couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jim Klo <jim....@sri.com>
Subject Re: userCtx extra information
Date Thu, 30 Aug 2012 20:02:00 GMT

> 
> Like I said there are some features actually missing in couch that
> would ease that. (partial updates and partial fetch). Of course a view
> could be use to see only profile but that would be a hack.
> 

I won't argue there - I've not thought enough about how one designs a secure document db in
CouchDB vs how one designs secure RDBMS.  Both are fundamentally different, and require you
to cast preconceived notions aside on how security is should be applied by using reference
from one model and trying to force it onto the other.

> And this is not only about being a db or something alternative. This
> is more about security here. Even in the old world data for
> authentication and profiles are generally separated. For a good
> reason. This isn't generally the same person that have access to them.
> And personnaly I would see a profile linked to a user but not in the
> same doc.
> 

I think that's a false premise to base separation upon. Security should *always* apply to
the effective user consistently to determine access control. In a large system, if I don't
do that, one could unintentionally elevate user permissions. This 'idea' that users being
separated from 'logins' come from traditional RDBMS design through normalization for enhancing
the performance of lookups and relationships, not because it couldn't be combined. Systems
like MS SQL, MySQL and others allow you to map an external user (which is typ tied to system
access rights) and to an internal user (which is tied to schema access rights).  All of this
which was coined and conceived before document db's ever really existed.  And even in those
models - they are following an extremely orthogonal design.  Even in more modern solutions,
that use SSO/LDAP for authentication - there's one user object that's used everywhere from
user application login down to validating access for an app that connects to the SQL DB (think
Active Directory + MS SQL + SharePoint, many users have access to the same data - defined
by roles stored in AD; access authenticated by Roles/Groups assigned in MSSQL). 

CouchDB works with documents, users should be documents and not require a "JOIN" of any sort
to fetch a different document relating to that user.  That "JOIN" operation isn't native to
CouchDB and would break orthogonality.

Nothing that I'm describing prevents  you from adopting an old world view mentality, I'm just
encouraging that we follow the CouchDB design pattern and like RDBMS - be consistent and orthogonal
in the way objects are stored and organized, including internal system objects, and store
users as a single document. In CouchDB, would you store an ISBN for a book document in separate
document from it's title, language, and abstract? Why would you do that with a user then?

> Also I'm pretty sure that the reason people are asking about
> populating this userCtx is because they lack the possibility to query
> internally the db. This can be changed.
> 

Agreed. However the general argument for non-separation of user login from profile is more
at question. userCtx is more than likely just a result of what is minimally expected to be
present in the data model.  It's analogous to requesting a view with just ids or with docs.
 Extending to add internal query, which would enable lookup, seems like a more involved solution,
and creates potential for real performance issues (think cascading triggers) and security
breach. 

Adding query is to get additional data from a specific object already being partially injected,
feels like you're bringing a machine gun to knife fight.

- Jim  

> - benoît
> 
> 
>> Wendall
>> 
>> 
>> On 08/30/2012 10:06 AM, Jim Klo wrote:
>>> 
>>> Been following this thread a bit, and thought there's a unique
>>> characteristic of CouchDB here that's being overlooked that should allow us
>>> to address this problem in potentially a unique way.
>>> 
>>> There's basically 2 camps (dare I say religions) here:
>>> 1) Traditional DB folks CouchDB's _users is special an should be reserved
>>> for "applications as clients" and that end users should be stored
>>> differently and separately.  Most clients that connect to the db will some
>>> application that will manage data access.  App logic will decide who gets to
>>> read/write - not the user who is actually connecting to the DB; their the
>>> conductor.
>>> 2) App developers: Who just want an HTTP API to save user info, utilize
>>> for authentication, session management, and access control - they don't want
>>> to rebuild this for every app, and certainly aren't interested in having to
>>> write an additional server-side tier to make this work securely. Real people
>>> use apps, real people are the ones who actually have where with all to fill
>>> out a form and place it into some storage area, read some bit of privileged
>>> or non-privileged info, etc... they don't really give a rip about some
>>> middle-man user who's logic decides if they have the right credentials to
>>> access.
>>> 
>>> CouchDB is a unique product: It can act as both a HTTP application server
>>> and an indexed document store. I can't think of another solution that does
>>> this so elegantly. Because of this, I think _users should be treated
>>> differently than a traditional DB that uses JDBC/Named Pipe/ODBC/pick your
>>> db connectivity poison to access the DB. Ultimately, in both camps, users
>>> are being used to restrict the level of access to documents.  CouchDB for
>>> the most part allows simple apps to 'eliminate the middle tier', which is
>>> what the traditional db camp is trying, i think, to argue is less secure,
>>> and thus one needs to mimic the middle tier and use isolation of data to
>>> protect that _users db. The reality the way the CouchDB user model, IMHO can
>>> actually be more secure, because the same user who is connecting to the app
>>> via web, can be the exact same user that reads/writes to the document
>>> store... no more middle tier application with god like powers having to
>>> mediate user permissions.
>>> 
>>> CouchDB, for the most part is very orthogonally designed.  Configuration /
>>> Users / Replication / Document DB are all essentially stored/treated in the
>>> same manner (or at least shares a common API) that is more or less
>>> schema-less (config is a tad different in that it persists to an INI file).
>>> The _users db should continue to follow the same schema-less design it has
>>> been; documents to be used for authentication should need to follow nothing
>>> more than a duck typed convention any extra data should just be treated as
>>> noise to parts of the core app that don't need it. If you want to store
>>> additional info in user document, go for it - authentication should just
>>> ignore it - its not hurting anyone, really.
>>> 
>>> Those arguing about making the _users db holy by hiding it completely and
>>> limiting what you can store in the object breaks the fantastic orthogonality
>>> that CouchDB enjoys today.
>>> 
>>> My vote? Continue keeping CouchDB orthogonal, in fact make it even more
>>> so! IMHO - I think being able to store additional data that can be retrieved
>>> via userCtx would be good; however I think there should be a flag that is
>>> used when querying to include the extra data,(like include_docs=true, but
>>> maybe full_userCtx=true)?.  As an app developer - you shouldn't need access
>>> to all of the users additional attributes for every request.  That is
>>> potentially a lot of extra data (equates to memory use/etc) that would go
>>> unused on probably 90% of the requests made.  The middle tier / traditional
>>> db guys & gals are happy - they can still do their own thing through god
and
>>> demi-god apps and manage access via their middle-tier (god help them!); and
>>> the CouchApp guys & gals can go on with their own thing and leverage built
>>> in _users functionality within CouchDB and get them access to the additional
>>> userCtx as needed in their list/show/etc functions (heck they are already
>>> double-dipping on the appserver / db combo).
>>> 
>>> Cheers,
>>> 
>>> *
>>> *
>>> *
>>> *Jim Klo*
>>> 
>>> Senior Software Engineer
>>> Center for Software Engineering
>>> SRI International
>>> *
>>> *
>>> *
>>> *
>>> *
>>> *
>>> t.@nsomnac
>>> *
>>> *
>>> 
>>> *
>>> 
>>> On Aug 30, 2012, at 8:13 AM, Gabriel Mancini wrote:
>>> 
>>>> I think this can be flexible to offer some aditional information as
>>>> optional feature in userCtx.
>>>> and with possibility to create a design document with validation login
>>>> and
>>>> use this info to auth/reject the user.
>>>> this way we can trigger an 403 http error. this is more "REST" approach i
>>>> guess
>>>> 
>>>> 
>>>> On Thu, Aug 30, 2012 at 9:14 AM, john.tiger <john.tigernassau@gmail.com
>>>> <mailto:john.tigernassau@gmail.com>>wrote:
>>>> 
>>>>> 
>>>>> 
>>>>> On 08/30/2012 05:28 AM, Robert Newson wrote:
>>>>> 
>>>>>> In CouchDB 1.2.0 mandatory filters have been added when accessing
>>>>>> _users
>>>>>> such that no user can see another users document. So, you can allow
>>>>>> signup
>>>>>> (jquery.couch.js includes such a function) safely.
>>>>>> 
>>>>>> As for expanding the userCtx, it's not something we're rejecting
out of
>>>>>> hand or to annoy you. The userCtx object is provided to allow
>>>>>> authorization
>>>>>> decisions and the CouchDB security model is that these decisions
are
>>>>>> made
>>>>>> on the user's name and/or the user's roles. Any additional information
>>>>>> in
>>>>>> there could also be used for authorization purposes which would alter
>>>>>> the
>>>>>> security model. Now, that might not be a bad thing, but it's not
>>>>>> something
>>>>>> to be done lightly.
>>>>>> 
>>>>>> Further, practical points to consider. The name and roles are needed
>>>>>> often and so they are stored in an in-memory cache. The larger you
make
>>>>>> userCtx, the less caching will be possible and performance will suffer.
>>>>>> If
>>>>>> you then further consider a sharded CouchDB system (after the BigCouch
>>>>>> merge) then the user information you need, on a cache miss, might
be on
>>>>>> another machine.
>>>>>> 
>>>>>> The _users db exists to provide authentication/authorization services
>>>>>> to
>>>>>> CouchDB, it's unfortunate that it's also a CouchDB database as well
as
>>>>>> this
>>>>>> leads to misconceptions. Most RDBMS's use system databases to provide
a
>>>>>> similar service and, likewise, don't expand those things for
>>>>>> application
>>>>>> use.
>>>>>> 
>>>>> 
>>>>> from a good practice standpoint, Robert and Benoit are correct.  from
a
>>>>> coding standpoint, maybe someone has a ready made example of a view that
>>>>> shows a check for auth/auth then a query on the associated user doc.
>>>>> Then
>>>>> the reverse, a post that creates a user and the user doc from the one
>>>>> form.
>>>>> It's fairly simple stuff but would save time and effort for everyone
who
>>>>> is using something like this (many apps).   This "how-to" could go into
>>>>> the
>>>>> couch guide or simply a couch how-to or cookbook wiki, something Couch
>>>>> needs.
>>>>> 
>>>>> 
>>>>> 
>>>>> 
>>>>> 
>>>>> 
>>>>>> B.
>>>>>> 
>>>>>> 
>>>>>> On 30 Aug 2012, at 12:06, Aliaksandr Barysiuk wrote:
>>>>>> 
>>>>>> We have a simple form for user registration with standard fields:
>>>>>> fname,
>>>>>>> 
>>>>>>> lname, email, avatar url, etc. We want to have some section on
the
>>>>>>> page
>>>>>>> that shows some of these field after user successfully logged
in (like
>>>>>>> 'Logged in as $fname + #lname', show avatar). Nothing special.
For me
>>>>>>> userCtx is perfect place to store such information, because logged
>>>>>>> user
>>>>>>> always have access to it (we don't need to make any additional
call).
>>>>>>> It's
>>>>>>> simple. What is the reason to keep user information separately?
>>>>>>> 
>>>>>>> I have already discuss this question with R.Newson and i can't
say
>>>>>>> that
>>>>>>> i'm satisfied with the answers. Of course to create new user
(signup)
>>>>>>> you
>>>>>>> need to be an admin, but we cannot specify admin credential on
the
>>>>>>> client
>>>>>>> because everyone can use them, right? Other option is to open
_usersdb
>>>>>>> for
>>>>>>> reading and writing which is crazy and totally unacceptable.
As a
>>>>>>> result
>>>>>>> now we use node.js with JSONP call that sent all user info to
node
>>>>>>> server
>>>>>>> and there user entry is saved into _usersdb with admin credentials.
Or
>>>>>>> we
>>>>>>> need to write own authorization logic and use Couchdb as usual
storage
>>>>>>> which may takes a lot of time. Signup, reset password functions
-
>>>>>>> usual
>>>>>>> operations on every web site.
>>>>>>> 
>>>>>>> Conclusion: couchdb makes development of web apps very convinient.
But
>>>>>>> user management is the important part of each web app and now
we HAVE
>>>>>>> to
>>>>>>> use external system (node.js in our case). It would be great
if
>>>>>>> Couchdb can
>>>>>>> handle it without any other service.
>>>>>>> 
>>>>>>> PS. Sorry for long posts. I have a lot of thoughts to write down)
>>>>>>> 
>>>>>>> Alex
>>>>>>> 
>>>>>>> On 30/08/12 12:56, Benoit Chesneau wrote:
>>>>>>> 
>>>>>>>> On Thu, Aug 30, 2012 at 8:22 AM, Aliaksandr Barysiuk
>>>>>>>> <a.barysiuk@gmail.com <mailto:a.barysiuk@gmail.com>>
 wrote:
>>>>>>>> 
>>>>>>>> 
>>>>>>>>> I don't understand your position. Now Couchdb user management
is in
>>>>>>>>> rudimentary state. It even doesn't allow to create new
user easily
>>>>>>>>> (without
>>>>>>>>> opening access to _users db or using external service).
I understand
>>>>>>>>> that
>>>>>>>>> Couchdb developers bothers more about performance, scalability
etc.
>>>>>>>>> But you
>>>>>>>>> totally forgot about simple cases. I'm sure that a big
percent of
>>>>>>>>> web
>>>>>>>>> apps
>>>>>>>>> use user management. So what is the problem to develop
convenient
>>>>>>>>> authentication/authorization system? It is normal to
have extra
>>>>>>>>> information
>>>>>>>>> in users db.
>>>>>>>>> 
>>>>>>>> Well you can't put extra informations in the postgresql user
>>>>>>>> database.
>>>>>>>> The question is more why do you want such extra information
in this
>>>>>>>> db?
>>>>>>>> 
>>>>>>>> - Is this because you don't have access to couchdb internally
in your
>>>>>>>> couchapps?
>>>>>>>> - Because you need to be an admin to create a user? (ie lacking
of a
>>>>>>>> permission to create a user for anyone)
>>>>>>>> - other? (and why in this case)
>>>>>>>> 
>>>>>>>> And it's also normal that people want to see this information
>>>>>>>> 
>>>>>>>>> in session. Using roles for that just shows you that
the problem
>>>>>>>>> exists. And
>>>>>>>>> it is not a solution 'fix this bug' since it provides
easiest way do
>>>>>>>>> what
>>>>>>>>> Couchdb doesnt' provide out-of-the-box. And even you
fix roles to
>>>>>>>>> have
>>>>>>>>> only
>>>>>>>>> strings it is not a big deal to concatenate all my data
and put it
>>>>>>>>> as
>>>>>>>>> string
>>>>>>>>> in a role. It just adds some extra work on parsing that.
>>>>>>>>> 
>>>>>>>> Everything is possible. But doing it the clean way is hard.
I do
>>>>>>>> think
>>>>>>>> that roles should be anything but roles. ie things to manage
>>>>>>>> permissions.
>>>>>>>> 
>>>>>>>> Conclusion: I don't want to reinvent the wheel and develop
own
>>>>>>>>> 
>>>>>>>>> authentication/authorization mechanism. The existing
works well
>>>>>>>>> except
>>>>>>>>> 2
>>>>>>>>> things:
>>>>>>>>> - add extra info to _users and being able to see it in
the userCtx
>>>>>>>>> - add signup function to Session API
>>>>>>>>> 
>>>>>>>>> Why sending a user doc isn't enough?
>>>>>>>> 
>>>>>>>> 
>>>>>>>> 
>>>>>>>> I'm sure these things make Couchdb better and easy to use
for
>>>>>>>>> 
>>>>>>>>> commercial
>>>>>>>>> development.
>>>>>>>>> 
>>>>>>>>> Alex
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> On 29/08/12 23:05, Benoit Chesneau wrote:
>>>>>>>>> 
>>>>>>>>>> On Wed, Aug 29, 2012 at 9:54 PM, Gabriel Mancini
>>>>>>>>>> <gabriel.mancini@gmail.com <mailto:gabriel.mancini@gmail.com>>
>>>>>>>>>> wrote:
>>>>>>>>>> 
>>>>>>>>>> 
>>>>>>>>>>> but can be nice have sume enable/disable behaviour
for user. We
>>>>>>>>>>> could
>>>>>>>>>>> have anything once partial updates and fetch
 will be here. But
>>>>>>>>>>> it's
>>>>>>>>>>> not the
>>>>>>>>>>> case right now :)
>>>>>>>>>>> 
>>>>>>>>>> - benoit
>>>>>>>>>> 
>>>>>>>>>>> On Wed, Aug 29, 2012 at 4:39 PM, Benoit Chesneau
>>>>>>>>>>> <bchesneau@gmail.com <mailto:bchesneau@gmail.com>>wrote:
>>>>>>>>>>> 
>>>>>>>>>>> On Wed, Aug 29, 2012 at 9:24 PM, Dave Cottlehuber<dave@muse.net.nz
>>>>>>>>>>> <mailto:dave@muse.net.nz>>
>>>>>>>>>>>> 
>>>>>>>>>>>> wrote:
>>>>>>>>>>>> 
>>>>>>>>>>>>> On 29 August 2012 08:21, Benoit Chesneau<bchesneau@gmail.com
>>>>>>>>>>>>> <mailto:bchesneau@gmail.com>>
>>>>>>>>>>>>> 
>>>>>>>>>>>>> wrote:
>>>>>>>>>>>>> 
>>>>>>>>>>>>>> On Tuesday, August 28, 2012, Aliaksandr
Barysiuk wrote:
>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> Hello,
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>>> We store some extra information
in _users db and now we are
>>>>>>>>>>>>>>> looking a
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> way
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>>> to populate session.userCtx with these
extra values. Is it
>>>>>>>>>>>>> possible
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>>> at
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> all?
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>>> Thank you
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>>> Alex
>>>>>>>>>>>>>>> 
>>>>>>>>>>>>>>> user db isn't done for that.
this db exists to authenticate
>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> users and
>>>>>>>>>>>>>> 
>>>>>>>>>>>>> only
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>>> that. You should better save the profiles
in another db. Also
>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> there is
>>>>>>>>>>>>>> 
>>>>>>>>>>>>> no
>>>>>>>>>>>> 
>>>>>>>>>>>> 
>>>>>>>>>>>>> such things like session in couchdb by
itself.
>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> 
>>>>>>>>>>>>>> benoît
>>>>>>>>>>>>>> 
>>>>>>>>>>>>> Any good reasons why we couldn't / shouldn't
support something
>>>>>>>>>>>>> that
>>>>>>>>>>>>> eases this pain? Putting in a second
db simply to store some
>>>>>>>>>>>>> basic
>>>>>>>>>>>>> profile info seems daft. And as others
have found, you can store
>>>>>>>>>>>>> anything you like in roles.
>>>>>>>>>>>>> 
>>>>>>>>>>>> Well I think that storing anything in a role
is a bug. We
>>>>>>>>>>>> shouldn't
>>>>>>>>>>>> allow that and it should be fixed. Only a
list of strings is
>>>>>>>>>>>> expected
>>>>>>>>>>>> in the roles member. We should enforce that.
>>>>>>>>>>>> 
>>>>>>>>>>>> For security reasons I don't think it's good
to have more data in
>>>>>>>>>>>> the
>>>>>>>>>>>> doc other than the login, roles, password
and possibly anything
>>>>>>>>>>>> about
>>>>>>>>>>>> permissions ( some would argue that the users
db shouldn't exist
>>>>>>>>>>>> at
>>>>>>>>>>>> all). You don't protect the same the access
to a user doc or a a
>>>>>>>>>>>> profile doc. And the way it is designed right
now  prevent any
>>>>>>>>>>>> use
>>>>>>>>>>>> of
>>>>>>>>>>>> this profile by others. Only the user or
an admin can have access
>>>>>>>>>>>> to
>>>>>>>>>>>> the doc. Which is good imo.
>>>>>>>>>>>> 
>>>>>>>>>>>> - benoit
>>>>>>>>>>>> 
>>>>>>>>>>>> --
>>>>>>>>>>> 
>>>>>>>>>>> Gabriel Mancini de Campos
>>>>>>>>>>> Arquiteto de Soluções
>>>>>>>>>>> 
>>>>>>>>>>> +55 (11) 9449-1706
>>>>>>>>>>> gabriel.mancini@gmail.com <mailto:gabriel.mancini@gmail.com>
>>>>>>>>>>> 
>>>>>>>>>>> São Paulo - SP - Brasil
>>>>>>>>>>> 
>>>>>>>>>> 
>>>>>> 
>>>>> 
>>>> 
>>>> 
>>>> --
>>>> Gabriel Mancini de Campos
>>>> Arquiteto de Soluções
>>>> 
>>>> +55 (11) 9449-1706
>>>> gabriel.mancini@gmail.com <mailto:gabriel.mancini@gmail.com>
>>>> 
>>>> São Paulo - SP - Brasil
>>> 
>>> 
>> 


Mime
View raw message