graphql-engine/docs/graphql/core/api-reference/schema-api/run-sql.rst
Rikin Kachhia fa90e9d004 docs: update feature version support notes
GitOrigin-RevId: 4627c876b2771c84602f4bfaa1ab0646b2650f12
2021-03-04 19:21:21 +00:00

325 lines
7.8 KiB
ReStructuredText

.. meta::
:description: Execute SQL with the Hasura schema/metadata API
:keywords: hasura, docs, schema/metadata API, API reference, run_sql
.. _schema_api_run_sql:
Schema API Reference: Run SQL (v2.0 and above)
==============================================
.. contents:: Table of contents
:backlinks: none
:depth: 2
:local:
.. _schema_run_sql:
run_sql
-------
``run_sql`` can be used to run arbitrary SQL statements.
Multiple SQL statements can be separated by a ``;``, however, only the result of the last SQL statement will be
returned.
.. admonition:: Admin-only
This is an admin-only request, i.e. the request can only be executed with ``X-Hasura-Role: admin``. This can be set by passing
``X-Hasura-Admin-Secret`` or by setting the right role in webhook/JWT
authorization mode.
This is deliberate as it is hard to enforce any sort of permissions on arbitrary SQL. If
you find yourself in the need of using ``run_sql`` to run custom DML requests,
consider creating a view. You can now define permissions on that particular view
for various roles.
.. admonition:: Supported from
The schema API is supported for versions ``v2.0.0`` and above and replaces the older
:ref:`schema/metadata API <schema_metadata_apis>`.
Use cases
^^^^^^^^^
1. To execute DDL operations that are not supported by the console (e.g. managing indexes).
2. Run custom DML requests from backend microservices instead of installing libraries to speak to Postgres.
An example:
.. code-block:: http
POST /v2/query HTTP/1.1
Content-Type: application/json
X-Hasura-Role: admin
{
"type": "run_sql",
"args": {
"source": "default",
"sql": "CREATE UNIQUE INDEX ON films (title);"
}
}
While ``run_sql`` lets you run any SQL, it tries to ensure that the Hasura GraphQL engine's
state (relationships, permissions etc.) is consistent i.e. you
cannot drop a column on which any metadata is dependent on (say a permission or
a relationship). The effects, however, can be cascaded.
Example: If we were to drop the 'bio' column from the author table (let's say
the column is used in some permission), you would see an error.
.. code-block:: http
POST /v2/query HTTP/1.1
Content-Type: application/json
X-Hasura-Role: admin
{
"type": "run_sql",
"args": {
"source": "default",
"sql": "ALTER TABLE author DROP COLUMN name"
}
}
.. code-block:: http
HTTP/1.1 400 BAD REQUEST
Content-Type: application/json
{
"path": "$.args",
"error": "cannot drop due to the following dependent objects : permission author.user.select"
}
We can however, cascade these changes.
.. code-block:: http
:emphasize-lines: 9
POST /v2/query HTTP/1.1
Content-Type: application/json
X-Hasura-Role: admin
{
"type": "run_sql",
"args": {
"source": "default",
"sql": "ALTER TABLE author DROP COLUMN bio",
"cascade" : true
}
}
.. code-block:: http
HTTP/1.1 200 OK
Content-Type: application/json
{
"result_type": "CommandOk"
}
With the above request, the dependent permission is also dropped.
Example: If we were to drop a foreign key constraint from the article table
(let's say the column involved in the foreign key is used to define a relationship),
you would see an error.
.. code-block:: http
POST /v2/query HTTP/1.1
Content-Type: application/json
X-Hasura-Role: admin
{
"type": "run_sql",
"args": {
"source": "default",
"sql": "ALTER TABLE article DROP CONSTRAINT article_author_id_fkey"
}
}
.. code-block:: http
HTTP/1.1 400 BAD REQUEST
Content-Type: application/json
{
"path": "$.args",
"error": "cannot drop due to the following dependent objects : constraint article.article_author_id_fkey"
}
We can however, cascade these changes.
.. code-block:: http
:emphasize-lines: 9
POST /v2/query HTTP/1.1
Content-Type: application/json
X-Hasura-Role: admin
{
"type": "run_sql",
"args": {
"source": "default",
"sql": "ALTER TABLE article DROP CONSTRAINT article_author_id_fkey",
"cascade" : true
}
}
.. code-block:: http
HTTP/1.1 200 OK
Content-Type: application/json
{
"result_type": "CommandOk"
}
With the above request, the dependent permission is also dropped.
In general, the SQL operations that will affect Hasura metadata are:
1. Dropping columns
2. Dropping tables
3. Dropping foreign keys
4. Altering types of columns
5. Dropping SQL functions
6. Overloading SQL functions
In case of 1, 2 and 3 the dependent objects (if any) can be dropped using ``cascade``.
However, when altering type of columns, if any objects are affected, the change
cannot be cascaded. So, those dependent objects have to be manually dropped before
executing the SQL statement. Dropping SQL functions will cascade the functions in
metadata even without using ``cascade`` since no other objects depend on them.
Overloading tracked SQL functions is not allowed.
Set ``check_metadata_consistency`` field to ``false`` to force the server to not consider metadata dependencies.
.. _schema_run_sql_syntax:
Args syntax
^^^^^^^^^^^
.. list-table::
:header-rows: 1
* - Key
- Required
- Schema
- Description
* - sql
- true
- String
- The sql to be executed
* - source
- false
- String
- The database on which the sql is to be executed (default: 'default' database)
* - cascade
- false
- Boolean
- When set to ``true``, the effect (if possible) is cascaded to any hasuradb dependent objects (relationships, permissions, templates).
* - check_metadata_consistency
- false
- Boolean
- When set to ``false``, the sql is executed without checking metadata dependencies.
* - read_only
- false
- Boolean
- When set to ``true``, the request will be run in ``READ ONLY`` transaction access mode which means only ``select`` queries will be successful. This flag ensures that the GraphQL schema is not modified and is hence highly performant.
Response
^^^^^^^^
The response is a JSON Object with the following structure.
.. list-table::
:header-rows: 1
* - Key
- Always present
- Schema
- Description
* - result_type
- true
- String
- One of "CommandOk" or "TuplesOk"
* - result
- false
- ``[[Text]]`` (An array of rows, each row an array of columns)
- This is present only when the ``result_type`` is "TuplesOk"
.. note::
The first row in the ``result`` (when present) will be the names of the columns.
Some examples
^^^^^^^^^^^^^
An SQL query returning results.
.. code-block:: http
POST /v2/query HTTP/1.1
Content-Type: application/json
X-Hasura-Role: admin
{
"type": "run_sql",
"args": {
"source": "default",
"sql": "select user_id, first_name from author limit 2;"
}
}
.. code-block:: http
HTTP/1.1 200 OK
Content-Type: application/json
{
"result_type": "TuplesOk",
"result": [
[
"user_id",
"first_name"
],
[
"1",
"andre"
],
[
"2",
"angela"
]
]
}
An SQL query to create a table:
.. code-block:: http
POST /v2/query HTTP/1.1
Content-Type: application/json
X-Hasura-Role: admin
{
"type":"run_sql",
"args": {
"source": "default",
"sql": "create table item ( id serial, name text, category text, primary key (id))",
"check_metadata_consistency": false
}
}
.. code-block:: http
HTTP/1.1 200 OK
Content-Type: application/json
{
"result_type": "CommandOk",
"result": null
}