users@jersey.java.net

[Jersey] Re: stateless REST

From: Robert Gacki <robert.gacki_at_contenttrace.org>
Date: Sat, 11 Jun 2016 00:59:41 +0200

Answers below.

Am Freitag, den 10.06.2016, 16:13 -0600 schrieb Trenton D. Adams:
> Hi Jason,
>
> Thanks for the response.
>
> So, you're using REST, but not maintaining state server side, yes?
>  i.e. no sessions.  So, when you can't commit data to a database,
> until all of the data is obtained, in your experience, how do you
> maintain that data until the process is complete?

We are not maintaining the state of the client on the server, like in a
terminal application. We use Java, but we could write most of the stuff
in PHP or pipe to shell scripts (SCNR ;).

So yes, usually the server only processes things that makes sense to
him (never trust the client, which is fundamental in HTTP and
implicitely in REST). For instance, we have a checkout. The client asks
for the checkout and the server will respond with its representation.
If the client sends data that can complete the checkout, then we
complete it. Otherwise, we tell the client that it was stupid and
should reconsider. ;)

> Do you use HTML5 session storage?  

Yes, but for little things. The client saves previous searches so the
user can access those.

> Do you not use JAX-RS for user responses (e.g. Jersey templates)?  

What is "Jersey templates"? We wrote something similar to JSR 371 and
use Freemarker to render HTML views.

And what are user responses? We have a service architecture
("microservices"). A user response can be comprised of serveral
responses from different services.

> I'm curious.  Partly because it seems like a lot of work to not have
> the server maintain state.  When the server maintains state, I can
> set all variables from page 1 to page n on the server side, and use
> only a reference to that object, and continue doing so until I'm
> complete, and I commit.  This is what makes EJB look so good to me.

Maybe that helps: This entire conversation is based on a stateless
protocol (mail). "Mail" (lots of RFCs) specifies how a mail client
interacts with a mail server and a mail server with another mail
server. Somehow you send me your state (your text) and I received it.
But time goes on, you may have found a flaw in your logic while I type
this response. But I don't have knowledge about that flaw. But my mail
client saved your text anyway, so I can re-open it, even when you send
me another mail, while I type this response. Anyhow, I type this
response and it has objections about your logic and press "send". So I
did not accept your logic and you get my response anyway. Now you can
accept my state or just ignore it (because you already found the
flaw)......

This could go on and on. But it's REST. We use URNs to address ourselfs
(email addresses). And we send state and transition into another
(hopefully, better state).

Robert

> On Fri, Jun 10, 2016 at 3:55 PM, Jason Lee <jason_at_steeplesoft.com>
> wrote:
> > I don't think that's a true statement ("The use of REST precludes
> > maintaining state server side"). Of course the server-side
> > maintains state: it has to have SOMETHING to transfer to the
> > client, right? :) What the server-side should NOT do, in a pure
> > RESTful architecture, is store some sort of client-specific state
> > such as we might be used to used to doing with session-scoped beans
> > in, say, a JSF application.  In other words, it shouldn't have the
> > notion of a client's *session* (including things like a client's
> > current step in a wizard), but it certainly needs *state* ("no
> > client context being stored on the server between requests"). 
> > To address the "where to put the application logic" question from
> > earlier in the thread, the server can (and should! :) react when
> > the state changes in certain ways to perform system-specific
> > processing. For example, the client updates the state of an Order
> > entity to, say, SUBMITTED, the server notices this change (through
> > means that aren't relevant here), and begins some sort of
> > processing in the background.
> > Moving all of the app's logic to the client is probably not a wise
> > move for a number of reasons, though there will almost certainly be
> > SOME logic there, such as validation, page flow coordination, etc.
> > Having said all of that, we've tended to be pretty pragmatic on the
> > systems I've worked on, adhering as strictly as possible to
> > Fielding's notions, but deviating, grudgingly, when it made sense
> > to do so (i.e., a strictly RESTful implementation would make the
> > APIs usage slow, chatty, cumbersome, etc). For example, for
> > authentication, one system had an endpoint to authenticate a user
> > that would return an auth token (it might actually have been called
> > a session ID :) that subsequent calls would then send in lieu of
> > credentials. This was done, in part, so that if the token were
> > intercepted and used, all that would be exposed to an attacker is
> > that session and not the user's credentials (SSL helps, of
> > course). 
> > Oh, and if I'm ever asked to implement HATEOAS again, I'll be
> > tempted to rage quit. That was a terrible experience. :)
> > At any rate, history (and personal experience ;) shows you CAN
> > build complex applications with JAX-RS, but you do have to change
> > how you think about designing the system. 
> >
> > On 6/10/16 2:38 PM, Trenton D. Adams wrote:
> > > Your descriptions of what REST is what I would like it to be, but
> > > it isn't that, accrording to most everything I've read on it.  If
> > > you're storing state, you're not using REST.  The use of REST
> > > precludes maintaing state server side.  I'm tempted to use JAX-RS
> > > regardless, and just not use the statelessness part.
> > >
> > > https://en.wikipedia.org/wiki/Representational_state_transfer#Sta
> > > teless
> > > Stateless
> > > See also: Stateless protocol
> > > The client–server communication is further constrained by no
> > > client context being stored on the server between requests. Each
> > > request from any client contains all the information necessary to
> > > service the request, and session state is held in the client. The
> > > session state can be transferred by the server to another service
> > > such as a database to maintain a persistent state for a period
> > > and allow authentication. The client begins sending requests when
> > > it is ready to make the transition to a new state. While one or
> > > more requests are outstanding, the client is considered to be in
> > > transition. The representation of each application state contains
> > > links that may be used the next time the client chooses to
> > > initiate a new state-transition.[12]
> > >
> > >
> > > On Fri, Jun 10, 2016 at 1:04 PM, Robert Gacki <robert.gacki_at_conte
> > > nttrace.org> wrote:
> > > > Hi Trenton,
> > > >
> > > > I think, you misunderstand what REST is. REST is about how a
> > > > server and
> > > > a client interacts to transfer state. The transport itself is
> > > > stateless
> > > > ("dumb pipes"). It is an architectural style. The notion of
> > > > state is
> > > > something the client and server have to implement themself.
> > > > What REST
> > > > describes is nothing else what HTTP specifies, but is not
> > > > limited to
> > > > HTTP (think of mail). In fact, Roy Fielding just took HTTP as
> > > > an
> > > > example for REST in his dissertation.
> > > >
> > > > OSI Layer
> > > > ----------------------------------------------------
> > > > | Your server and/or client application      | L   |
> > > > ---------------------------------------------- a   |
> > > > | REST as architectural style                | y 7 |
> > > > ---------------------------------------------- e   |
> > > > | HTTP|MAIL|<other req / res based protocols>| r   |
> > > > ----------------------------------------------------
> > > >   Layer 6 and below
> > > >
> > > > JAX RS is only about modelling REST principles and Jersey
> > > > provides the
> > > > implementation. That does not include state management. So you
> > > > should
> > > > see JAX RS as the protocol toolset ("dumb pipes") to build
> > > > applications. What you call stateful is application state and
> > > > that is
> > > > always application / implementation specific. For example, the
> > > > notion
> > > > of a session is purely application specific.
> > > >
> > > > So Jersey, at its core, does not provide you with components to
> > > > build
> > > > things like a wizard. But it provides you the ability to build
> > > > or use
> > > > existing components, like JSR 371 and its implementations.
> > > >
> > > > To your wizard example: You can solve wizards with different
> > > > methods.
> > > > You described one solution, where the server does not store
> > > > results of
> > > > previous wizard steps. Each solution has advantages and
> > > > disadvantages.
> > > > But all solutions have something in common: they are
> > > > application
> > > > specific. If you use HTTP as a communication protocol, you can
> > > > use REST
> > > > principles to have a well-defined architectural style. But you
> > > > still
> > > > need to figure out how to maintain state between a request and
> > > > response.
> > > >
> > > > An important part of REST is Hypermedia. For HTTP, this is just
> > > > payload. For an application built by REST principles, it is the
> > > > state
> > > > that is transfered. And that state should include all
> > > > information
> > > > necessary for server or client to support continous
> > > > interactions. If
> > > > you use your browser and surf around, clicking on links or
> > > > signing in
> > > > into your Google account, you are doing exactly what REST
> > > > proposes for
> > > > all other media types - not just HTML.
> > > >
> > > > So your wizard solution is based on HTML (a representation) and
> > > > is
> > > > constraint by the HTML specification and how browsers behave.
> > > > Thus, you
> > > > may lose form field values when you use the browser's back
> > > > button. But
> > > > you can do three things to counter that without storing state
> > > > server-
> > > > side (outside request-response bounds): 1) You can provide a
> > > > back
> > > > button inside the form that just submits to the server - the
> > > > server
> > > > renders the previous page with the values from the hidden
> > > > fields. 2) In
> > > > modern browsers, with Javascript enabled, you can store the
> > > > state
> > > > client side and restore it when the user goes back. 3) SPAs
> > > > mostly
> > > > avoid the problem by avoiding round-trips to the server, with
> > > > the
> > > > disadvantage of cloning lots of server-side logic for the
> > > > client
> > > > (validation, business logic).
> > > >
> > > > Robert
> > > >  
> > > >
> > > > Am Freitag, den 10.06.2016, 12:11 -0600 schrieb Trenton D.
> > > > Adams:
> > > > > Thanks for the responses guys.  I'm thinking more along the
> > > > lines of
> > > > > JAX-RS implementations, such as Jersey, providing a template
> > > > based
> > > > > approach, where the result can be served by JSP. 
> > > > Technically, that's
> > > > > almost useless when doing it statelessly, unless your service
> > > > is
> > > > > intended to return HTML to a client program, or you intend to
> > > > violate
> > > > > the statelessness of REST.  Cause like you say, writing BBAs,
> > > > is
> > > > > actually quite tough without some nice statefulness.
> > > > >
> > > > > JSR 371 will be coming out too.  Are they still expecting
> > > > stateless? 
> > > > > Cause it suddenly becomes less useful.  Maintaining all the
> > > > state in
> > > > > the browser's store is a lot of extra work, and just a plain
> > > > weird
> > > > > way of dealing with it.
> > > > >
> > > > > On Fri, Jun 10, 2016 at 10:07 AM, <lenny_at_flowlogix.com>
> > > > wrote:
> > > > > > Trenton, I feel your pain.
> > > > > > In all honesty, the technology architecture didn’t change
> > > > all that
> > > > > > much in the last 10 years.
> > > > > > Don’t get caught up in the Rest/Microservices hype too
> > > > much.
> > > > > > For BBAs (boring business apps) which I believe that you
> > > > are
> > > > > > dealing with (and I also like developing),
> > > > > > it makes very little sense migrating to Rest-based
> > > > architecture.
> > > > > > Concentrate on cleaning, refactoring, modularizing your
> > > > code, using
> > > > > > the latest iterations of Java EE specs that you are already
> > > > using.
> > > > > >
> > > > > > So, why all the hype?
> > > > > > Current Rest “state-of-the-art” today is actually in it’s
> > > > very
> > > > > > infancy. It makes sense for some applications to do it
> > > > though, but
> > > > > > not for BBAs.
> > > > > > What apps does the Stateless Rest APIs make sense for?
> > > > > > - Multiple clients written separately by different teams
> > > > (i.e.
> > > > > > native iOS, native Android, native JavaScript)
> > > > > > - APIs that other companies use in different (non-Java)
> > > > languages
> > > > > > - There are some other use cases also, but the above are
> > > > the major
> > > > > > ones
> > > > > >
> > > > > > What’s the negative impact of Rest services today, as it
> > > > relates to
> > > > > > Java client and server development?
> > > > > > - Massive violation of DRY principle, as it’s really meant
> > > > for
> > > > > > client and server to be written in different languages,
> > > > > > i.e. client in JavaScript/iOS/Android and server in Java
> > > > > > - Data models are duplicated both in Java (on the server)
> > > > and on
> > > > > > the client (JS/iOS/Android)
> > > > > > - Validation is duplicated on the client and server
> > > > > > - Some business logic may need to be duplicated in the
> > > > client and
> > > > > > server
> > > > > >
> > > > > > The above negatives are something that are not solved by
> > > > the
> > > > > > industry quite yet, and for BBAs, (IMHO) is not worth the
> > > > effort.
> > > > > >
> > > > > > As far as answering your specific question, in “pure” Rest
> > > > API, All
> > > > > > state is indeed handled by the client, including
> > > > authentication.
> > > > > > Authentication is usually handled by “bearer token”
> > > > paradigm, is
> > > > > > sent with every request to the server, and thus re-
> > > > authenticated by
> > > > > > the server every time.
> > > > > > Yes, it sounds (and is) a bit less efficient on per-request
> > > > basis
> > > > > > (throughput per instance), but does (horizontally) scale
> > > > better to
> > > > > > millions of users.
> > > > > >
> > > > > >
> > > > > > > On Jun 9, 2016, at 4:40 PM, Trenton D. Adams <trenton.d.a
> > > > dams_at_gma
> > > > > > > il.com> wrote:
> > > > > > >
> > > > > > > One thing I didn't mention, is that I'm considering
> > > > updating one
> > > > > > > of our enterprise apps to more modern technologies, where
> > > > it
> > > > > > > actually saves effort.  It is currently based on RMI,
> > > > with a
> > > > > > > custom web front-end/mvc framework based on a the command
> > > > > > > pattern.  So, I'm trying to determine whether we should
> > > > do EJB or
> > > > > > > JAX-RS for the back-end, and possibly JAX-RS for the
> > > > front end. 
> > > > > > > One of the issues is that EJB is almost a drop in
> > > > replacement for
> > > > > > > RMI.  It would require significant more effort to switch
> > > > to JAX-
> > > > > > > RS.
> > > > > > >
> > > > > > > When the back end is stateless, it's simply pushing the
> > > > > > > complexity to the client.  It is now the client that must
> > > > do all
> > > > > > > the work to know what it needs to send; i.e. it keeps
> > > > it's own
> > > > > > > state.  For the back end, that's HIGHLY scale-able
> > > > technology
> > > > > > > wise, but has other drawbacks.  With a stateful back-end,
> > > > you
> > > > > > > just set the firstname, lastname, birthdate, etc, and
> > > > pass around
> > > > > > > a reference to the back-end object, such as with EJB. 
> > > > Neither
> > > > > > > the front end nor the back end have to maintain much
> > > > state,
> > > > > > > programmatically speaking; it's the service that does
> > > > that (EJB
> > > > > > > for example).  This saves developer time, does it not?
> > > > > > >
> > > > > > > It seems that using the html5 stuff would be a pain in
> > > > the butt. 
> > > > > > > Mainly because html5's storage system can't store a
> > > > javascript
> > > > > > > object even, so you can't even abstract your data
> > > > storage.  Plus,
> > > > > > > we'd end up not supporting people with older
> > > > machines/browsers. 
> > > > > > > And then, if you have a series of web pages that a person
> > > > is
> > > > > > > going through, you'd have to write code to grab all of
> > > > that, and
> > > > > > > pass it to the server.
> > > > > > >
> > > > > > > So, should I not use JAX-RS if I'm wanting to maintain
> > > > state?  I
> > > > > > > mean it kind of goes against the "ST" in REST.  Nothing
> > > > actually
> > > > > > > prevents you from maintaining state though.
> > > > > > >
> > > > > > > I've read some articles where people say you shouldn't
> > > > even be
> > > > > > > maintaining state of authentication.  That I don't agree
> > > > with,
> > > > > > > because some services don't even have access to the
> > > > user's
> > > > > > > credentials.  So at some point, someone is going to have
> > > > to
> > > > > > > maintain state.  So, you could either not use JAX-RS, or
> > > > use it
> > > > > > > and maintain state while doing so, but essentially
> > > > violate it's
> > > > > > > principles.
> > > > > > >
> > > > > > > Also, doesn't statelessness become very complex when you
> > > > have
> > > > > > > larger enterprise applications?
> > > > > > >
> > > > > > > I see a lot of benefits of JAX-RS, and a lot of
> > > > drawbacks.  Am I
> > > > > > > missing something?
> > > > > > >
> > > > > > > On Wed, Jun 8, 2016 at 5:46 PM, cowwoc <cowwoc_at_bbs.darkte
> > > > ch.org>
> > > > > > > wrote:
> > > > > > > > Define "application logic".
> > > > > > > >
> > > > > > > > In the case you mentioned below (storing the user's
> > > > last name
> > > > > > > > somewhere) I would favor using localStorage and
> > > > sessionStore to
> > > > > > > > store this information: http://www.w3schools.com/html/h
> > > > tml5_web
> > > > > > > > storage.asp
> > > > > > > > The client would read the information from the local
> > > > store and
> > > > > > > > use it to make AJAX calls.
> > > > > > > >
> > > > > > > > I misspoke earlier when I talked about Cookies. These
> > > > are
> > > > > > > > typically used to reference to a server-side state that
> > > > is
> > > > > > > > present across all calls. If some REST calls need one
> > > > piece of
> > > > > > > > information and others need another, I would pull them
> > > > from the
> > > > > > > > local/sessionStore and pass them to the AJAX calls as
> > > > needed.
> > > > > > > >
> > > > > > > > Gili
> > > > > > > >
> > > > > > > >
> > > > > > > > On 2016-06-08 7:40 PM, Trenton D. Adams wrote:
> > > > > > > > > So are you saying push all the application logic to
> > > > the
> > > > > > > > > browser, using javascript?  Are cookies really
> > > > intended to
> > > > > > > > > store a whole bunch of user data?
> > > > > > > > >
> > > > > > > > > I know with HTML5, you can use
> > > > > > > > > sessionStorage.setItem("lastname", "last name"). 
> > > > But, I
> > > > > > > > > don't think moving most application logic into a
> > > > browser is
> > > > > > > > > very maintainable, maybe I'm wrong though.
> > > > > > > > >
> > > > > > > > > On Wed, Jun 8, 2016 at 5:29 PM, cowwoc <cowwoc_at_bbs.da
> > > > rktech.o
> > > > > > > > > rg> wrote:
> > > > > > > > > > Without commenting on the specifics of Jersey, I
> > > > agree:
> > > > > > > > > > REST is for computers, not humans.
> > > > > > > > > >
> > > > > > > > > > I typically expose REST APIs for computers and use
> > > > cookies
> > > > > > > > > > to maintain browser sessions. The browser can then
> > > > read
> > > > > > > > > > stateful information from the Cookie and serve it
> > > > to
> > > > > > > > > > stateless REST APIs. Not all clients are web-
> > > > browsers, so
> > > > > > > > > > your REST API should be designed around non-
> > > > browsers.
> > > > > > > > > >
> > > > > > > > > > Gili
> > > > > > > > > >
> > > > > > > > > >
> > > > > > > > > > On 2016-06-08 5:58 PM, Trenton D. Adams wrote:
> > > > > > > > > > > Good day,
> > > > > > > > > > >
> > > > > > > > > > > I'm a bit confused.  I actually have two separate
> > > > > > > > > > > questions.  I understand that REST is supposed to
> > > > be done
> > > > > > > > > > > in a stateless way.  For regular web services
> > > > that's
> > > > > > > > > > > easy.  I mean it really shifts a lot of the work
> > > > to the
> > > > > > > > > > > client, where it seems to be more difficult to
> > > > deal with,
> > > > > > > > > > > but as far as the server goes, it's simple.
> > > > > > > > > > >
> > > > > > > > > > > However, how is it even possible to use jersey
> > > > templates
> > > > > > > > > > > without state (sessions), in a reasonable way? 
> > > > The
> > > > > > > > > > > browser isn't going to maintain the state.  It
> > > > seems that
> > > > > > > > > > > one would need to make sure each and every page
> > > > puts
> > > > > > > > > > > hidden inputs from the previous form, in the html
> > > > output,
> > > > > > > > > > > so that it is re-submitted with the new request. 
> > > > That
> > > > > > > > > > > would be a lot of work.  If the user presses the
> > > > back
> > > > > > > > > > > button, all that state vanishes, and the user
> > > > must re-
> > > > > > > > > > > enter any screens they go forward to again.  This
> > > > doesn't
> > > > > > > > > > > make for a very good user experience.
> > > > > > > > > > >
> > > > > > > > > > > Can someone explain to me how the use of JAX-RS
> > > > as an MVC
> > > > > > > > > > > framework is even possible in a reasonable way,
> > > > while
> > > > > > > > > > > being stateless?
> > > > > > > > > > >
> > > > > > > > > > > Then, can someone explain to me how statelessness
> > > > in a
> > > > > > > > > > > back-end REST web service, promotes good code
> > > > design,
> > > > > > > > > > > where user interaction is a necessity?  It seems
> > > > to me
> > > > > > > > > > > that the client would then need to maintain all
> > > > the
> > > > > > > > > > > state, thereby tightly coupling all the data
> > > > points
> > > > > > > > > > > between the different controllers on the client. 
> > > > > > > > > > > Something like EJB allows you to pass around the
> > > > stateful
> > > > > > > > > > > pointer, and you simply add data as you go.
> > > > > > > > > > >
> > > > > > > > > > > After reading this stack exchange post, it's
> > > > sounding
> > > > > > > > > > > like everyone thinks that REST is NOT for users,
> > > > but for
> > > > > > > > > > > services only.
> > > > > > > > > > > http://stackoverflow.com/questions/3105296/if-res
> > > > t-applic
> > > > > > > > > > > ations-are-supposed-to-be-stateless-how-do-you-
> > > > manage-
> > > > > > > > > > > sessions
> > > > > > > > > > >
> > > > > > > > > > > I understand that it's more scalable, as the
> > > > server
> > > > > > > > > > > always knows exactly what you want, because
> > > > you're
> > > > > > > > > > > telling it every time.  But it seems like that
> > > > would come
> > > > > > > > > > > with a lot more boilerplate coding.
> > > > > > > > > > >
> > > > > > > > > > > Thanks.
> > > > > > > > > > >
> > > > > > > > > >
> > > > > > > >
> > > > > >
> > > >
> >  
> > -- 
> > Jason Lee
> > http://cubtracker.com 
> > http://blogs.steeplesoft.com
> > http://twitter.com/jasondlee
> > http://blogs.steeplesoft.com/+
> > http://blogs.steeplesoft.com/in
> >