In short, why am I getting an "sqlalchemy.exc.InvalidRequestError: A transaction is already begun. Use subtransactions=True to allow subtransactions" error?
Following the best practices of separating and keeping the session external, I created foo(input) with a context manager instead of using the try / except / else. If I use foo(user) instead of it I get the above error. My guess is that foo() isn't committing and closing the connection. Howevere the documentation states otherwise.
Flask documentation uses a scoped_session but the SQLAlchemy documentation says "It is however strongly recommended that the integration tools provided with the web framework itself be used, if available, instead of scoped_session." Perhaps the scoped_session is causing errors across threads with the requests?
Here is my main code:
#__init__.py
import os
from flask import Flask, render_template, redirect, request, url_for
def create_app(test_config=None):
# create and configure the app
app = Flask(__name__, instance_relative_config=False)
app.config.from_object('config.DevelopmentConfig')
# set up extensions
# all flask extensions must support factory pattern
# can run these two steps from the cli
from app.database import init_db
init_db()
#app.route('/')
def index():
return render_template('index.html')
from app.auth import RegistrationForm
from app.models import User
from app.database import db_session, foo
#app.route('/register', methods=['GET', 'POST'])
def register():
form = RegistrationForm(request.form)
if request.method == 'POST' and form.validate():
user = User(form.name.data, form.email.data,
form.password.data)
foo(user)
# try:
# db_session.add(user)
# except:
# db_session.rollback()
# raise
# else:
# db_session.commit()
return redirect(url_for('login'))
return render_template('register.html', form=form)
#app.route('/login', methods=['GET'])
def login():
return render_template('login.html')
#app.teardown_appcontext
def shutdown_session(exception=None):
db_session.remove()
return app
Here is my database code:
#database.py
from sqlalchemy import create_engine
from sqlalchemy.orm import scoped_session, sessionmaker
from sqlalchemy.ext.declarative import declarative_base
_database_uri = os.environ['DATABASE_URL'] engine = create_engine(_database_uri)
db_session = scoped_session(sessionmaker(autocommit=False,
autoflush=False,
bind=engine))
Base = declarative_base() Base.query = db_session.query_property()
def init_db():
# import all modules here that might define models so that
# they will be registered properly on the metadata. Otherwise
# you will have to import them first before calling init_db()
import app.models
Base.metadata.create_all(bind=engine)
def foo(input):
with db_session.begin() as session:
session.add(input)
I'm not sure whether this will actually answer your question or not but I think it worth mentioning.
Near the last line in your database.py file, I suggest you to not alias db_session.begin() as session because you'll then be confused thinking that session is an object of Session class while it's an object of SessionTransaction class which is:
largely an internal object that in modern use provides a context manager for session transactions. SessionTransaction
You can switch to either:
with db_session() as session, session.begin():
session.add(input)
or shorter version
with db_session.begin():
db_session.add(input)
Also you need to wrap your User object creation with Session.begin() context like below:
def register():
form = RegistrationForm(request.form)
if request.method == 'POST' and form.validate():
with db_session.begin():
user = User(form.name.data, form.email.data,
form.password.data)
foo(user)
Because User model is just a proxy object that will actually execute database query under the hood. Therefore A transaction is already begun. in the creation process. The exception itself will be raised upon the next transactional query calls.
When using a Session, it’s useful to consider the ORM mapped objects that it maintains as proxy objects to database rows, which are local to the transaction being held by the Session. In order to maintain the state on the objects as matching what’s actually in the database, there are a variety of events that will cause objects to re-access the database in order to keep synchronized. It is possible to “detach” objects from a Session, and to continue using them, though this practice has its caveats. It’s intended that usually, you’d re-associate detached objects with another Session when you want to work with them again, so that they can resume their normal task of representing database state.
Session Basic
As additional answer to your last question: Perhaps the scoped_session is causing errors across threads with the requests?
No. SQLAlchemy scoped_session is actually a helper function that act as Registry for the global Session object. It is very useful in multithreading application, helping to ensure the same Session object is being used accross threads while each keeping their own data to local via threading.local api provided by python. Most web framework uses threading strategies to cope with many web requests at once hence most of them provide some integration with/without this helper.
I am using an application factory to add views to my flask application like so :
(this is not my actual application factory, and has been shortened for the sake of brevity)
def create_app(config_name='default'):
app = Flask(__name__, template_folder="templates", static_folder='static')
admin_instance = Admin(app, name='Admin')
admin_instance.add_view(EntityAdmin(Entity, db.session))
My EntityAdmin class looks like this :
class EntityAdmin(ModelView):
column_filters = [
MyCustomFilter(column=None, name='Custom')
]
My custom filter looks like this :
class MyCustomFilter(BaseSQLAFilter):
def get_options(self, view):
entities = Entity.query.filter(Entity.active == True).all()
return [(entity.id, entity.name) for entity in entities]
The problem is that it seems that the get_options function is called when the app is instantiated, running a select query every time the create_app function gets called.
So if I update my database schema and run the flask db migrate command, I get an error because the new column I added does not exist when the select query is run. The query raises an error because my database schema is not in sync with the actual database.
Can I register my views only when an actual HTTP request is made ? How can I differentiate between a request and a command ?
You have one more problem with this filter: its options are created on the application instantiation so if your list of entities was changed during the application running it would still return the same list of options.
To fix both problems you don't need to postpone views registrations. You need the filter to get the list of options every time it is used.
This SO answer to the question "Resetting generator object in Python" describes a way to reuse a generator (in your case — a database query):
from flask import has_app_context
def get_entities():
# has_app_context is used to prevent database access
# when application is not ready yet
if has_app_context():
for entity in Entity.query.filter(Entity.active.is_(True)):
yield entity.id, entity.name
class ReloadingIterator:
def __init__(self, iterator_factory):
self.iterator_factory = iterator_factory
def __iter__(self):
return self.iterator_factory()
class MyCustomFilter(BaseSQLAFilter):
def get_options(self, view):
# This will return a generator which is
# reloaded every time it is used
return ReloadingIterator(get_entities)
The problem is that the query to the Entity table can be called multiple times during request. So I usually cache the result for a single request using Flask globals:
def get_entities():
if has_app_context():
if not hasattr(g, 'entities'):
query = Entity.query.filter(Entity.active.is_(True))
g.entities = [(entity.id, entity.name) for entity in query]
for entity_id, entity_name in g.entities:
yield entity_id, entity_name
I'm using Flask, Flask-SQLAlchemy, Flask-Marshmallow + marshmallow-sqlalchemy, trying to implement REST api PUT method. I haven't found any tutorial using SQLA and Marshmallow implementing update.
Here is the code:
class NodeSchema(ma.Schema):
# ...
class NodeAPI(MethodView):
decorators = [login_required, ]
model = Node
def get_queryset(self):
if g.user.is_admin:
return self.model.query
return self.model.query.filter(self.model.owner == g.user)
def put(self, node_id):
json_data = request.get_json()
if not json_data:
return jsonify({'message': 'Invalid request'}), 400
# Here is part which I can't make it work for me
data, errors = node_schema.load(json_data)
if errors:
return jsonify(errors), 422
queryset = self.get_queryset()
node = queryset.filter(Node.id == node_id).first_or_404()
# Here I need some way to update this object
node.update(data) #=> raises AttributeError: 'Node' object has no attribute 'update'
# Also tried:
# node = queryset.filter(Node.id == node_id)
# node.update(data) <-- It doesn't if know there is any object
# Wrote testcase, when user1 tries to modify node of user2. Node doesn't change (OK), but user1 gets status code 200 (NOT OK).
db.session.commit()
return jsonify(), 200
UPDATED, 2022-12-08
Extending the ModelSchema from marshmallow-sqlalchemy instead of Flask-Marshmallow you can use the load method, which is defined like this:
load(data, *, session=None, instance=None, transient=False, **kwargs)
Putting that to use, it should look like that (or similar query):
node_schema.load(json_data, session= current_app.session, instance=Node().query.get(node_id))
And if you want to load without all required fields of Model, you can add the partial=True argument, like this:
node_schema.load(json_data, instance=Node().query.get(node_id), partial=True)
See the docs for more info (does not include definition of ModelSchema.load).
See the code for the load definition.
I wrestled with this issue for some time, and in consequence came back again and again to this post. In the end what made my situation difficult was that there was a confounding issue involving SQLAlchemy sessions. I figure this is common enough to Flask, Flask-SQLAlchemy, SQLAlchemy, and Marshmallow, to put down a discussion. I certainly, do not claim to be an expert on this, and yet I believe what I state below is essentially correct.
The db.session is, in fact, closely tied to the process of updating the DB with Marshmallow, and because of that decided to to give the details, but first the short of it.
Short Answer
Here is the answer I arrived at for updating the database using Marshmallow. It is a different approach from the very helpful post of Jair Perrut. I did look at the Marshmallow API and yet was unable to get his solution working in the code presented, because at the time I was experimenting with his solution I was not managing my SQLAlchemy sessions properly. To go a bit further, one might say that I wasn't managing them at all. The model can be updated in the following way:
user_model = user_schema.load(user)
db.session.add(user_model.data)
db.session.commit()
Give the session.add() a model with primary key and it will assume an update, leave the primary key out and a new record is created instead. This isn't all that surprising since MySQL has an ON DUPLICATE KEY UPDATE clause which performs an update if the key is present and creates if not.
Details
SQLAlchemy sessions are handled by Flask-SQLAlchemy during a request to the application. At the beginning of the request the session is opened, and when the request is closed that session is also closed. Flask provides hooks for setting up and tearing down the application where code for managing sessions and connections may be found. In the end, though, the SQLAlchemy session is managed by the developer, and Flask-SQLAlchemy just helps. Here is a particular case that illustrates the management of sessions.
Consider a function that gets a user dictionary as an argument and uses that with Marshmallow() to load the dictionary into a model. In this case, what is required is not the creation of a new object, but the update of an existing object. There are 2 things to keep in mind at the start:
The model classes are defined in a python module separate from any code, and these models require the session. Often the developer (Flask documentation) will put a line db = SQLAlchemy() at the head of this file to meet this requirement. This in fact, creates a session for the model.
from flask_sqlalchemy import SQLAlchemy
db = SQLAlchemy()
In some other separate file there may be a need for a SQLAlchemy session as well. For example, the code may need to update the model, or create a new entry, by calling a function there. Here is where one might find db.session.add(user_model) and db.session.commit(). This session is created in the same way as in the bullet point above.
There are 2 SQLAlchemy sessions created. The model sits in one (SignallingSession) and the module uses its own (scoped_session). In fact, there are 3. The Marshmallow UserSchema has sqla_session = db.session: a session is attached to it. This then is the third, and the details are found in the code below:
from marshmallow_sqlalchemy import ModelSchema
from donate_api.models.donation import UserModel
from flask_sqlalchemy import SQLAlchemy
db = SQLAlchemy()
class UserSchema(ModelSchema):
class Meta(object):
model = UserModel
strict = True
sqla_session = db.session
def some_function(user):
user_schema = UserSchema()
user['customer_id'] = '654321'
user_model = user_schema.load(user)
# Debug code:
user_model_query = UserModel.query.filter_by(id=3255161).first()
print db.session.object_session(user_model_query)
print db.session.object_session(user_model.data)
print db.session
db.session.add(user_model.data)
db.session.commit()
return
At the head of this module the model is imported, which creates its session, and then the module will create its own. Of course, as pointed out there is also the Marshmallow session. This is entirely acceptable to some degree because SQLAlchemy allows the developer to manage the sessions. Consider what happens when some_function(user) is called where user['id'] is assigned some value that exists in the database.
Since the user includes a valid primary key then db.session.add(user_model.data) knows that it is not creating a new row, but updating an existing one. This behavior should not be surprising, and is to be at least somewhat expected since from the MySQL documentation:
13.2.5.2 INSERT ... ON DUPLICATE KEY UPDATE Syntax
If you specify an ON DUPLICATE KEY UPDATE clause and a row to be inserted would cause a duplicate value in a UNIQUE index or PRIMARY KEY, an UPDATE of the old row occurs.
The snippet of code is then seen to be updating the customer_id on the dictionary for the user with primary key 32155161. The new customer_id is '654321'. The dictionary is loaded with Marshmallow and a commit done to the database. Examining the database it can be found that it was indeed updated. You might try two ways of verifying this:
In the code: db.session.query(UserModel).filter_by(id=325516).first()
In MySQL: select * from user
If you were to consider the following:
In the code: UserModel.query.filter_by(id=3255161).customer_id
You would find that the query brings back None. The model is not synchronized with the database. I have failed to manage our SQLAlchemy sessions correctly. In an attempt to bring clarity to this consider the output of the print statements when separate imports are made:
<sqlalchemy.orm.session.SignallingSession object at 0x7f81b9107b90>
<sqlalchemy.orm.session.SignallingSession object at 0x7f81b90a6150>
<sqlalchemy.orm.scoping.scoped_session object at 0x7f81b95eac50>
In this case the UserModel.query session is different from the Marshmallow session. The Marshmallow session is what gets loaded and added. This means that querying the model will not show our changes. In fact, if we do:
db.session.object_session(user_model.data).commit()
The model query will now bring back the updated customer_id! Consider the second alternative where the imports are done through flask_essentials:
from flask_sqlalchemy import SQLAlchemy
from flask_marshmallow import Marshmallow
db = SQLAlchemy()
ma = Marshmallow()
<sqlalchemy.orm.session.SignallingSession object at 0x7f00fe227910>
<sqlalchemy.orm.session.SignallingSession object at 0x7f00fe227910>
<sqlalchemy.orm.scoping.scoped_session object at 0x7f00fed38710>
And the UserModel.query session is now the same as the user_model.data (Marshmallow) session. Now the UserModel.query does reflect the change in the database: the Marshmallow and UserModel.query sessions are the same.
A note: the signalling session is the default session that Flask-SQLAlchemy uses. It extends the default session system with bind selection and modification tracking.
I have rolled out own solution. Hope it helps someone else. Solution implements update method on Node model.
Solution:
class Node(db.Model):
# ...
def update(self, **kwargs):
# py2 & py3 compatibility do:
# from six import iteritems
# for key, value in six.iteritems(kwargs):
for key, value in kwargs.items():
setattr(self, key, value)
class NodeAPI(MethodView):
decorators = [login_required, ]
model = Node
def get_queryset(self):
if g.user.is_admin:
return self.model.query
return self.model.query.filter(self.model.owner == g.user)
def put(self, node_id):
json_data = request.get_json()
if not json_data:
abort(400)
data, errors = node_schema.load(json_data) # validate with marshmallow
if errors:
return jsonify(errors), 422
queryset = self.get_queryset()
node = queryset.filter(self.model.id == node_id).first_or_404()
node.update(**data)
db.session.commit()
return jsonify(message='Successfuly updated'), 200
Latest Update [2020]:
You might facing the issue of mapping keys to the database models. Your request body have only updated fields so, you want to change only those without affecting others. There is an option to write multiple if conditions but that's not a good approach.
Solution
You can implement patch or put methods using sqlalchemy library only.
For example:
YourModelName.query.filter_by(
your_model_column_id = 12 #change 12: where condition to find particular row
).update(request_data)
request_data should be dict object. For ex.
{
"your_model_column_name_1": "Hello",
"your_model_column_name_2": "World",
}
In above case, only two columns will be updated that is: your_model_column_name_1 and your_model_column_name_2
Update function maps request_data to the database models and creates update query for you. Checkout this: https://docs.sqlalchemy.org/en/13/core/dml.html#sqlalchemy.sql.expression.update
Previous answer seems to be outdated as ModelSchema is now deprecated.
You should instead SQLAlchemyAutoSchema with the proper options.
class NodeSchema(SQLAlchemyAutoSchema):
class Meta:
model = Node
load_instance = True
sqla_session = db.session
node_schema = NodeSchema()
# then when you need to update a Node orm instance :
node_schema.load(node_data, instance=node, partial=True)
db.session.update()
Below is my solution with Flask-Marshmallow + marshmallow-sqlalchemy bundle as the author requested initially.
schemas.py
from flask import current_app
from flask_marshmallow import Marshmallow
from app.models import Node
ma = Marshmallow(current_app)
class NodeSchema(ma.SQLAlchemyAutoSchema):
class Meta:
model = Node
load_instance = True
load_instance is a key point here to make an update further.
routes.py
from flask import jsonify, request
from marshmallow import ValidationError
from app import db
#bp.route("/node/<node_uuid>/edit", methods=["POST"])
def edit_node(node_uuid):
json_data = request.get_json(force=True, silent=True)
node = Node.query.filter_by(
node_uuid=node_uuid
).first()
if node:
try:
schema = NodeSchema()
json_data["node_uuid"] = node_uuid
node = schema.load(json_data, instance=node)
db.session.commit()
return schema.jsonify(node)
except ValidationError as err:
return jsonify(err.messages), 422
else:
return jsonify("Not found"), 404
You have to check for existence of Node first, otherwise the new instance will be created.
I'm storing some variables in the session when the user logs in, to use later to populate a field.
from flask_wtf import Form
from wtforms import SelectField
from flask import session
class InstitutionForm(Form):
city = session['city']
city_tuples = [(x, x) for x in city]
organisation = SelectField(
'organisation',
choices=city_tuples
)
class Institution(View):
methods = ['POST']
def dispatch_request(self):
form = InstitutionForm()
return render_template(
'form/institution.html',
form=form)
When I run the app I get the error 'working outside of request context'. What is the correct way to populate this field from the session data?
Code in a class definition is executed at import time, not when the class is instantiated. You need to move the access to session to the __init__ method so that it will be accessed when creating a form in a view function.
class Institution(Form):
organization = SelectField()
def __init__(self, *args, **kwargs):
self.organization.kwargs['choices'] = [(x, x) for x in session.get('city', ('not set',))]
super().__init__(*args, **kwargs)
This applies to anything that needs an application or request context, such as a database query, not just the session.
I'm using pyramid_formalchemy 0.4.1...well I'm trying to use it.
When requests come in to my app I keep seeing pyramid_formalchemy making strange assumptions about what will be in the request object. My requests fail because in pyramid_formalchemy.views (starting at line: 58) the ModelView class has a constructor with the following code:
def __init__(self, context, request):
self.context = context
self.request = request
self.session = request.session_factory
self.fieldset_class = request.forms.FieldSet
self.grid_class = request.forms.Grid
The thing is my request object has a 'session' not a 'session_factory'. It also doesn't have 'forms'
Here is what I've done:
Create a RootFactory that extends pyramid_formalchemy.resources.Models
I call config.formalchemy_admin() passing it my RootFactory and my DBSession
I created an empty forms.py file.
What am I missing in my setup? Any ideas?
Thanks.
This stuff is configurable.
See the source
Not sure it's really documented..