LTI 1.3 Advantage Tool implementation in Python

This project is a Python implementation of the similar PHP tool. Library contains adapters for usage from Django and Flask web frameworks but there is no difficulty to adapt it to other frameworks: you should just re-implement OIDCLogin and MessageLaunch classes as it is already done in existing adapters.

Examples of usage

Django: https://github.com/dmitry-viskov/pylti1.3-django-example

Flask: https://github.com/dmitry-viskov/pylti1.3-flask-example

Configuration

To configure your own tool you may use built-in adapters:

from pylti1p3.tool_config import ToolConfJsonFile
tool_conf = ToolConfJsonFile('path/to/json')

from pylti1p3.tool_config import ToolConfDict
settings = {
    "<issuer_1>": { },  # one issuer ~ one client-id (outdated and not recommended way)
    "<issuer_2>": [{ }, { }]  # one issuer ~ many client-ids (preferable way)
}
private_key = '...'
public_key = '...'
tool_conf = ToolConfDict(settings)

client_id = '...' # must be set in case of "one issuer ~ many client-ids" concept

tool_conf.set_private_key(iss, private_key, client_id=client_id)
tool_conf.set_public_key(iss, public_key, client_id=client_id)

or create your own implementation. The pylti1p3.tool_config.ToolConfAbstract interface must be fully implemented for this to work. Concept of one issuer ~ many client-ids is a preferable way to organize configs and may be useful in case of integration with Canvas (https://canvas.instructure.com) or other Cloud LMS-es where platform doesn't change iss for each customer.

In case of Django Framework you may use DjangoDbToolConf (see Configuration using Django Admin UI section below)

Example of JSON config:

{
    "iss1": [{
        "default": true,
        "client_id": "client_id1",
        "auth_login_url": "auth_login_url1",
        "auth_token_url": "auth_token_url1",
        "auth_audience": null,
        "key_set_url": "key_set_url1",
        "key_set": null,
        "private_key_file": "private.key",
        "public_key_file": "public.key",
        "deployment_ids": ["deployment_id1", "deployment_id2"]
    }, {
        "default": false,
        "client_id": "client_id2",
        "auth_login_url": "auth_login_url2",
        "auth_token_url": "auth_token_url2",
        "auth_audience": null,
        "key_set_url": "key_set_url2",
        "key_set": null,
        "private_key_file": "private.key",
        "public_key_file": "public.key",
        "deployment_ids": ["deployment_id3", "deployment_id4"]
    }],
    "iss2": [ ],
    "iss3": { }
}

default (bool) - this iss config will be used in case if client-id was not passed on the login step

client_id - this is the id received in the 'aud' during a launch

auth_login_url - the platform's OIDC login endpoint

auth_token_url - the platform's service authorization endpoint

auth_audience - the platform's OAuth2 Audience (aud). Is used to get platform's access token. Usually the same as "auth_token_url" and could be skipped but in the common case could be a different url

key_set_url - the platform's JWKS endpoint

key_set - in case if platform's JWKS endpoint somehow unavailable you may paste JWKS here

private_key_file - relative path to the tool's private key

public_key_file - relative path to the tool's public key

deployment_ids (list) - The deployment_id passed by the platform during launch

Usage with Django

Configuration using Django Admin UI

# settings.py

INSTALLED_APPS = [
    'django.contrib.admin',
    ...
    'pylti1p3.contrib.django.lti1p3_tool_config'
]

# urls.py

urlpatterns = [
    ...
    path('admin/', admin.site.urls),
    ...
]

# views.py

from pylti1p3.contrib.django import DjangoDbToolConf

tool_conf = DjangoDbToolConf()

Open Id Connect Login Request

LTI 1.3 uses a modified version of the OpenId Connect third party initiate login flow. This means that to do an LTI 1.3 launch, you must first receive a login initialization request and return to the platform.

To handle this request, you must first create a new OIDCLogin (or DjangoOIDCLogin) object:

from pylti1p3.contrib.django import DjangoOIDCLogin

oidc_login = DjangoOIDCLogin(request, tool_conf)

Now you must configure your login request with a return url (this must be preconfigured and white-listed on the tool). If a redirect url is not given or the registration does not exist an pylti1p3.exception.OIDC_Exception will be thrown.

try:
    oidc_login.redirect(get_launch_url(request))
except OIDC_Exception:
    # display error page
    log.error('Error doing OIDC login')

With the redirect, we can now redirect the user back to the tool. There are three ways to do this:

This will add a HTTP 302 location header:

oidc_login.redirect(get_launch_url(request))

This will display some javascript to do the redirect instead of using a HTTP 302:

oidc_login.redirect(get_launch_url(request), js_redirect=True)

You can also get the url you need to redirect to, with all the necessary query parameters (if you would prefer to redirect in a custom way):

redirect_obj = oidc_login.get_redirect_object()
redirect_url = redirect_obj.get_redirect_url()

Redirect is done, we can move onto the launch.

LTI Message Launches

Now that we have done the OIDC log the platform will launch back to the tool. To handle this request, first we need to create a new MessageLaunch (or DjangoMessageLaunch) object.

message_launch = DjangoMessageLaunch(request, tool_conf)

Once we have the message launch, we can validate it. Validation is transparent - it's done once before you try to access the message body:

try:
    launch_data = message_launch.get_launch_data()
except LtiException:
    log.error('Launch validation failed')

You may do it more explicitly:

try:
    launch_data = message_launch.set_auto_validation(enable=False).validate()
except LtiException:
    log.error('Launch validation failed')

Now we know the launch is valid we can find out more information about the launch.

Check if we have a resource launch or a deep linking launch:

if message_launch.is_resource_launch():
    # Resource Launch!
elif message_launch.is_deep_link_launch():
    # Deep Linking Launch!
else:
    # Unknown launch type

Check which services we have access to:

if message_launch.has_ags():
    # Has Assignments and Grades Service
if message_launch.has_nrps():
    # Has Names and Roles Service

Accessing Cached Launch Requests

It is likely that you will want to refer back to a launch later during subsequent requests. This is done using the launch id to identify a cached request. The launch id can be found using:

launch_id = message_launch.get_launch_id()

Once you have the launch id, you can link it to your session and pass it along as a query parameter.

Retrieving a launch using the launch id can be done using:

message_launch = DjangoMessageLaunch.from_cache(launch_id, request, tool_conf)

Once retrieved, you can call any of the methods on the launch object as normal, e.g.

if message_launch.has_ags():
    # Has Assignments and Grades Service

Deep Linking Responses

If you receive a deep linking launch, it is very likely that you are going to want to respond to the deep linking request with resources for the platform.

To create a deep link response you will need to get the deep link for the current launch:

deep_link = message_launch.get_deep_link()

Now we need to create pylti1p3.deep_link_resource.DeepLinkResource to return:

resource = DeepLinkResource()
resource.set_url("https://my.tool/launch")\
    .set_custom_params({'my_param': my_param})\
    .set_title('My Resource')

Everything is set to return the resource to the platform. There are two methods of doing this.

The following method will output the html for an aut-posting form for you.

deep_link.output_response_form([resource1, resource2])

Alternatively you can just request the signed JWT that will need posting back to the platform by calling.

deep_link.get_response_jwt([resource1, resource2])

Names and Roles Service

Before using names and roles you should check that you have access to it:

if not message_launch.has_nrps():
    raise Exception("Don't have names and roles!")

Once we know we can access it, we can get an instance of the service from the launch.

nrps = message_launch.get_nrps()

From the service we can get list of all members by calling:

members = nrps.get_members()

Assignments and Grades Service

Before using assignments and grades you should check that you have access to it:

if not launch.has_ags():
    raise Exception("Don't have assignments and grades!")

Once we know we can access it, we can get an instance of the service from the launch:

ags = launch.get_ags()

To pass a grade back to the platform, you will need to create a pylti1p3.grade.Grade object and populate it with the necessary information:

gr = Grade()
gr.set_score_given(earned_score)\
     .set_score_maximum(100)\
     .set_timestamp(datetime.datetime.utcnow().strftime('%Y-%m-%dT%H:%M:%S+0000'))\
     .set_activity_progress('Completed')\
     .set_grading_progress('FullyGraded')\
     .set_user_id(external_user_id)

To send the grade to the platform we can call:

ags.put_grade(gr)

This will put the grade into the default provided lineitem. If no default lineitem exists it will create one.

If you want to send multiple types of grade back, that can be done by specifying a pylti1p3.lineitem.LineItem:

line_item = LineItem()
line_item.set_tag('grade')\
    .set_score_maximum(100)\
    .set_label('Grade')

ags.put_grade(gr, line_item)

If a lineitem with the same tag exists, that lineitem will be used, otherwise a new lineitem will be created.

Check user's role after LTI launch

user_is_staff = message_launch.check_staff_access()
user_is_student = message_launch.check_student_access())
user_is_teacher = message_launch.check_teacher_access()
user_is_teaching_assistant = message_launch.check_teaching_assistant_access()
user_is_designer = message_launch.check_designer_access()
user_is_observer = message_launch.check_observer_access()

Usage with Flask

Open Id Connect Login Request

This is draft of API endpoint. Wrap it in library of your choice.

Create FlaskRequest adapter. Then create instance of FlaskOIDCLogin. redirect method will return instance of werkzeug.wrappers.Response that points to LTI platform if login was successful. Handle exceptions.

from flask import request, session
from pylti1p3.flask_adapter import (FlaskRequest, FlaskOIDCLogin)

def login(request_params_dict):

    tool_conf = ... # See Configuration chapter above

    # FlaskRequest by default use flask.request and flask.session
    # so in this case you may define request object without any arguments:

    request = FlaskRequest()

    # in case of using different request object (for example webargs or something like this)
    # you may pass your own values:

    request = FlaskRequest(
        cookies=request.cookies,
        session=session,
        request_data=request_params_dict,
        request_is_secure=request.is_secure
    )

    oidc_login = FlaskOIDCLogin(
        request=request,
        tool_config=tool_conf,
        session_service=FlaskSessionService(request),
        cookie_service=FlaskCookieService(request)
    )

    return oidc_login.redirect(request.get_param('target_link_uri'))

LTI Message Launches

This is draft of API endpoint. Wrap it in library of your choice.

Create FlaskRequest adapter. Then create instance of FlaskMessageLaunch. It lets you access data from LTI launch message if launch was successful. Handle exceptions.

from flask import request, session
from werkzeug.utils import redirect
from pylti1p3.flask_adapter import (FlaskRequest, FlaskMessageLaunch)

def launch(request_params_dict):

    tool_conf = ... # See Configuration chapter above

    request = FlaskRequest()

    # or

    request = FlaskRequest(
        cookies=...,
        session=...,
        request_data=...,
        request_is_secure=...
    )

    message_launch = FlaskMessageLaunch(
        request=request,
        tool_config=tool_conf
    )

    email = message_launch.get_launch_data().get('email')

    # Place your user creation/update/login logic
    # and redirect to tool content here

Cookies issue in the iframes

Some browsers may deny to save cookies in the iframes. For example Google Chrome from ver.80 deny to save all cookies in the iframes except cookies with flags Secure (i.e HTTPS usage) and SameSite=None. Safari deny to save all third-party cookies by default. pylti1p3 library contains workaround for such behaviour:

def login():
    ...
    return oidc_login\
        .enable_check_cookies()\
        .redirect(target_link_uri)

After this the special JS code will try to write and then read test cookie instead of redirect. User will see special page with asking to open current URL in the new window in case if cookies are unavailable. In case if cookies are allowed user will be transparently redirected to the next page. All texts are configurable with passing arguments:

oidc_login.enable_check_cookies(main_msg, click_msg, loading_msg)

Also you may have troubles with default framework sessions (because pylti1p3 library can't control your framework settings connected with the session ID cookie). So without necessary settings user's session could be unavailable in case of iframe usage. To avoid this troubles it is recommended to change default session adapter to the new cache adapter (with memcache/redis backend) and as a consequence allow library to set it's own LTI1.3 session id cookie (that will be set with all necessary params like Secure and SameSite=None).

Django cache data storage

from pylti1p3.contrib.django import DjangoCacheDataStorage

def login(request):
    ...
    launch_data_storage = DjangoCacheDataStorage(cache_name='default')
    oidc_login = DjangoOIDCLogin(request, tool_conf, launch_data_storage=launch_data_storage)

def launch(request):
    ...
    launch_data_storage = DjangoCacheDataStorage(cache_name='default')
    message_launch = DjangoMessageLaunch(request, tool_conf, launch_data_storage=launch_data_storage)

def restore_launch(request):
    ...
    launch_data_storage = get_launch_data_storage(cache_name='default')
    message_launch = DjangoMessageLaunch.from_cache(launch_id, request, tool_conf,
                                                    launch_data_storage=launch_data_storage)

Flask cache data storage

from flask_caching import Cache
from pylti1p3.contrib.flask import FlaskCacheDataStorage

cache = Cache(app)

def login():
    ...
    launch_data_storage = FlaskCacheDataStorage(cache)
    oidc_login = DjangoOIDCLogin(request, tool_conf, launch_data_storage=launch_data_storage)

def launch():
    ...
    launch_data_storage = FlaskCacheDataStorage(cache)
    message_launch = DjangoMessageLaunch(request, tool_conf, launch_data_storage=launch_data_storage)

def restore_launch():
    ...
    launch_data_storage = FlaskCacheDataStorage(cache)
    message_launch = DjangoMessageLaunch.from_cache(launch_id, request, tool_conf,
                                                    launch_data_storage=launch_data_storage)

Cache for Public Key

Library try to fetch platform's public key everytime on the message launch step. This public key may be stored in cache (memcache/redis) to speed-up launch process:

# Django cache storage:
launch_data_storage = DjangoCacheDataStorage()

# Flask cache storage:
launch_data_storage = FlaskCacheDataStorage(cache)

message_launch.set_public_key_caching(launch_data_storage, cache_lifetime=7200)

API to get JWKS

You may generate JWKS from Tool Config object:

tool_conf.set_public_key(iss, public_key, client_id=client_id)
jwks_dict = tool_conf.get_jwks()  # {"keys": [{ ... }]}

# or you may specify iss and client_id:
jwks_dict = tool_conf.get_jwks(iss, client_id)  # {"keys": [{ ... }]}

Don't forget to set public key because without it JWKS can't be generated. Also you may generate JWK for any public key using construction below:

from pylti1p3.registration import Registration

jwk_dict = Registration.get_jwk(public_key)
# {"e": ..., "kid": ..., "kty": ..., "n": ..., "alg": ..., "use": ...}