEV)
.setGithubOrganization("myOrg")
.setSlackTeamId("SLACK123")
.build();

Protobuf Extensions

In order to extract even more metadata from the protobuf schema we can extend the FieldOptions and MessageOptions to be able to annotate Messages and Fields with arbitrary information. This information can then be emitted as DataHub primary key information, tags, glossary terms or properties on the dataset.

An annotated protobuf schema would look like the following, except for the is_primary_key all annotations are configurable for individual needs.

Note: Extending FieldOptions and MessageOptions does not change the messages themselves. The metadata is not included in messages being sent over the wire.

syntax = "proto3";
import "meta.proto";

message Department {
int32 id = 1 [(meta.fld.is_primary_key) = true];
string name = 2;
}

message Person {
option(meta.msg.type) = ENTITY;
option(meta.msg.classification_enum) = HighlyConfidential;
option(meta.msg.team) = "TeamB";
option(meta.msg.bool_feature) = true;
option(meta.msg.alert_channel) = "#alerts";

string name = 1 [(meta.fld.classification) = "Classification.HighlyConfidential"];

int32 id = 2
[(meta.fld.is_primary_key) = true];

string email = 3
[(meta.fld.classification_enum) = Confidential];

Department dept = 4;

string test_coverage = 5
[(meta.fld.product_type_bool) = true, (meta.fld.product_type) = "my type", (meta.fld.product_type_enum) = EVENT];
}

meta.proto

In order to use the annotations above, create a proto file called meta.proto. Feel free to customize the kinds of metadata and how it is emitted to DataHub for your use cases.

syntax = "proto3";
package meta;

import "google/protobuf/descriptor.proto";

/*
This is assigned to metadata fields. It describes how the metadata field should be represented
in DataHub. This enum must be used in the `meta` package. Multiple can be used for the same
metadata annotation. This allows a single piece of information to be captured in DataHub
as a property, tag and/or term.

Tags can be strings, enums, or booleans
Terms can be strings or enums
Properties should be strings

*/
enum DataHubMetadataType {
PROPERTY = 0; // Datahub Custom Property
TAG = 1; // Datahub Tag
TAG_LIST = 2; // Datahub Tags from comma delimited string
TERM = 3; // Datahub Term
OWNER = 4; // Datahub Owner
DOMAIN = 5; // Datahub Domain
DEPRECATION = 6; // Datahub Deprecation
}

/*
Example below: The following is not required for annotation processing. This is an example
of creating an annotation using an enum.
*/

enum MetaEnumExample {
UNKNOWN = 0;
ENTITY = 1;
EVENT = 2;
}

// Assuming Glossary Term defined from bootstrap example
enum Classification {
HighlyConfidential = 0;
Confidential = 1;
Sensitive = 2;
}

FieldOptions

Define possible annotations on fields and how they are exported to DataHub.


message fld {
extend google.protobuf.FieldOptions {
// Required: Mark option field with how to export to DataHub in one or more places.
repeated meta.DataHubMetadataType type = 6000;

/*
Examples below: The following is not required for annotation processing.
*/

// Set true if the field is a primary key. This works for any boolean with `primary_key` in it.
bool is_primary_key = 6010;

// Extract classification field option as a Term, either works
string classification = 6001 [(meta.fld.type) = TERM];
meta.Classification classification_enum = 6002 [(meta.fld.type) = TERM];

// Expose this option as a tag on the field.
string product_type = 70004 [(meta.fld.type) = TAG];
bool product_type_bool = 70005 [(meta.fld.type) = TAG];
meta.MetaEnumExample product_type_enum = 70006 [(meta.fld.type) = TAG];
}
}

MessageOptions

Define possible annotations on messages and how they are exported to DataHub.


message msg {
extend google.protobuf.MessageOptions {
/*
Examples below: The following is not required for annotation processing.
*/

// Place the classification term at the Message/Dataset level, either string or enum is supported
string classification = 4000 [(meta.fld.type) = TERM, (meta.fld.type) = PROPERTY];
meta.Classification classification_enum = 4001 [(meta.fld.type) = TERM, (meta.fld.type) = PROPERTY];

// Attach these Message/Dataset options as a tag and property.
string product = 5001 [(meta.fld.type) = TAG, (meta.fld.type) = PROPERTY];
string project = 5002 [(meta.fld.type) = TAG, (meta.fld.type) = PROPERTY];
string team = 5003 [(meta.fld.type) = OWNER, (meta.fld.type) = PROPERTY];

string domain = 60003 [(meta.fld.type) = DOMAIN, (meta.fld.type) = PROPERTY];
meta.MetaEnumExample type = 60004 [(meta.fld.type) = TAG, (meta.fld.type) = PROPERTY];
bool bool_feature = 60005 [(meta.fld.type) = TAG];
string alert_channel = 60007 [(meta.fld.type) = PROPERTY];

repeated string deprecation_note = 60008 [(meta.fld.type) = DEPRECATION, (meta.fld.type) = PROPERTY];
uint64 deprecation_time = 60009 [(meta.fld.type) = DEPRECATION, (meta.fld.type) = PROPERTY];
}
}

DataHubMetadataType

DataHubMetadataTypeStringBoolEnumRepeatedUint64
PROPERTYXXXX
TAGXXX
TAG_LISTX
TERMXX
OWNERXX
DOMAINXX
DEPRECATIONX (notes)X (notes)X (time)
PROPERTY

Custom properties can be captured as key/value pairs where the protobuf option name is the key and the option value is the option's value.

For example, generating a custom property with key prop1 and value value1.

   message msg {
extend google.protobuf.MessageOptions {
string prop1 = 5000 [(meta.fld.type) = PROPERTY];
}
}

message Message {
option(meta.msg.prop1) = "value1";
}

Booleans are converted to a value of either true or false.

   message msg {
extend google.protobuf.MessageOptions {
bool prop1 = 5000 [(meta.fld.type) = PROPERTY];
}
}

message Message {
option(meta.msg.prop1) = true;
}

Enum values are similarly converted to their string representation.

   enum MetaEnumExample {
UNKNOWN = 0;
ENTITY = 1;
EVENT = 2;
}

message msg {
extend google.protobuf.MessageOptions {
MetaEnumExample prop1 = 5000 [(meta.fld.type) = PROPERTY];
}
}

message Message {
option(meta.msg.prop1) = ENTITY;
}

Repeated values will be collected and the value will be stored as a serialized json array. The following example would result in the value of ["a","b","c"].

   message msg {
extend google.protobuf.MessageOptions {
repeated string prop1 = 5000 [(meta.fld.type) = PROPERTY];
}
}

message Message {
option(meta.msg.prop1) = "a";
option(meta.msg.prop1) = "b";
option(meta.msg.prop1) = "c";
}
TAG & TAG_LIST

The tag list assumes a string that contains the comma delimited values of the tags. In the example below, tags would be added as a, b, c.

   message msg {
extend google.protobuf.MessageOptions {
string tags = 5000 [(meta.fld.type) = TAG_LIST];
}
}

message Message {
option(meta.msg.tags) = "a, b, c";
}

Tags could also be represented as separate boolean options. Only the true options result in tags. In this example, a single tag of tagA would be added to the dataset.

   message msg {
extend google.protobuf.MessageOptions {
bool tagA = 5000 [(meta.fld.type) = TAG];
bool tagB = 5001 [(meta.fld.type) = TAG];
}
}

message Message {
option(meta.msg.tagA) = true;
option(meta.msg.tagB) = false;
}

Alternatively, tags can be separated into different fields with the option name as a dot delimited prefix. The following would produce two tags with values of tagA.a and tagB.a.

   message msg {
extend google.protobuf.MessageOptions {
string tagA = 5000 [(meta.fld.type) = TAG];
string tagB = 5001 [(meta.fld.type) = TAG];
}
}

message Message {
option(meta.msg.tagA) = "a";
option(meta.msg.tagB) = "a";
}

The dot delimited prefix also works with enum types where the prefix is the enum type name. In this example two tags are created, MetaEnumExample.ENTITY.

  enum MetaEnumExample {
UNKNOWN = 0;
ENTITY = 1;
EVENT = 2;
}

message msg {
extend google.protobuf.MessageOptions {
MetaEnumExample tag = 5000 [(meta.fld.type) = TAG];
}
}

message Message {
option(meta.msg.tag) = ENTITY;
}

In addition, tags can be added to fields as well as messages. The following is a consolidated example for all the possible tag options on fields.

  enum MetaEnumExample {
UNKNOWN = 0;
ENTITY = 1;
EVENT = 2;
}

message fld {
extend google.protobuf.FieldOptions {
string tags = 6000 [(meta.fld.type) = TAG_LIST];
string tagString = 6001 [(meta.fld.type) = TAG];
bool tagBool = 6002 [(meta.fld.type) = TAG];
MetaEnumExample tagEnum = 6003 [(meta.fld.type) = TAG];
}
}

message Message {
uint32 my_field = 1
[(meta.fld.tags) = "a, b, c",
(meta.fld.tagString) = "myTag",
(meta.fld.tagBool) = true,
(meta.fld.tagEnum) = ENTITY];
}
TERM

Terms are specified by either a fully qualified string value or an enum where the enum type's name is the first element in the fully qualified term name.

The following example shows both methods, either of which would result in the term Classification.HighlyConfidential being applied.

   enum Classification {
HighlyConfidential = 0;
Confidential = 1;
Sensitive = 2;
}

message msg {
extend google.protobuf.MessageOptions {
Classification term = 5000 [(meta.fld.type) = TERM];
string class = 5001 [(meta.fld.type) = TERM];
}
}

message Message {
option(meta.msg.term) = HighlyConfidential;
option(meta.msg.class) = "Classification.HighlyConfidential";
}

The following is a consolidated example for the possible field level term options.

   enum Classification {
HighlyConfidential = 0;
Confidential = 1;
Sensitive = 2;
}

message fld {
extend google.protobuf.FieldOptions {
Classification term = 5000 [(meta.fld.type) = TERM];
string class = 5001 [(meta.fld.type) = TERM];
}
}

message Message {
uint32 my_field = 1
[(meta.fld.term) = HighlyConfidential,
(meta.fld.class) = "Classification.HighlyConfidential"];
}
OWNER

One or more owners can be specified and can be any combination of corpUser and corpGroup entities. The default entity type is corpGroup. By default, the ownership type is set to technical_owner, see the second example for setting the ownership type.

The following example assigns the ownership to a group of myGroup and a user called myName.

   message msg {
extend google.protobuf.MessageOptions {
repeated string owner = 5000 [(meta.fld.type) = OWNER];
}
}

message Message {
option(meta.msg.owner) = "corpUser:myName";
option(meta.msg.owner) = "myGroup";
}

In this example, the option name determines the ownership type. User myName is assigned as the Technical Owner and myGroup as the Data Steward.

   message msg {
extend google.protobuf.MessageOptions {
repeated string technical_owner = 5000 [(meta.fld.type) = OWNER];
repeated string data_steward = 5001 [(meta.fld.type) = OWNER];
}
}

message Message {
option(meta.msg.technical_owner) = "corpUser:myName";
option(meta.msg.data_steward) = "myGroup";
}
DOMAIN

Set the domain id for the dataset. The domain should exist already. Note that the id of the domain is the value. If not specified during domain creation it is likely a random string.

   message msg {
extend google.protobuf.MessageOptions {
string domain = 5000 [(meta.fld.type) = DOMAIN];
}
}

message Message {
option(meta.msg.domain) = "engineering";
}
DEPRECATION

Deprecation of fields and messages are natively supported by protobuf options. The standard "Deprecation" aspect is used for a dataset generated from a protobuf message. Field deprecation adds a tag with the following urn urn:li:tag:deprecated (red, #FF000).

   message msg {
extend google.protobuf.MessageOptions {
repeated string deprecation_note = 5620 [(meta.fld.type) = DEPRECATION];
uint64 deprecation_time = 5621 [(meta.fld.type) = DEPRECATION];
}
}

message Message {
option deprecated = true;
option (meta.msg.deprecation_note) = "Deprecated for this other message.";
option (meta.msg.deprecation_note) = "Drop in replacement.";
option (meta.msg.deprecation_time) = 1649689387;
}

The field deprecation tag works without definition in meta.proto using the native protobuf option.

message Message {
uint32 my_field = 1 [deprecated = true];
}

Installation

Follow the specific instructions for your build system to declare a dependency on the appropriate version of the package.

Note: Check the Maven repository for the latest version of the package before following the instructions below.

Gradle

Add the following to your build.gradle.

implementation 'io.acryl:datahub-protobuf:__version__'

Maven

Add the following to your pom.xml.

<!-- https://mvnrepository.com/artifact/io.acryl/datahub-protobuf -->
<dependency>
<groupId>io.acryl</groupId>
<artifactId>datahub-protobuf</artifactId>
<!-- replace __version__ with the latest version number -->
<version>__version__</version>
</dependency>

Example Application (embedded)

An example application Proto2DataHub is included as part of this project. You can also set up a standalone project that works with the protobuf-gradle-plugin, see the standalone example project as an example of such a project.

Usage

Standalone Application: Proto2DataHub

shell
java -jar build/libs/datahub-protobuf-0.8.45-SNAPSHOT.jar --help
usage: Proto2DataHub
--datahub_api <arg> [Optional] The API endpoint for DataHub GMS.
(defaults to https://localhost:8080)
--datahub_token <arg> [Optional] The authentication token for
DataHub API access. (defaults to empty)
--datahub_user <arg> [Optional] The datahub user to attribute this
ingestion to. (defaults to ..)
--descriptor <arg> [Required] The generated protobuf descriptor
file. Typically a single .dsc file for the
repo or a .protoc file (1:1 with each src
file)
--directory <arg> [Optional if using --file] The root directory
containing protobuf source files.
--env <arg> [Optional] The environment to attach all
entities to. Typically, DEV, PROD etc.
(defaults to DEV)
--exclude <arg> [Optional] Exclude patterns to avoid
processing all source files, separated by ,.
Typically used with --directory option.
Follows glob patterns: e.g. --exclude
"build/**,generated/**" will exclude all files
in the build and generated directories under
the rootDirectory given by the --directory
option
--file <arg> [Optional if using --directory] The protobuf
source file. Typically a .proto file.
--filename <arg> [Required if using transport file] Filename to
write output to.
--github_org <arg> [Optional] The GitHub organization that this
schema repository belongs to. We will
translate comments in your protoc files like
@datahub-project/data-team to GitHub team urls
like:
https://github.com/orgs/datahub-project/teams/
data-team
--help Print this help message
--platform <arg> [Optional] The data platform to produce
schemas for. e.g. kafka, snowflake, etc.
(defaults to kafka)
--slack_id <arg> [Optional] The Slack team id if your protobuf
files contain comments with references to
channel names. We will translate comments like
#data-eng in your protobuf file to slack urls
like:
https://slack.com/app_redirect?channel=data-en
g&team=T1234 following the documentation at
(https://api.slack.com/reference/deep-linking#
deep-linking-into-your-slack-app__opening-a-ch
annel-by-name-or-id) The easiest way to find
your Slack team id is to open your workspace
in your browser. It should look something
like:
https://app.slack.com/client/TUMKD5EGJ/... In
this case, the team-id is TUMKD5EGJ.
--subtype [Optional] A custom subtype to attach to all
entities produced. e.g. event, schema, topic
etc.(Default is schema)
--transport <arg> [Optional] What transport to use to
communicate with DataHub. Options are: rest
(default), kafka and file.

You can run it like a standard java jar application:


java -jar build/libs/datahub-protobuf-0.8.45-SNAPSHOT.jar --descriptor ../datahub-protobuf-example/build/descriptors/main.dsc --directory ../datahub-protobuf-example/schema/protobuf/ --transport rest

or using gradle

../../../gradlew run --args="--descriptor ../datahub-protobuf-example/build/descriptors/main.dsc --directory ../datahub-protobuf-example/schema/protobuf/ --transport rest"

Result:

java -jar build/libs/datahub-protobuf-0.8.45-SNAPSHOT.jar --descriptor ../datahub-protobuf-example/build/descriptors/main.dsc --directory ../datahub-protobuf-example/schema/protobuf/ --transport rest
SLF4J: Failed to load class "org.slf4j.impl.StaticLoggerBinder".
SLF4J: Defaulting to no-operation (NOP) logger implementation
SLF4J: See http://www.slf4j.org/codes.html#StaticLoggerBinder for further details.
✅ Successfully emitted 90 events for 5 files to DataHub REST

You can also route results to a file by using the --transport file --filename events.json options.

Important Flags

Here are a few important flags to use with this command

  • --env : Defaults to DEV, you should use PROD once you have ironed out all the issues with running this command.
  • --platform: Defaults to Kafka (as most people use protobuf schema repos with Kafka), but you can provide a custom platform name for this e.g. (schema_repo or <company_name>_schemas). If you use a custom platform, make sure to provision the custom platform on your DataHub instance with a logo etc, to get a native experience. See how to use the put platform command to accomplish this.
  • --subtype : This gives your entities a more descriptive category than Dataset in the UI. Defaults to schema, but you might find topic, event or message more descriptive.

Example Application (separate project)

The standalone example project shows you how you can create an independent project that uses this as part of a build task.

Sample Usage:

export DATAHUB_API=...
export DATAHUB_TOKEN=...

# Optional parameters
# export DATAHUB_ENV=PROD
# export DATAHUB_GITHUBORG=datahub-project
# export DATAHUB_SLACKID=

# publishSchema task will publish all the protobuf files into DataHub
./gradlew publishSchema