Internals for plugins#

Many Plugin hooks are passed objects that provide access to internal Datasette functionality. The interface to these objects should not be considered stable with the exception of methods that are documented here.

Request object#

The request object is passed to various plugin hooks. It represents an incoming HTTP request. It has the following properties:

.scope - dictionary

The ASGI scope that was used to construct this request, described in the ASGI HTTP connection scope specification.

.method - string

The HTTP method for this request, usually GET or POST.

.url - string

The full URL for this request, e.g.

.scheme - string

The request scheme - usually https or http.

.headers - dictionary (str -> str)

A dictionary of incoming HTTP request headers. Header names have been converted to lowercase.

.cookies - dictionary (str -> str)

A dictionary of incoming cookies

.host - string

The host header from the incoming request, e.g. or localhost.

.path - string

The path of the request excluding the query string, e.g. /fixtures.

.full_path - string

The path of the request including the query string if one is present, e.g. /fixtures?sql=select+sqlite_version().

.query_string - string

The query string component of the request, without the ? - e.g. name__contains=sam&age__gt=10.

.args - MultiParams

An object representing the parsed query string parameters, see below.

.url_vars - dictionary (str -> str)

Variables extracted from the URL path, if that path was defined using a regular expression. See register_routes(datasette).

.actor - dictionary (str -> Any) or None

The currently authenticated actor (see actors), or None if the request is unauthenticated.

The object also has two awaitable methods:

await request.post_vars() - dictionary

Returns a dictionary of form variables that were submitted in the request body via POST. Don't forget to read about CSRF protection!

await request.post_body() - bytes

Returns the un-parsed body of a request submitted by POST - useful for things like incoming JSON data.

And a class method that can be used to create fake request objects for use in tests:

fake(path_with_query_string, method="GET", scheme="http", url_vars=None)

Returns a Request instance for the specified path and method. For example:

from datasette import Request
from pprint import pprint

request = Request.fake(
    url_vars={"database": "fixtures", "table": "facetable"},

This outputs:

{'http_version': '1.1',
 'method': 'GET',
 'path': '/fixtures/facetable/',
 'query_string': b'',
 'raw_path': b'/fixtures/facetable/',
 'scheme': 'http',
 'type': 'http',
 'url_route': {'kwargs': {'database': 'fixtures', 'table': 'facetable'}}}

The MultiParams class#

request.args is a MultiParams object - a dictionary-like object which provides access to query string parameters that may have multiple values.

Consider the query string ?foo=1&foo=2&bar=3 - with two values for foo and one value for bar.

request.args[key] - string

Returns the first value for that key, or raises a KeyError if the key is missing. For the above example request.args["foo"] would return "1".

request.args.get(key) - string or None

Returns the first value for that key, or None if the key is missing. Pass a second argument to specify a different default, e.g. q = request.args.get("q", "").

request.args.getlist(key) - list of strings

Returns the list of strings for that key. request.args.getlist("foo") would return ["1", "2"] in the above example. request.args.getlist("bar") would return ["3"]. If the key is missing an empty list will be returned.

request.args.keys() - list of strings

Returns the list of available keys - for the example this would be ["foo", "bar"].

key in request.args - True or False

You can use if key in request.args to check if a key is present.

for key in request.args - iterator

This lets you loop through every available key.

len(request.args) - integer

Returns the number of keys.

Response class#

The Response class can be returned from view functions that have been registered using the register_routes(datasette) hook.

The Response() constructor takes the following arguments:

body - string

The body of the response.

status - integer (optional)

The HTTP status - defaults to 200.

headers - dictionary (optional)

A dictionary of extra HTTP headers, e.g. {"x-hello": "world"}.

content_type - string (optional)

The content-type for the response. Defaults to text/plain.

For example:

from datasette.utils.asgi import Response

response = Response(
    "<xml>This is XML</xml>",
    content_type="application/xml; charset=utf-8",

The quickest way to create responses is using the Response.text(...), Response.html(...), Response.json(...) or Response.redirect(...) helper methods:

from datasette.utils.asgi import Response

html_response = Response.html("This is HTML")
json_response = Response.json({"this_is": "json"})
text_response = Response.text(
    "This will become utf-8 encoded text"
# Redirects are served as 302, unless you pass status=301:
redirect_response = Response.redirect(

Each of these responses will use the correct corresponding content-type - text/html; charset=utf-8, application/json; charset=utf-8 or text/plain; charset=utf-8 respectively.

Each of the helper methods take optional status= and headers= arguments, documented above.

Returning a response with .asgi_send(send)#

In most cases you will return Response objects from your own view functions. You can also use a Response instance to respond at a lower level via ASGI, for example if you are writing code that uses the asgi_wrapper(datasette) hook.

Create a Response object and then use await response.asgi_send(send), passing the ASGI send function. For example:

async def require_authorization(scope, receive, send):
    response = Response.text(
        "401 Authorization Required",
            "www-authenticate": 'Basic realm="Datasette", charset="UTF-8"'
    await response.asgi_send(send)

Datasette class#

This object is an instance of the Datasette class, passed to many plugin hooks as an argument called datasette.

You can create your own instance of this - for example to help write tests for a plugin - like so:

from import Datasette

# With no arguments a single in-memory database will be attached
datasette = Datasette()

# The files= argument can load files from disk
datasette = Datasette(files=["/path/to/my-database.db"])

# Pass metadata as a JSON dictionary like this
datasette = Datasette(
        "databases": {
            "my-database": {
                "description": "This is my database"

Constructor parameters include:

  • files=[...] - a list of database files to open

  • immutables=[...] - a list of database files to open in immutable mode

  • metadata={...} - a dictionary of Metadata

  • config_dir=... - the configuration directory to use, stored in datasette.config_dir


Property exposing a collections.OrderedDict of databases currently connected to Datasette.

The dictionary keys are the name of the database that is used in the URL - e.g. /fixtures would have a key of "fixtures". The values are Database class instances.

All databases are listed, irrespective of user permissions.


Property exposing a dictionary of permissions that have been registered using the register_permissions(datasette) plugin hook.

The dictionary keys are the permission names - e.g. view-instance - and the values are Permission() objects describing the permission. Here is a description of that object.

.plugin_config(plugin_name, database=None, table=None)#

plugin_name - string

The name of the plugin to look up configuration for. Usually this is something similar to datasette-cluster-map.

database - None or string

The database the user is interacting with.

table - None or string

The table the user is interacting with.

This method lets you read plugin configuration values that were set in datasette.yaml. See Writing plugins that accept configuration for full details of how this method should be used.

The return value will be the value from the configuration file - usually a dictionary.

If the plugin is not configured the return value will be None.

await .render_template(template, context=None, request=None)#

template - string, list of strings or jinja2.Template

The template file to be rendered, e.g. my_plugin.html. Datasette will search for this file first in the --template-dir= location, if it was specified - then in the plugin's bundled templates and finally in Datasette's set of default templates.

If this is a list of template file names then the first one that exists will be loaded and rendered.

If this is a Jinja Template object it will be used directly.

context - None or a Python dictionary

The context variables to pass to the template.

request - request object or None

If you pass a Datasette request object here it will be made available to the template.

Renders a Jinja template using Datasette's preconfigured instance of Jinja and returns the resulting string. The template will have access to Datasette's default template functions and any functions that have been made available by other plugins.

await .actors_from_ids(actor_ids)#

actor_ids - list of strings or integers

A list of actor IDs to look up.

Returns a dictionary, where the keys are the IDs passed to it and the values are the corresponding actor dictionaries.

This method is mainly designed to be used with plugins. See the actors_from_ids(datasette, actor_ids) documentation for details.

If no plugins that implement that hook are installed, the default return value looks like this:

    "1": {"id": "1"},
    "2": {"id": "2"}

await .permission_allowed(actor, action, resource=None, default=...)#

actor - dictionary

The authenticated actor. This is usually

action - string

The name of the action that is being permission checked.

resource - string or tuple, optional

The resource, e.g. the name of the database, or a tuple of two strings containing the name of the database and the name of the table. Only some permissions apply to a resource.

default - optional: True, False or None

What value should be returned by default if nothing provides an opinion on this permission check. Set to True for default allow or False for default deny. If not specified the default from the Permission() tuple that was registered using register_permissions(datasette) will be used.

Check if the given actor has permission to perform the given action on the given resource.

Some permission checks are carried out against rules defined in datasette.yaml, while other custom permissions may be decided by plugins that implement the permission_allowed(datasette, actor, action, resource) plugin hook.

If neither metadata.json nor any of the plugins provide an answer to the permission query the default argument will be returned.

See Built-in permissions for a full list of permission actions included in Datasette core.

await .ensure_permissions(actor, permissions)#

actor - dictionary

The authenticated actor. This is usually

permissions - list

A list of permissions to check. Each permission in that list can be a string action name or a 2-tuple of (action, resource).

This method allows multiple permissions to be checked at once. It raises a datasette.Forbidden exception if any of the checks are denied before one of them is explicitly granted.

This is useful when you need to check multiple permissions at once. For example, an actor should be able to view a table if either one of the following checks returns True or not a single one of them returns False:

await self.ds.ensure_permissions(,
        ("view-table", (database, table)),
        ("view-database", database),

await .check_visibility(actor, action=None, resource=None, permissions=None)#

actor - dictionary

The authenticated actor. This is usually

action - string, optional

The name of the action that is being permission checked.

resource - string or tuple, optional

The resource, e.g. the name of the database, or a tuple of two strings containing the name of the database and the name of the table. Only some permissions apply to a resource.

permissions - list of action strings or (action, resource) tuples, optional

Provide this instead of action and resource to check multiple permissions at once.

This convenience method can be used to answer the question "should this item be considered private, in that it is visible to me but it is not visible to anonymous users?"

It returns a tuple of two booleans, (visible, private). visible indicates if the actor can see this resource. private will be True if an anonymous user would not be able to view the resource.

This example checks if the user can access a specific table, and sets private so that a padlock icon can later be displayed:

visible, private = await self.ds.check_visibility(,
    resource=(database, table),

The following example runs three checks in a row, similar to await .ensure_permissions(actor, permissions). If any of the checks are denied before one of them is explicitly granted then visible will be False. private will be True if an anonymous user would not be able to view the resource.

visible, private = await self.ds.check_visibility(,
        ("view-table", (database, table)),
        ("view-database", database),

.create_token(actor_id, expires_after=None, restrict_all=None, restrict_database=None, restrict_resource=None)#

actor_id - string

The ID of the actor to create a token for.

expires_after - int, optional

The number of seconds after which the token should expire.

restrict_all - iterable, optional

A list of actions that this token should be restricted to across all databases and resources.

restrict_database - dict, optional

For restricting actions within specific databases, e.g. {"mydb": ["view-table", "view-query"]}.

restrict_resource - dict, optional

For restricting actions to specific resources (tables, SQL views and Canned queries) within a database. For example: {"mydb": {"mytable": ["insert-row", "update-row"]}}.

This method returns a signed API token of the format dstok_... which can be used to authenticate requests to the Datasette API.

All tokens must have an actor_id string indicating the ID of the actor which the token will act on behalf of.

Tokens default to lasting forever, but can be set to expire after a given number of seconds using the expires_after argument. The following code creates a token for user1 that will expire after an hour:

token = datasette.create_token(

The three restrict_* arguments can be used to create a token that has additional restrictions beyond what the associated actor is allowed to do.

The following example creates a token that can access view-instance and view-table across everything, can additionally use view-query for anything in the docs database and is allowed to execute insert-row and update-row in the attachments table in that database:

token = datasette.create_token(
    restrict_all=("view-instance", "view-table"),
    restrict_database={"docs": ("view-query",)},
        "docs": {
            "attachments": ("insert-row", "update-row")


name_or_abbr - string

The name or abbreviation of the permission to look up, e.g. view-table or vt.

Returns a Permission object representing the permission, or raises a KeyError if one is not found.


name - string, optional

The name of the database - optional.

Returns the specified database object. Raises a KeyError if the database does not exist. Call this method without an argument to return the first connected database.


Returns a database object for reading and writing to the private internal database.

.add_database(db, name=None, route=None)#

db - datasette.database.Database instance

The database to be attached.

name - string, optional

The name to be used for this database . If not specified Datasette will pick one based on the filename or memory name.

route - string, optional

This will be used in the URL path. If not specified, it will default to the same thing as the name.

The datasette.add_database(db) method lets you add a new database to the current Datasette instance.

The db parameter should be an instance of the datasette.database.Database class. For example:

from datasette.database import Database


This will add a mutable database and serve it at /my-new-database.

Use is_mutable=False to add an immutable database.

.add_database() returns the Database instance, with its name set as the attribute. Any time you are working with a newly added database you should use the return value of .add_database(), for example:

db = datasette.add_database(
    Database(datasette, memory_name="statistics")
await db.execute_write(
    "CREATE TABLE foo(id integer primary key)"


Adds a shared in-memory database with the specified name:


This is a shortcut for the following:

from datasette.database import Database

    Database(datasette, memory_name="statistics")

Using either of these pattern will result in the in-memory database being served at /statistics.


name - string

The name of the database to be removed.

This removes a database that has been previously added. name= is the unique name of that database.

await .track_event(event)#

event - Event

An instance of a subclass of

Plugins can call this to track events, using classes they have previously registered. See Event tracking for details.

The event will then be passed to all plugins that have registered to receive events using the track_event(datasette, event) hook.

Example usage, assuming the plugin has previously registered the BanUserEvent class:

await datasette.track_event(
    BanUserEvent(user={"id": 1, "username": "cleverbot"})

.sign(value, namespace="default")#

value - any serializable type

The value to be signed.

namespace - string, optional

An alternative namespace, see the itsdangerous salt documentation.

Utility method for signing values, such that you can safely pass data to and from an untrusted environment. This is a wrapper around the itsdangerous library.

This method returns a signed string, which can be decoded and verified using .unsign(value, namespace="default").

.unsign(value, namespace="default")#

signed - any serializable type

The signed string that was created using .sign(value, namespace="default").

namespace - string, optional

The alternative namespace, if one was used.

Returns the original, decoded object that was passed to .sign(value, namespace="default"). If the signature is not valid this raises a itsdangerous.BadSignature exception.

.add_message(request, message, type=datasette.INFO)#

request - Request

The current Request object

message - string

The message string

type - constant, optional

The message type - datasette.INFO, datasette.WARNING or datasette.ERROR

Datasette's flash messaging mechanism allows you to add a message that will be displayed to the user on the next page that they visit. Messages are persisted in a ds_messages cookie. This method adds a message to that cookie.

You can try out these messages (including the different visual styling of the three message types) using the /-/messages debugging tool.

.absolute_url(request, path)#

request - Request

The current Request object

path - string

A path, for example /dbname/table.json

Returns the absolute URL for the given path, including the protocol and host. For example:

absolute_url = datasette.absolute_url(
    request, "/dbname/table.json"
# Would return "http://localhost:8001/dbname/table.json"

The current request object is used to determine the hostname and protocol that should be used for the returned URL. The force_https_urls configuration setting is taken into account.


key - string

The name of the setting, e.g. base_url.

Returns the configured value for the specified setting. This can be a string, boolean or integer depending on the requested setting.

For example:

downloads_are_allowed = datasette.setting("allow_download")


request - Request object

A request object

If you are implementing your own custom views, you may need to resolve the database that the user is requesting based on a URL path. If the regular expression for your route declares a database named group, you can use this method to resolve the database object.

This returns a Database instance.

If the database cannot be found, it raises a datasette.utils.asgi.DatabaseNotFound exception - which is a subclass of datasette.utils.asgi.NotFound with a .database_name attribute set to the name of the database that was requested.


request - Request object

A request object

This assumes that the regular expression for your route declares both a database and a table named group.

It returns a ResolvedTable named tuple instance with the following fields:

db - Database

The database object

table - string

The name of the table (or view)

is_view - boolean

True if this is a view, False if it is a table

If the database or table cannot be found it raises a datasette.utils.asgi.DatabaseNotFound exception.

If the table does not exist it raises a datasette.utils.asgi.TableNotFound exception - a subclass of datasette.utils.asgi.NotFound with .database_name and .table attributes.


request - Request object

A request object

This method assumes your route declares named groups for database, table and pks.

It returns a ResolvedRow named tuple instance with the following fields:

db - Database

The database object

table - string

The name of the table

sql - string

SQL snippet that can be used in a WHERE clause to select the row

params - dict

Parameters that should be passed to the SQL query

pks - list

List of primary key column names

pk_values - list

List of primary key values decoded from the URL

row - sqlite3.Row

The row itself

If the database or table cannot be found it raises a datasette.utils.asgi.DatabaseNotFound exception.

If the table does not exist it raises a datasette.utils.asgi.TableNotFound exception.

If the row cannot be found it raises a datasette.utils.asgi.RowNotFound exception. This has .database_name, .table and .pk_values attributes, extracted from the request path.


Plugins can make internal simulated HTTP requests to the Datasette instance within which they are running. This ensures that all of Datasette's external JSON APIs are also available to plugins, while avoiding the overhead of making an external HTTP call to access those APIs.

The datasette.client object is a wrapper around the HTTPX Python library, providing an async-friendly API that is similar to the widely used Requests library.

It offers the following methods:

await datasette.client.get(path, **kwargs) - returns HTTPX Response

Execute an internal GET request against that path.

await, **kwargs) - returns HTTPX Response

Execute an internal POST request. Use data={"name": "value"} to pass form parameters.

await datasette.client.options(path, **kwargs) - returns HTTPX Response

Execute an internal OPTIONS request.

await datasette.client.head(path, **kwargs) - returns HTTPX Response

Execute an internal HEAD request.

await datasette.client.put(path, **kwargs) - returns HTTPX Response

Execute an internal PUT request.

await datasette.client.patch(path, **kwargs) - returns HTTPX Response

Execute an internal PATCH request.

await datasette.client.delete(path, **kwargs) - returns HTTPX Response

Execute an internal DELETE request.

await datasette.client.request(method, path, **kwargs) - returns HTTPX Response

Execute an internal request with the given HTTP method against that path.

These methods can be used with datasette.urls - for example:

table_json = (
    await datasette.client.get(
            "fixtures", "facetable", format="json"

datasette.client methods automatically take the current base_url setting into account, whether or not you use the datasette.urls family of methods to construct the path.

For documentation on available **kwargs options and the shape of the HTTPX Response object refer to the HTTPX Async documentation.


The datasette.urls object contains methods for building URLs to pages within Datasette. Plugins should use this to link to pages, since these methods take into account any base_url configuration setting that might be in effect.


Returns the URL to the Datasette instance root page. This is usually "/".

datasette.urls.path(path, format=None)

Takes a path and returns the full path, taking base_url into account.

For example, datasette.urls.path("-/logout") will return the path to the logout page, which will be "/-/logout" by default or /prefix-path/-/logout if base_url is set to /prefix-path/


Returns the URL to the logout page, usually "/-/logout"


Returns the URL of one of Datasette's default static assets, for example "/-/static/app.css"

datasette.urls.static_plugins(plugin_name, path)

Returns the URL of one of the static assets belonging to a plugin.

datasette.urls.static_plugins("datasette_cluster_map", "datasette-cluster-map.js") would return "/-/static-plugins/datasette_cluster_map/datasette-cluster-map.js"


Returns the URL of one of Datasette's default static assets, for example "/-/static/app.css"

datasette.urls.database(database_name, format=None)

Returns the URL to a database page, for example "/fixtures"

datasette.urls.table(database_name, table_name, format=None)

Returns the URL to a table page, for example "/fixtures/facetable"

datasette.urls.query(database_name, query_name, format=None)

Returns the URL to a query page, for example "/fixtures/pragma_cache_size"

These functions can be accessed via the {{ urls }} object in Datasette templates, for example:

<a href="{{ urls.instance() }}">Homepage</a>
<a href="{{ urls.database("fixtures") }}">Fixtures database</a>
<a href="{{ urls.table("fixtures", "facetable") }}">facetable table</a>
<a href="{{ urls.query("fixtures", "pragma_cache_size") }}">pragma_cache_size query</a>

Use the format="json" (or "csv" or other formats supported by plugins) arguments to get back URLs to the JSON representation. This is the path with .json added on the end.

These methods each return a datasette.utils.PrefixedUrlString object, which is a subclass of the Python str type. This allows the logic that considers the base_url setting to detect if that prefix has already been applied to the path.

Database class#

Instances of the Database class can be used to execute queries against attached SQLite databases, and to run introspection against their schemas.

Database(ds, path=None, is_mutable=True, is_memory=False, memory_name=None)#

The Database() constructor can be used by plugins, in conjunction with .add_database(db, name=None, route=None), to create and register new databases.

The arguments are as follows:

ds - Datasette class (required)

The Datasette instance you are attaching this database to.

path - string

Path to a SQLite database file on disk.

is_mutable - boolean

Set this to False to cause Datasette to open the file in immutable mode.

is_memory - boolean

Use this to create non-shared memory connections.

memory_name - string or None

Use this to create a named in-memory database. Unlike regular memory databases these can be accessed by multiple threads and will persist an changes made to them for the lifetime of the Datasette server process.

The first argument is the datasette instance you are attaching to, the second is a path=, then is_mutable and is_memory are both optional arguments.


If the database was opened in immutable mode, this property returns the 64 character SHA-256 hash of the database contents as a string. Otherwise it returns None.

await db.execute(sql, ...)#

Executes a SQL query against the database and returns the resulting rows (see Results).

sql - string (required)

The SQL query to execute. This can include ? or :named parameters.

params - list or dict

A list or dictionary of values to use for the parameters. List for ?, dictionary for :named.

truncate - boolean

Should the rows returned by the query be truncated at the maximum page size? Defaults to True, set this to False to disable truncation.

custom_time_limit - integer ms

A custom time limit for this query. This can be set to a lower value than the Datasette configured default. If a query takes longer than this it will be terminated early and raise a dataette.database.QueryInterrupted exception.

page_size - integer

Set a custom page size for truncation, over-riding the configured Datasette default.

log_sql_errors - boolean

Should any SQL errors be logged to the console in addition to being raised as an error? Defaults to True.


The db.execute() method returns a single Results object. This can be used to access the rows returned by the query.

Iterating over a Results object will yield SQLite Row objects. Each of these can be treated as a tuple or can be accessed using row["column"] syntax:

info = []
results = await db.execute("select name from sqlite_master")
for row in results:

The Results object also has the following properties and methods:

.truncated - boolean

Indicates if this query was truncated - if it returned more results than the specified page_size. If this is true then the results object will only provide access to the first page_size rows in the query result. You can disable truncation by passing truncate=False to the db.query() method.

.columns - list of strings

A list of column names returned by the query.

.rows - list of sqlite3.Row

This property provides direct access to the list of rows returned by the database. You can access specific rows by index using results.rows[0].

.first() - row or None

Returns the first row in the results, or None if no rows were returned.


Returns the value of the first column of the first row of results - but only if the query returned a single row with a single column. Raises a datasette.database.MultipleValues exception otherwise.


Calling len(results) returns the (truncated) number of returned results.

await db.execute_fn(fn)#

Executes a given callback function against a read-only database connection running in a thread. The function will be passed a SQLite connection, and the return value from the function will be returned by the await.

Example usage:

def get_version(conn):
    return conn.execute(
        "select sqlite_version()"

version = await db.execute_fn(get_version)

await db.execute_write(sql, params=None, block=True)#

SQLite only allows one database connection to write at a time. Datasette handles this for you by maintaining a queue of writes to be executed against a given database. Plugins can submit write operations to this queue and they will be executed in the order in which they are received.

This method can be used to queue up a non-SELECT SQL query to be executed against a single write connection to the database.

You can pass additional SQL parameters as a tuple or dictionary.

The method will block until the operation is completed, and the return value will be the return from calling conn.execute(...) using the underlying sqlite3 Python library.

If you pass block=False this behavior changes to "fire and forget" - queries will be added to the write queue and executed in a separate thread while your code can continue to do other things. The method will return a UUID representing the queued task.

Each call to execute_write() will be executed inside a transaction.

await db.execute_write_script(sql, block=True)#

Like execute_write() but can be used to send multiple SQL statements in a single string separated by semicolons, using the sqlite3 conn.executescript() method.

Each call to execute_write_script() will be executed inside a transaction.

await db.execute_write_many(sql, params_seq, block=True)#

Like execute_write() but uses the sqlite3 conn.executemany() method. This will efficiently execute the same SQL statement against each of the parameters in the params_seq iterator, for example:

await db.execute_write_many(
    "insert into characters (id, name) values (?, ?)",
    [(1, "Melanie"), (2, "Selma"), (2, "Viktor")],

Each call to execute_write_many() will be executed inside a transaction.

await db.execute_write_fn(fn, block=True, transaction=True)#

This method works like .execute_write(), but instead of a SQL statement you give it a callable Python function. Your function will be queued up and then called when the write connection is available, passing that connection as the argument to the function.

The function can then perform multiple actions, safe in the knowledge that it has exclusive access to the single writable connection for as long as it is executing.


fn needs to be a regular function, not an async def function.

For example:

def delete_and_return_count(conn):
    conn.execute("delete from some_table where id > 5")
    return conn.execute(
        "select count(*) from some_table"

    num_rows_left = await database.execute_write_fn(
except Exception as e:
    print("An error occurred:", e)

The value returned from await database.execute_write_fn(...) will be the return value from your function.

If your function raises an exception that exception will be propagated up to the await line.

By default your function will be executed inside a transaction. You can pass transaction=False to disable this behavior, though if you do that you should be careful to manually apply transactions - ideally using the with conn: pattern, or you may see OperationalError: database table is locked errors.

If you specify block=False the method becomes fire-and-forget, queueing your function to be executed and then allowing your code after the call to .execute_write_fn() to continue running while the underlying thread waits for an opportunity to run your function. A UUID representing the queued task will be returned. Any exceptions in your code will be silently swallowed.

await db.execute_isolated_fn(fn)#

This method works is similar to execute_write_fn() but executes the provided function in an entirely isolated SQLite connection, which is opened, used and then closed again in a single call to this method.

The prepare_connection() plugin hook is not executed against this connection.

This allows plugins to execute database operations that might conflict with how database connections are usually configured. For example, running a VACUUM operation while bypassing any restrictions placed by the datasette-sqlite-authorizer plugin.

Plugins can also use this method to load potentially dangerous SQLite extensions, use them to perform an operation and then have them safely unloaded at the end of the call, without risk of exposing them to other connections.

Functions run using execute_isolated_fn() share the same queue as execute_write_fn(), which guarantees that no writes can be executed at the same time as the isolated function is executing.

The return value of the function will be returned by this method. Any exceptions raised by the function will be raised out of the await line as well.


Closes all of the open connections to file-backed databases. This is mainly intended to be used by large test suites, to avoid hitting limits on the number of open files.

Database introspection#

The Database class also provides properties and methods for introspecting the database. - string

The name of the database - usually the filename without the .db prefix.

db.size - integer

The size of the database file in bytes. 0 for :memory: databases.

db.mtime_ns - integer or None

The last modification time of the database file in nanoseconds since the epoch. None for :memory: databases.

db.is_mutable - boolean

Is this database mutable, and allowed to accept writes?

db.is_memory - boolean

Is this database an in-memory database?

await db.attached_databases() - list of named tuples

Returns a list of additional databases that have been connected to this database using the SQLite ATTACH command. Each named tuple has fields seq, name and file.

await db.table_exists(table) - boolean

Check if a table called table exists.

await db.view_exists(view) - boolean

Check if a view called view exists.

await db.table_names() - list of strings

List of names of tables in the database.

await db.view_names() - list of strings

List of names of views in the database.

await db.table_columns(table) - list of strings

Names of columns in a specific table.

await db.table_column_details(table) - list of named tuples

Full details of the columns in a specific table. Each column is represented by a Column named tuple with fields cid (integer representing the column position), name (string), type (string, e.g. REAL or VARCHAR(30)), notnull (integer 1 or 0), default_value (string or None), is_pk (integer 1 or 0).

await db.primary_keys(table) - list of strings

Names of the columns that are part of the primary key for this table.

await db.fts_table(table) - string or None

The name of the FTS table associated with this table, if one exists.

await db.label_column_for_table(table) - string or None

The label column that is associated with this table - either automatically detected or using the "label_column" key from Metadata, see Specifying the label column for a table.

await db.foreign_keys_for_table(table) - list of dictionaries

Details of columns in this table which are foreign keys to other tables. A list of dictionaries where each dictionary is shaped like this: {"column": string, "other_table": string, "other_column": string}.

await db.hidden_table_names() - list of strings

List of tables which Datasette "hides" by default - usually these are tables associated with SQLite's full-text search feature, the SpatiaLite extension or tables hidden using the Hiding tables feature.

await db.get_table_definition(table) - string

Returns the SQL definition for the table - the CREATE TABLE statement and any associated CREATE INDEX statements.

await db.get_view_definition(view) - string

Returns the SQL definition of the named view.

await db.get_all_foreign_keys() - dictionary

Dictionary representing both incoming and outgoing foreign keys for this table. It has two keys, "incoming" and "outgoing", each of which is a list of dictionaries with keys "column", "other_table" and "other_column". For example:

    "incoming": [],
    "outgoing": [
            "other_table": "attraction_characteristic",
            "column": "characteristic_id",
            "other_column": "pk",
            "other_table": "roadside_attractions",
            "column": "attraction_id",
            "other_column": "pk",

CSRF protection#

Datasette uses asgi-csrf to guard against CSRF attacks on form POST submissions. Users receive a ds_csrftoken cookie which is compared against the csrftoken form field (or x-csrftoken HTTP header) for every incoming request.

If your plugin implements a <form method="POST"> anywhere you will need to include that token. You can do so with the following template snippet:

<input type="hidden" name="csrftoken" value="{{ csrftoken() }}">

If you are rendering templates using the await .render_template(template, context=None, request=None) method the csrftoken() helper will only work if you provide the request= argument to that method. If you forget to do this you will see the following error:

form-urlencoded POST field did not match cookie

You can selectively disable CSRF protection using the skip_csrf(datasette, scope) hook.

Datasette's internal database#

Datasette maintains an "internal" SQLite database used for configuration, caching, and storage. Plugins can store configuration, settings, and other data inside this database. By default, Datasette will use a temporary in-memory SQLite database as the internal database, which is created at startup and destroyed at shutdown. Users of Datasette can optionally pass in a --internal flag to specify the path to a SQLite database to use as the internal database, which will persist internal data across Datasette instances.

Datasette maintains tables called catalog_databases, catalog_tables, catalog_columns, catalog_indexes, catalog_foreign_keys with details of the attached databases and their schemas. These tables should not be considered a stable API - they may change between Datasette releases.

The internal database is not exposed in the Datasette application by default, which means private data can safely be stored without worry of accidentally leaking information through the default Datasette interface and API. However, other plugins do have full read and write access to the internal database.

Plugins can access this database by calling internal_db = datasette.get_internal_database() and then executing queries using the Database API.

Plugin authors are asked to practice good etiquette when using the internal database, as all plugins use the same database to store data. For example:

  1. Use a unique prefix when creating tables, indices, and triggera in the internal database. If your plugin is called datasette-xyz, then prefix names with datasette_xyz_*.

  2. Avoid long-running write statements that may stall or block other plugins that are trying to write at the same time.

  3. Use temporary tables or shared in-memory attached databases when possible.

  4. Avoid implementing features that could expose private data stored in the internal database by other plugins.

The datasette.utils module#

The datasette.utils module contains various utility functions used by Datasette. As a general rule you should consider anything in this module to be unstable - functions and classes here could change without warning or be removed entirely between Datasette releases, without being mentioned in the release notes.

The exception to this rule is anythang that is documented here. If you find a need for an undocumented utility function in your own work, consider opening an issue requesting that the function you are using be upgraded to documented and supported status.


This function accepts a string containing either JSON or YAML, expected to be of the format described in Metadata. It returns a nested Python dictionary representing the parsed data from that string.

If the metadata cannot be parsed as either JSON or YAML the function will raise a utils.BadMetadataError exception.

datasette.utils.parse_metadata(content: str) dict#

Detects if content is JSON or YAML and parses it appropriately.


Utility function for calling await on a return value if it is awaitable, otherwise returning the value. This is used by Datasette to support plugin hooks that can optionally return awaitable functions. Read more about this function in The “await me maybe” pattern for Python asyncio.

async datasette.utils.await_me_maybe(value: Any) Any#

If value is callable, call it. If awaitable, await it. Otherwise return it.

derive_named_parameters(db, sql)#

Derive the list of named parameters referenced in a SQL query, using an explain query executed against the provided database.

async datasette.utils.derive_named_parameters(db: Database, sql: str) List[str]#

Given a SQL statement, return a list of named parameters that are used in the statement

e.g. for select * from foo where id=:id this would return ["id"]

Tilde encoding#

Datasette uses a custom encoding scheme in some places, called tilde encoding. This is primarily used for table names and row primary keys, to avoid any confusion between / characters in those values and the Datasette URLs that reference them.

Tilde encoding uses the same algorithm as URL percent-encoding, but with the ~ tilde character used in place of %.

Any character other than ABCDEFGHIJKLMNOPQRSTUVWXYZ abcdefghijklmnopqrstuvwxyz0123456789_- will be replaced by the numeric equivalent preceded by a tilde. For example:

  • / becomes ~2F

  • . becomes ~2E

  • % becomes ~25

  • ~ becomes ~7E

  • Space becomes +

  • polls/2022.primary becomes polls~2F2022~2Eprimary

Note that the space character is a special case: it will be replaced with a + symbol.

datasette.utils.tilde_encode(s: str) str#

Returns tilde-encoded string - for example /foo/bar -> ~2Ffoo~2Fbar

datasette.utils.tilde_decode(s: str) str#

Decodes a tilde-encoded string, so ~2Ffoo~2Fbar -> /foo/bar


Running Datasette with --setting trace_debug 1 enables trace debug output, which can then be viewed by adding ?_trace=1 to the query string for any page.

You can see an example of this at the bottom of The JSON output shows full details of every SQL query that was executed to generate the page.

The datasette-pretty-traces plugin can be installed to provide a more readable display of this information. You can see a demo of that here.

You can add your own custom traces to the JSON output using the trace() context manager. This takes a string that identifies the type of trace being recorded, and records any keyword arguments as additional JSON keys on the resulting trace object.

The start and end time, duration and a traceback of where the trace was executed will be automatically attached to the JSON object.

This example uses trace to record the start, end and duration of any HTTP GET requests made using the function:

from datasette.tracer import trace
import httpx

async def fetch_url(url):
    with trace("fetch-url", url=url):
        async with httpx.AsyncClient() as client:
            return await client.get(url)

Tracing child tasks#

If your code uses a mechanism such as asyncio.gather() to execute code in additional tasks you may find that some of the traces are missing from the display.

You can use the trace_child_tasks() context manager to ensure these child tasks are correctly handled.

from datasette import tracer

with tracer.trace_child_tasks():
    results = await asyncio.gather(
        # ... async tasks here

This example uses the register_routes() plugin hook to add a page at /parallel-queries which executes two SQL queries in parallel using asyncio.gather() and returns their results.

from datasette import hookimpl
from datasette import tracer

def register_routes():
    async def parallel_queries(datasette):
        db = datasette.get_database()
        with tracer.trace_child_tasks():
            one, two = await asyncio.gather(
                db.execute("select 1"),
                db.execute("select 2"),
        return Response.json(
                "one": one.single_value(),
                "two": two.single_value(),

    return [
        (r"/parallel-queries$", parallel_queries),

Note that running parallel SQL queries in this way has been known to cause problems in the past, so treat this example with caution.

Adding ?_trace=1 will show that the trace covers both of those child tasks.

Import shortcuts#

The following commonly used symbols can be imported directly from the datasette module:

from datasette import Response
from datasette import Forbidden
from datasette import NotFound
from datasette import hookimpl
from datasette import actor_matches_allow