select_related() ConnectionError

[radulucaciu]

Hi,

I am having an interesting situation. We have been using MongoEngine in our project for a good few years now and this issue appeared out of nowhere (no update, no change to the code).

Calling select_related() on the results of a query raises a mongoengine.connection.ConnectionError: You have not defined a default connection Exception

The call that causes the issue:
credentials = Credential.objects(id__in=ids).select_related()

I noticed that I am getting the exception only the first time time I call select_related() and only on this object. Any subsequent calls (on the same collection) work as expected.

While trying to debug, I got to this:

> /usr/local/lib/python2.7/dist-packages/mongoengine/dereference.py(129)_fetch_objects()
    127         object_map = {}
    128         for collection, dbrefs in self.reference_map.iteritems():
--> 129             if hasattr(collection, 'objects'):  # We have a document class for the refs
    130                 col_name = collection._get_collection_name()
    131                 refs = [dbref for dbref in dbrefs

ipdb> collection
<class 'insights.models.company.Company'>
ipdb> hasattr(collection, 'objects')
False
ipdb> hasattr(collection, 'objects')
True

Can this be caused by some lazy loading or some configuration issue?

Thank you!

[erdenezul]

can you provide sample code to reproduce this error?

[radulucaciu]

Hi @erdenezul so sorry it took so long to reply. It took forever to figure out how to reproduce the issue on anything but our live environment and then to trim it down where it could be easily reproduced on other systems.

I am adding a zip file containing the collection dumps and the three models for the respective collections.

• credentials: 1 entry in collection
• companies: 4 entries in collection
• users: 1 entry in collection

Once imported into MongoDB, simply calling the following line will result in the exception being raised. Calling the same line again, will work as expected.

Credential.objects(id__in=['553a30b05552290a2d0481e9']).select_related()

Peculiar aspect:
The companies collection contains 4 items with 3 of them not being related to the credential. However, if we remove any of these items from the collection, the error does not show up.

files.zip

[erdenezul]

I got no error

[jozsi]

I was able to reproduce using the files provided by @radulucaciu

>>> from mongoengine import *
>>> connect('meta', 'meta-db')
MongoClient(host=['localhost:27017'], document_class=dict, tz_aware=False, connect=True, read_preference=Primary())
>>> from company import Company
>>> from credential import Credential
>>> from users import User
>>> Credential.objects(id__in=['553a30b05552290a2d0481e9']).select_related()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Library/Python/2.7/site-packages/mongoengine/queryset/base.py", line 756, in select_related
    return queryset._dereference(queryset, max_depth=max_depth)
  File "/Library/Python/2.7/site-packages/mongoengine/dereference.py", line 76, in __call__
    self.object_map = self._fetch_objects(doc_type=doc_type)
  File "/Library/Python/2.7/site-packages/mongoengine/dereference.py", line 149, in _fetch_objects
    references = get_db()[collection].find({'_id': {'$in': refs}})
  File "/Library/Python/2.7/site-packages/mongoengine/connection.py", line 226, in get_db
    conn = get_connection(alias)
  File "/Library/Python/2.7/site-packages/mongoengine/connection.py", line 146, in get_connection
    raise MongoEngineConnectionError(msg)
mongoengine.connection.MongoEngineConnectionError: You have not defined a default connection
>>> Credential.objects(id__in=['553a30b05552290a2d0481e9']).select_related()
[<Credential: Credential object>]
>>> 

Fresh install of everything (mongodb, pip, mongoengine), latest stable versions.

I wasn't sure how to import the collection data, so I manually created the database and connections and inserted each document (replacing {"$oid": "..."} with ObjectId("...")). I created a mongodump, attaching.
meta.zip

cc @erdenezul

[radulucaciu]

Thank you @jozsi

I wonder if this is influenced by the version of MongoDB - I am using v. 3.2.6

Package versions are:

mongoengine==0.14.3
pymongo==3.2.2

[bagerard]

OK It was a tricky one, actually the problem lies in the Company class definition that conflicts with the existing data in the database.

The model you sent us is:

class Company(Document):
    meta = {
        'db_alias': 'meta-db',
        'collection': 'companies',
    }
    name = StringField(unique=True)

And the data in the db (restored from the file you attached):

In [3]: Company.objects.as_pymongo()
Out[3]: [{'_id': ObjectId('53a44802833be3e812e4ec3f'), 'name': 'WH'}, {'_id': ObjectId('59fc5b1b5998373e8c320ba2'), 'name': 'BMW'}, {'_id': ObjectId('5a02f42e5998373e8c320bc3'), 'name': 'MvP'}, {'_id': ObjectId('59fc55a75998373e8c320b9e'), 'name': 'BMW'}]

You actually have 2 companies with the same 'name' (being BMW) ubut your model defines a unique=True, which creates an index behind the scene.

You've debugged it properly until the line:

> /usr/local/lib/python2.7/dist-packages/mongoengine/dereference.py(129)_fetch_objects()
    127         object_map = {}
    128         for collection, dbrefs in self.reference_map.iteritems():
--> 129             if hasattr(collection, 'objects'):  # We have a document class for the refs
    130                 col_name = collection._get_collection_name()
    131                 refs = [dbref for dbref in dbrefs

What happens here (under python2):
Mongoengine will access .objects, which will call ._get_collection() and then .ensure_indexes() which will finally attempt to create the unique index there https://github.com/MongoEngine/mongoengine/blob/master/mongoengine/document.py#L885.
The creation of the index fails and throws an exception but under Python2 hasattr swallows any exceptions (https://mail.python.org/pipermail/python-3000/2008-January/011771.html), thus it behaves like if hasattr simply returns False but it is actually swallowing an exception before returning False.

The behavior of hasattr changed in Python3 and if you run the same code in Python3, it fails as followed:

~/.pyenv/versions/3.6.5/envs/cmp/lib/python3.6/site-packages/mongoengine/document.py in ensure_indexes(cls)
    876                 if IS_PYMONGO_3:
--> 877                     collection.create_index(fields, background=background, **opts)
~/.pyenv/versions/3.6.5/envs/cmp/lib/python3.6/site-packages/pymongo/collection.py in create_index(self, keys, session, **kwargs)
-> 1958         self.__create_index(keys, kwargs, session, **cmd_options)
...
~/.pyenv/versions/3.6.5/envs/cmp/lib/python3.6/site-packages/pymongo/helpers.py in _check_command_response(response, msg, allowable_errors, parse_write_concern_error)
--> 148                 raise DuplicateKeyError(errmsg, code, response)

DuplicateKeyError: E11000 duplicate key error collection: meta.companies index: name_1 dup key: { : "BMW" }

This is the root cause of your issue. It sounds like the unique index was added to the Document class at a later stage.
What we can do to improve this is try/except the access to .objects without relying on hasattr, that way any exceptions will get to the user in both Python2 and Python3. I'll push a PR shortly to improve this

[bagerard]

I believe this can be closed

[radulucaciu]

@bagerard Thank you so much! Tbh, I don't remember if the unique index was added at the beginning or later on and it's interesting that it was accepted with duplicate data.
Appreciate your help!