Version 2.4

A GSQL query is a sequence of data retrieval-and-computation statements executed as a single operation. Users can write queries to explore a data graph however they like, to read and make computations on the graph data along the way, to update the graph, and to deliver resulting data. A query is analogous to a user-defined procedure or function: it can have one or more input parameters, and it can produce output in two ways: by returning a value or by printing. A query can be run in one of three ways:

  1. Define and run an unnamed query immediately:

    1. INTERPRET QUERY: execute the query's statements

    Alternately, there is also a built-in REST++ endpoint to interpret a query string: POST /gsqlserver/interpreted_query See the RESTPP API User Guide for details.

  2. Define a named query and then run it.

    1. CREATE QUERY: define the functionality of the query

    2. INTERPRET QUERY: execute the query with input values

  3. Define a named query, compile it to optimize performance, and then run it.

    1. CREATE QUERY: define the functionality of the query

    2. INSTALL QUERY: compile the query

    3. RUN QUERY: execute the query with input values

There are some limitations to Interpreted mode. See the section on INTERPRET QUERY and the appendix section Interpreted GSQL Limitations.


CREATE QUERY defines the functionality of a query on a given graph schema.

A query has a name, a parameter list, the name of the graph being queried, an optional RETURNS type (see Section "RETURN Statement" for more details), an optional specifier for the output api, and a body. The body consists of an optional sequence of typedefs , followed by an optional sequence of declarations, then followed by one or more statements. The body defines the behavior of the query.

The DISTRIBUTED option applies only to installations where the graph has been distributed across a cluster . If specified, the query will run with a different execution model which may give better performance for queries which traverse a large portion of the cluster. Not all GSQL query language features are supported in DISTRIBUTED mode. For details, see the separate document: Distributed Query Mode.

Typedefs allow the programmer to define custom types for use within the body. The declarations support definition of accumulators (see Chapter "Accumulators" for more details) and global/local variables. All accumulators and global variables must be declared before any statements. There are various types of statements that can be used within the body. Typically, the core statement(s) in the body of a query is one or more SELECT, UPDATE, INSERT, DELETE statements. The language supports conditional statements such as an IF statement as well as looping constructs such as WHILE and FOREACH. It also supports calling functions, assigning variables, printing, and modifying the graph data.

The query body may include calls to other queries. That is, the other queries are treated as subquery functions. See the subsection on "Queries as Functions".

Query Parameter and Return Types

This table lists the supported data types for input parameters and return values.

Statement Types

A statement is a standalone instruction that expresses an action to be carried out. The most common statements are data manipulation language ( DML) statements . DML statements include the SELECT, UPDATE, INSERT INTO, DELETE FROM, and DELETE statements.

A GSQL query has two levels of statements. The upper-level statement type is called query-body-level statement , or query-body statement for short. This statement type is part of either the top-level block or a query-body control flow block. For example, each of the statements at the top level directly under CREATE QUERY is a query-body statement. If one of the statements is a CASE statement with several THEN blocks, each of the statements in the THEN blocks is also a query-body statement. Each query-body statement ends with a semicolon.

The lower-level statement type is called DML-sub-level statement or DML-sub-statement for short. This statement type is used inside certain query-body DML statements, to define particular data manipulation actions. DML-sub-statements are comma-separated. There is no comma or semicolon after the last DML-sub-statement in a block. For example, one of the top-level statements is a SELECT statement, each of the statements in its ACCUM clause is a DML-sub-statement. If one of those DML-sub-statements is a CASE statement, each of the statement in the THEN blocks is a DML-sub-statement.

There is some overlap in the types. For example, an assignStmt can be used either at the query-body level or the DML-sub-level.

Guidelines for understanding statement type hierarchy:

  • Top-level statements are Query-Body type (each statement ending with a semicolon).

  • The statements within a DML statement are DML-sub statements (comma-separated list).

  • The blocks within a Control Flow statement have the same type as the entire Control Flow statement itself.

Here is a descriptive list of query-body statements:

Here is a descriptive list of DML-sub-statements:


INTERPRET QUERY runs a query by translating it line-by-line. This is in contrast to the 2-step flow: INSTALL to pre-translate and optimize a query. then RUN to execute the installed query. The basic trade-off between INTERPRET QUERY and INSTALL/RUN QUERY is as follows:


    • Starts running immediately but may take longer to finish than running an INSTALLed query.

    • Suitable for ad hoc exploration of a graph or when developing and debugging an application, and rapid experimentation is desired.

    • Supports most but not all of the features of the full GSQL query language. See the Appendix section Interpreted GSQL Limitations.


    • Takes up to a minute to INSTALL.

    • Runs faster than INTERPRET, from only a few percent faster to twice as fast.

    • Should always be used for production environments with fixed queries.

There are two GSQL syntax options for Interpreted GSQL: Immediate mode and Interpret-only mode. In addition there is also a predefined RESTful endpoint for running interpreted GSQL: POST /gsqlserver/interpreted_query. The query body is sent as the payload of the request. The syntax is like the Immediate query option, except that it is possible to provide parameters, using the query string of the endpoint's request URL. The example below shows a parameterized query using the POST /gsqlserver/interpreted_query endpoint. For more details, see the RESTPP API User Guide.

Immediate Mode: Define and Interpret

This syntax is similar in concept to SQL queries. Queries are not named, do not accept parameters, and are not saved after being run. Syntax differences from compiled GSQL:

  1. The keyword CREATE is replaced with INTERPRET.

  2. The query is executed immediately by the INTERPRET statement. The INSTALL and RUN statements are not used.

  3. Parameters are not accepted.

Compare the example below to the example in the Create Query section:

  • No query name, no parameters, no RETURN

  • Because no parameter is allowed, the parameter uid is set within the query.

Interpret-Only Mode: Interpret a Predefined Query

This syntax is like RUN query, except

  1. The keyword RUN is replaced with INTERPRET.

  2. Run options are not supported.


A query must be installed before it can be executed. The INSTALL QUERY command will install the queries listed:

INSTALL QUERY queryName1, queryName2, ...

It can also install all uninstalled queries, using either of the following commands: INSTALL QUERY * INSTALL QUERY ALL

The following options are available:

-force Option

Reinstall the query even if the system indicates the query is already installed. This is useful for overwriting an installation that is corrupted or otherwise outdated, without having to drop and then recreate the query. If this option is not used, the GSQL shell will refuse to re-install a query that is already installed.


During standard installation, the user-defined queries are dynamically linked to the GSQL language code. Anytime after INSTALL QUERY has been performed, another statement, INSTALL QUERY -OPTIMIZE can be executed. The names of the individual queries are not needed. This operation optimizes all previously installed queries, reducing their run times by about 20%. Optimize a query if query run time is more important to you than query installation time.




If you have a distributed database deployment, installing the query in DISTRIBUTED mode can increase performance for single queries - using a single worker from each available machine to yield results. Certain cases may benefit more from this option than others -- more detailed information is available on the next page: Distributed Query Mode .

Running a Query

Installing a query creates a REST++ endpoint. Once a query is installed, there are two ways of executing a query. One way is through the GSQL shell: RUN QUERY query_name( parameterValues ) .

The query response time can be reduced by directly submitting an HTTP request to the REST++ server: send a GET request to " http://server_ip:9000/query/graphname/queryname ". If the REST++ server is local, then server_ip is localhost . The query parameter values are either included directly in the query string of the HTTP request's URL or supplied using a data payload.

The following two curl commands are each equivalent to the RUN QUERY command above. The first gives the parameter values in the query string in a URL. This example illustrates the simple format for primitive data types such as INT, DOUBLE, and STRING. The second gives the parameter values through the curl command's data payload -d option.

where RunQueryExPara.dat has the exact string as the query string in the first URL.

By using the data payload option, the user can avoid using a long and complex URL. In fact, to call the same query but with different parameters, only the data payload file contents need to be changed; the HTTP request can be the same. The file loader loads the entire file, appends multiple lines into one, and uses the resulting string as the URL query string. If both a query string and a data payload are given (which we strongly discourage), both are included, where the URL query string's parameter values overwrite the values given in the data payload.

Complex Type Parameter Passing

This subsection describes how to format the complex type parameter values when executing a query by RUN QUERY or curl command. More details about all parameter types are described in Section "Query Parameter Types"

Below are examples.

Payload Size Limit

This data payload option can accept a file up to 128MB by default. To increase this limit to xxx MB, use the following command:

The upper limit of this setting is 1024 MB. Raising the size limit for the data payload buffer reduces the memory available for other operations, so be cautious about increasing this limit.

For more detailed information about REST++ endpoints and requests, see the RESTPP API User Guide .

The following options are available when running a query:

All-Vertex Mode -av Option

Some queries run with all or almost all vertices in a SELECT statement s, e.g. PageRank algorithm. In this case, the graph processing engine can run much more efficiently in all-vertex mode. In the all-vertex mode, all vertices are always selected, and the following actions become ineffective:

  • Filtering with selected vertices or vertex types. The source vertex set must be all vertices.

  • Filtering with the WHERE clause.

  • Filtering with the HAVING clause.

  • Assigning designated vertex or designated type of vertexes. E.g. X = { vertex_type .*}

To run the query in all-vertex mode, use the -av option in shell mode or include __GQUERY__USING_ALL_ACTIVE_MODE=true in the query string of an HTTP request.

Diagnose -d Option

The diagnose option can be turned on in order to produce a diagnostic monitoring log, which contains the processing time of each SELECT block . To turn on the monitoring log, use the -d option in shell mode or __GQUERY__monitor=true in the query string of an HTTP request.

The path of the generated log file will be shown as a part of output message. An example log is shown below:

GSQL Query Output Format

The standard output of GSQL queries is in industry-standard JSON format. A JSON object is an unordered set of key:value pairs , enclosed in curly braces. Among the acceptable data types for a JSONvalue are array and object . A JSON array is an ordered list of values , enclosed in square brackets. Since values can be objects or arrays, JSON supports hierarchical, nested structures. Strings are enclosed in double quotation marks. We also use the term field to refer to a key (or a key:value pair) of a given object.

At the top level of the JSON structure are four required fields ("version", "error", "message", and "results") and one dependent field ("code"). If a query is successful, the value of "error" will be "false", the "message" value will be empty, and the "results" value will be the intended output of the query.If an error or exception occurred during query execution, the "error" value will be "true", the "message" value will be a string message describing the error condition, and the "results" field will be empty. Also, the "code" field will contain an error code.

Beginning with version 2 (v2) of the output specification, an additional top-level field is required: "version". The "version" value is an object with the following fields:

Other top-level objects, such as "code" may appear in certain circumstances. Note that the top-level objects are enclosed in curly braces, meaning that they form an unordered set. They may appear in any order.

Below is an example of the output of a successful query:

The following REST response misspells the name of the endpoint

For backward compatibility, TigerGraph platforms whose principal output API is v2 can also produce output with API v1.

The value of the "results" key-value pair is a sequential list of the data objects specified by the PRINT statements of the query. The list order follows the order of PRINT execution. The detailed format of the PRINT statement results is described in the Chapter "Output Statements".

Changing the Default Output API

and generates the following output:

The following GSQL statement can be used to set the JSON output API configuration.

Currently, the legal values for <version_string> are "v1" and "v2". This statement sets a persistent system parameter. Each version of the TigerGraph platform is pre-configured to what was the latest output API that at the time of release. For example, platform version 1.1 is configured so that each query will produce v2 output by default.


To show the GSQL text of a query, run "SHOW QUERY query_name ". Additionally, the "ls" GSQL command lists all created queries and identifies which queries have been installed.

As of v2.3, the query_name argument can now use * or ? wildcards from Linux globbing, or it can be a regular expression, when preceded by -r. See SHOW: View Parts of the Catalog


To drop a query, run "DROP QUERY query_name ". The query will be uninstalled (if it has been installed) and removed from the dictionary. The GSQL language will refuse to drop an installed query Q if another query R is installed which calls query Q . That is, all calling queries must be dropped before or at the same time that their called subqueries are dropped.

To drop all queries,, either of the following commands can be used: DROP QUERY ALL DROP QUERY *