Load from a Data Warehouse

After you have defined a graph schema, you can create a loading job, specify your data sources, and run the job to load data.

The steps are similar whether you are loading from local files, from cloud storage, or any of the other supported sources. We will call out whether a particular step is common for all loading or specific to a data source or loading mode.

Example Schema

This example uses part of the LDBC_SNB schema:

Example schema taken from LDBC_SNB
//Vertex Types:
  gender STRING, birthday DATETIME, creationDate DATETIME, locationIP STRING,
  browserUsed STRING, speaks SET<STRING>, email SET<STRING>)
  locationIP STRING, browserUsed STRING, content STRING, length UINT)
//Edge Types:

Create Data Source Object

A data source code provides a standard interface for all supported data source types, so that loading jobs can be written without regard for the data source.

When you create the object, you specify its details (type, access credentials, etc.) in the form of a JSON object. The JSON object can either be read in from a file or provided inline. Inline mode is required when creating data sources for TigerGraph Cloud instances.

In the following example, we create a data source named s1, and read its configuration information from a file called ds_config.json.

USE GRAPH ldbc_snb
CREATE DATA_SOURCE s1 = "ds_config.json" FOR GRAPH ldbc_snb

Older versions of TigerGraph required a keyword after DATA_SOURCE such as STREAM or KAFKA.

Inline JSON data format when creating a data source
type: <type>,
key: <value>
}" FOR GRAPH ldbc_snb

String literals can be enclosed with a double quote ", triple double quotes """, or triple single quotes '''. Double quotes " in the JSON can be omitted if the key name does not contain a colon : or comma ,.

Alternate quote syntax for inline JSON data
"type": "<type>",
"key": "<value>"
}""" FOR GRAPH ldbc_snb

Key names accept a separator of either a period . or underscore _, so for example, key_name and key.name are both valid key names.

BigQuery is the only data warehouse is supported in v3.9. More data warehouses will be supported in future releases.


TigerGraph’s BigQuery loader makes use of the BigQuery JDBC connector provided by Google, in collaboration with Simba. Use the following configuration for the DATA_SOURCE.

Data source configuration for BigQuery
  "parameters" : {
    "OAuthRefreshToken":"<refresh token>",
    "OAuthClientId":"<client ID>.apps.googleusercontent.com",
    "OAuthClientSecret":"<client secret>"
    #other Simba JDBC parameters
In addition, for large query results, we highly recommended specifying the following parameters:
Addition parameter setting for large BigQuery results
"EnableHighThroughputAPI":"1" -> Storage Read API
"AllowLargeResults":"1" -> Large Query Result Support
"LargeResultDataset":"<target_dataset>" -> Storage for Temp Result
"LargeResultsDatasetExpirationTime":"<time_ms>" -> Expiration of Temp Result

For more information about Simba/Google BigQuery JDBC connection parameters, please refer the BigQuery JDBC Installation and Configuration Guide.

Create a loading job

A loading job tells the database how to construct vertices and edges from data sources. The loading job body has two parts:

  1. DEFINE statements create variables to refer to data sources. These can refer to actual files or be placeholder names. The actual data sources can be given when running the loading job.

  2. LOAD statements specify how to take the data fields from files to construct vertices or edges.

Example loading job from a data warehouse

The following is an example loading job from Google BigQuery.

Example loading job for BigQuery
USE GRAPH ldbc_snb
CREATE LOADING JOB load_data FOR GRAPH ldbc_snb {
DEFINE FILENAME file_Comment =
  "$s1:SELECT *
       FROM tigergraph-ldbc-benchmark.snb_bi_sf01.Comment";
  "$s1:SELECT id, firstName, lastName, gender, birthday, creationDate,
         locationIP, browserUsed, language, email
       FROM tigergraph-ldbc-benchmark.snb_bi_sf01.Person";
DEFINE FILENAME file_Comment_hasCreator_Person =
  "$s1:SELECT *
       FROM tigergraph-ldbc-benchmark.snb_bi_sf01.Comment_hasCreator_Person";
LOAD file_Comment
  TO VERTEX Comment
    VALUES ($1, $0, $2, $3, $4, $5)
    USING separator="|";
LOAD file_Person
  TO VERTEX Person
    VALUES ($1, $2, $3, $4, $5, $0, $6, $7, SPLIT($8,";"), SPLIT($9,";"))
    USING separator="|";
LOAD file_Comment_hasCreator_Person
    VALUES ($1, $2) USING separator="|";

Define filenames

First we define filenames, which are local variables referring to data files (or data objects).

The terms FILENAME and filevar are used for legacy reasons, but a filevar can also be an object in a data object store.
DEFINE FILENAME filevar ["=" file_descriptor ];

The file descriptor can be specified at compile time or at runtime. Runtime settings override compile-time settings:

Specifying file descriptor at runtime
RUN LOADING JOB job_name USING filevar=file_descriptor_override
While a loading job may have multiple FILENAME variables , they must all refer to the same DATA_SOURCE object.

Data warehouse file descriptors

For data warehouses, you run a SQL query to get the data. The file descriptor has three valid formats. You can simply provide the file URI. Or, you can provide optional configuration details, either in a JSON file or as inline JSON content.

DEFINE FILENAME file_name = "$[data source name]:[SQL]";
DEFINE FILENAME file_name = "$[data source name]:[json config file]";
DEFINE FILENAME file_name = "$[data source name]:[inline json content]";

For example:

// Format 1: URI only
DEFINE FILENAME query_person = "$s1:SELECT id,name,gender FROM ldbc.person";

// Format 2: URI and configuration file
DEFINE FILENAME bq_inline_json = """$s1:myfile.json""";

// Format 3: URI and inline JSON
DEFINE FILENAME query_person="""$s1:{
  "query": "SELECT id,name,gender
            FROM ldbc.person where age < 10;
            SELECT id,name,gender
            FROM ldbc.person where age > 50",
  "num.partitions": 6,
  "tasks.max": 2

Filename parameters

These are the required and optional configuration parameters:

Parameter Description Required? Default value


One or more SQL queries separated by commas. To avoid timeout in a large query, you can break it into multiple smaller queries based on the partitioning key. These queries may be assigned to multiple tasks for execution, while the execution order is not guaranteed.




Maximum number of rows to include in a single batch when polling for new data from the query result.




The number of partitions to use. When loading data, each partition is distributed evenly across each node. If one filename contains much more data than others, consider using a larger partition number.




The maximum number of tasks used to execute queries. When query contains multiple queries, you can increase this parameter to execute queries in parallel.




Time interval in ms for periodic executoion of the query.



Specify the data mapping

Next, we use LOAD statements to describe how the incoming data will be loaded to attributes of vertices and edges. Each LOAD statement handles the data mapping, and optional data transformation and filtering, from one filename to one or more vertex and edge types.

LOAD statement syntax
LOAD [ source_object|filevar|TEMP_TABLE table_name ]
  destination_clause [, destination_clause ]*
  [ TAGS clause ] (1)
  [ USING clause ];
1 As of v3.9.3, TAGS are deprecated.

Let’s break down one of the LOAD statements in our example:

Example loading job for local files
LOAD file_Person TO VERTEX Person
    VALUES($1, $2, $3, $4, $5, $0, $6, $7,
       SPLIT($8, ";"), SPLIT($9, ";"))
    USING SEPARATOR="|", HEADER="true", EOL="\n";
  • $0, $1,…​ refer to the first, second, …​ columns in each line a data file.

  • SEPARATOR="|" says the column separator character is the pipe (|). The default is comma (,).

  • HEADER="true" says that the first line in the source contains column header names instead of data. These names can be used instead of the columnn numbers.

  • SPLIT is one of GSQL’s ETL functions. It says that there is a multi-valued column, which has a separator character to mark the subfields in that column.

Refer to Creating a Loading Job in the GSQL Language Reference for descriptions of all the options for loading jobs.

Data Mapping from BigQuery reuslts

The columns of SQL results are joined by a specified separator to form delimited content.

LOAD bq_sql TO VERTEX Comment VALUES ($1, $0, $2, $3, $4, $5) USING separator="|";

In order to load nested or repeated records from BigQuery, some conversion functions need to be applied to the SQL statement.

Querying STRUCT Data

  • Method 1:

    1. Apply the BigQuery TO_JSON_STRING function to the columns of the STRUCT, e.g.,

    2. Flatten the JSON object to CSV format.

  • Method 2:

    • Retrieve the fields from the STRUCT directly, e.g.,

      SELECT col.field1, col.field2, col.field3 FROM table

Querying Arrays

  1. Apply function ARRAY_TO_STRING to the columns of ARRAY type, e.g.,

    SELECT ARRAY_TO_STRING(col_arr,separator) FROM table
  2. In the LOAD statement, use the GSQL SPLIT function.

Run the loading job

Use the command RUN LOADING JOB to run the loading job.

RUN LOADING JOB basic syntax (some options omitted)
RUN LOADING JOB [-noprint] job_name [
  USING filevar [="file_descriptor"][, filevar [="file_descriptor"]]*


By default, the loading job will run in the foreground and print the loading status and statistics after you submit the job. If the -noprint option is specified, the job will run in the background after displaying the job ID and the location of the log file.

filevar list

The optional USING clause may contain a list of file variables. Each file variable may optionally be assigned a file_descriptor, obeying the same format as in CREATE LOADING JOB. This list of file variables determines which parts of a loading job are run and what data files are used.

When a loading job is compiled, it generates one RESTPP endpoint for each filevar and source_object. As a consequence, a loading job can be run in parts. When RUN LOADING JOB is executed, only those endpoints whose filevar or file identifier (GSQL_FILENAME_n) is mentioned in the USING clause will be used. However, if the USING clause is omitted, then the entire loading job will be run.

If a file_descriptor is given, it overrides the file_descriptor defined in the loading job. If a particular filevar is not assigned a file_descriptor either in the loading job or in the RUN LOADING JOB statement, an error is reported and the job exits.

Continuous Loading from Data Warehouses

If EOF="true" (the default), then the query is executed once, and its output will be loaded.

If EOF="false", the query will be executed periodically every poll.interval.ms and its output loaded. This will continuous indefinitely until the job is aborted.

Manage and monitor your loading job

When a loading job starts, the GSQL server assigns it a job ID and displays it for the user to see. There are three key commands to monitor and manage loading jobs:


SHOW LOADING STATUS shows the current status of either a specified loading job or all current jobs, this command should be within the scope of a graph:

GSQL > USE GRAPH graph_name

For each loading job, the above command reports the following information:

  • Loading status

  • Loaded lines/Loaded objects/Error lines

  • Average loading speed

  • Size of loaded data

  • Duration

When inspecting all current jobs with SHOW LOADING STATUS ALL, the jobs in the FINISHED state will be omitted as they are considered to have successfully finished. You can use SHOW LOADING STATUS job_id to check the historical information of finished jobs.

Manage loading job concurrency

See Loading Job Concurrency for how to manage the concurrency of loading jobs.

Known Issues with Loading

TigerGraph does not store NULL values. Therefore, your input data should not contain any NULLs.