protoc-gen-bq-schema is a plugin for ProtocolBuffer compiler. It converts messages written in .proto format into schema files in JSON for BigQuery. So you can reuse existing data definitions in .proto for BigQuery with this plugin.

go install github.com/GoogleCloudPlatform/protoc-gen-bq-schema@latest

protoc --bq-schema_out=path/to/outdir [--bq-schema_opt=single-message] foo.proto

protoc and protoc-gen-bq-schema commands must be found in $PATH.

The generated JSON schema files are suffixed with .schema and their base names are named after their package names and bq_table_name options.

If you do not already have the standard google protobuf libraries in your proto_path, you'll need to specify them directly on the command line (and potentially need to copy bq_schema.proto into a proto_path directory as well), like this:

protoc --bq-schema_out=path/to/out/dir foo.proto --proto_path=. --proto_path=<path_to_google_proto_folder>/src

Suppose that we have the following foo.proto.

syntax = "proto2";
package foo;
import "bq_table.proto";
import "bq_field.proto";

message Bar {
  option (gen_bq_schema.bigquery_opts).table_name = "bar_table";

  message Nested {
    repeated int32 a = 1;
  }

  // Description of field a -- this is an int32
  required int32 a = 1;

  // Nested b structure
  optional Nested b = 2;

  // Repeated c string
  repeated string c = 3;

  optional bool d = 4 [(gen_bq_schema.bigquery).ignore = true];

  // TIMESTAMP (uint64 in proto) - required in BigQuery
  optional uint64 e = 5 [
    (gen_bq_schema.bigquery) = {
      require: true
      type_override: 'TIMESTAMP'
    }
  ];
}

message Baz {
  required int32 a = 1;
}

protoc --bq-schema_out=. foo.proto will generate a file named foo/bar_table.schema. The message foo.Baz is ignored because it doesn't have option gen_bq_schema.bigquery_opts.

protoc --bq-schema_out=. --bq-schema_opt=single-message single_message.proto will generate a file named foo/single_message.schema. The message foo.Baz is also ignored because it is not the first message in the file.

protoc-gen-bq-schema now supports policyTags. You can define a Policy Tag for a field in .proto file.

Suppose that you have the following test_table.proto

syntax = "proto3";
package foo;
import "bq_table.proto";
import "bq_field.proto";

message TestTable{
    option (gen_bq_schema.bigquery_opts).table_name = "test_table";

    int32 a = 1 [
        (gen_bq_schema.bigquery) = {
          require: true
          policy_tags : "private"
        }
      ];

    string b = 2 [(gen_bq_schema.bigquery).policy_tags="public"];

    message Nested {
        int32 a = 1 [(gen_bq_schema.bigquery) = {
            require: true
            policy_tags : "private"
            }
        ];

        string b = 2;
    }

    repeated Nested nested = 3 [(gen_bq_schema.bigquery).require = true];

    message EmptyMessage {}

    repeated EmptyMessage hasMessage = 4;
}

protoc --bq-schema_out=. test_table.proto will generate a file named foo/test_table.schema. The field hasMessage is ignored because the message EmptyMessage is empty.

It will generate the following JSON schema

[
 {
  "name": "a",
  "type": "INTEGER",
  "mode": "REQUIRED",
  "policyTags": {
   "names": [
    "private"
   ]
  }
 },
 {
  "name": "b",
  "type": "STRING",
  "mode": "NULLABLE",
  "policyTags": {
   "names": [
    "public"
   ]
  }
 },
 {
  "name": "nested",
  "type": "RECORD",
  "mode": "REQUIRED",
  "fields": [
   {
    "name": "a",
    "type": "INTEGER",
    "mode": "REQUIRED",
    "policyTags": {
     "names": [
      "private"
     ]
    }
   },
   {
    "name": "b",
    "type": "STRING",
    "mode": "NULLABLE"
   }
  ]
 }
]

The policy tag name provided in test_table.proto file is taken as it is. According to Google Docs, the policy tag string should be of the following format

projects/project-id/locations/location/taxonomies/taxonomy-id/policyTags/policytag-id

protoc-gen-bq-schema is licensed under the Apache License version 2.0. This is not an official Google product.