Session management using json web tokens in microservices - json

I am trying to figure out how I will manage sessions using json web tokens in a microservice architecture.
Looking at the design in this article what I currently have in mind is that the client will send a request that first goes through a firewall. This request will contain an opaque/reference token which the firewall sends to an authorization server. The authorization server responds with a value token containing all the session information for the user. The firewall then passes the request along with the value token to the API, and the value token will then get propagated to all the different microservices required to fulfill the request.
I have 2 questions:
How should updates to the session information in the value token be handled? To elaborate, when the session info in a token gets updated, it needs to be updated in the authorization server. Should each service that changes the token talk to the authorization server?
Should all the microservices use this single token to store their session info? Or would it be better for each service to have a personalized token? If it's the latter, please explain how to adjust the design.

A very(!) significant "fly in the ointment" of this kind of design ... which requires careful advance thought on your part ... is: “precisely what is meant by ‘session’ information.” In this architecture, “everyone is racing with everyone else.” If the session information is updated, you do not and basically cannot(!) know which of the agents knows about that change and which does not. To further complicate things, new requests are arriving asynchronously and will overlap other requests in unpredictable ways.
Therefore, the Authorization Server must be exactly that ... and, no more. It validates (authenticates ...) the opaque token, and supplies a trustworthy description of what the request is authorized to do. But, the information that it harbors basically cannot change. And specifically, it cannot hold “session state” data in the web server sense of that term.
Each microservice provider must maintain its own “tote board” *(my term ... “its own particular subset of what in a web-server would be ‘the session pool’”), and it is desirable but not always feasible that its board would be independent of the others. Almost certainly, it must use a central database (with transactions) to coordinate with other service-providers similarly situated. And still, if the truth is that the content of any of these “totes” is causally related to any other, you now have an out-of-sync issue between them.
Although microservice architecture has a certain academic appeal, IMHO designs must be carefully studied to be certain that they are, in fact, compatible with this approach.

Related

What is the RESTful way to return a JSON + binary file in an API

I have to implement a REST endpoint that receives start and end dates (among other arguments). It does some computations to generate a result that is a kind of forecast according to the server state at invocation epoch and the input data (imagine a weather forecast for next few days).
Since the endpoint does not alter the system state, I plan to use GET method and return a JSON.
The issue is that the output includes also an image file (a plot). So my idea is to create a unique id for the file and include an URI in the JSON response to be consumed later (I think this is the way suggested by HATEOAS principle).
My question is, since this image file is a resource that is valid only as part of the response to a single invocation to the original endpoint, I would need a way to delete it once it was consumed.
Would it be RESTful to deleting it after serving it via a GET?
or expose it only via a DELETE?
or not delete it on consumption and keep it for some time? (purge should be performed anyway since I can't ensure the client consumes the file).
I would appreciate your ideas.
Would it be RESTful to deleting it after serving it via a GET?
Yes.
or expose it only via a DELETE?
Yes.
or not delete it on consumption and keep it for some time?
Yes.
The last of these options (caching) is a decent fit for REST in HTTP, since we have meta-data that we can use to communicate to general purpose components that a given representation has a finite lifetime.
So this reference of the report (which includes the link to the plot) could be accompanied by an Expires header that informs the client that the representation of the report has an expected shelf life.
You might, therefore, plan to garbage collect the image resource after 10 minutes, and if the client hasn't fetched it before then - poof, gone.
The reason that you might want to keep the image around after you send the response to the GET: the network is unreliable, and the GET message may never reach its destination. Having things in cache saves you the compute of trying to recalculate the image.
If you want confirmation that the client did receive the data, then you must introduce another message to the protocol, for the client to inform you that the image has been downloaded successfully.
It's reasonable to combine these strategies: schedule yourself to evict the image from the cache in some fixed amount of time, but also evict the image immediately if the consumer acknowledges receipt.
But REST doesn't make any promises about liveness - you could send a response with a link to the image, but 404 Not Found every attempt to GET it, and that's fine (not useful, of course, but fine). REST doesn't promise that resources have stable representations, or that the resource is somehow eternal.
REST gives us standards for how we request things, and how responses should be interpreted, but we get a lot of freedom in choosing which response is appropriate for any given request.
You could offer a download link in the JSON response to that binary resource that also contains the parameters that are required to generate that resource. Then you can decide yourself when to clean that file up (managing disk space) or cache it - and you can always regenerate it because you still have the parameters. I assume here that the generation doesn't take significant time.
It's a tricky one. Typically GET requests should be repeatable as an import HTTP feature, in case the original failed. Some people might rely on it.
It could also be construed as a 'non-safe' operation, GET resulting in what is effectively a DELETE.
I would be inclined to expire the image after X seconds/minutes instead, perhaps also supporting DELETE at that endpoint if the client got the result and wants to clean up early.

How do NServiceBus endpoint names work with pub/sub

I have been fighting the same very simple problem with NServiceBus all day today. The problem is that there is lots of documentation on how to change the configuration, but almost nothing that helps me to know what configuration I need.
There are sample applications, and they work, but there is nothing explaining how they work, what limitations they have, or how to do something just a little bit different than the sample. The sample applications also present a "Hello world" type simplicity, and in any real application you need something different from the sample application, but again there is no help on how to make these changes, or the implications of configuration choices.
From all the things that are very difficult to guess from the documentation, it is the relationship between the endpoint name, the UnicastBusConfig mappings, and pub/sub persistence that is causing the most frustration right now.
Is the endpoint name the name of the MSMQ queue? Does that mean that every application has only one input queue for all message types? Does adding a mapping in UnicastBusConfig cause a subscription message to be sent to the publisher, or does it add a subscription record in subscription DB? Why can't you add the same message type more than once to UnicastBusConfig? Why can't I just subscribe to messages of a certain type without having to know which server they come from?
For someone that understands NServiceBus this probably seems so simple that it wasn't worth documenting, but for someone coming to this for the first time, it's the very simple stuff that's the most difficult to infer from the morass of low level detail.
Is the endpoint name the name of the MSMQ queue?
Yes.
Does that mean that every application has only one input queue for all message types?
Yes. Each endpoint has a single queue associated with it, so all messages for that endpoint go through the same queue.
Does adding a mapping in UnicastBusConfig cause a subscription message to be sent to the publisher, or does it add a subscription record in subscription DB?
Neither really. The UnicastBusConfig section is for setting up the relationship between types (or assemblies) and endpoints. So it doesn't actually cause a subscription to be set up (per se), but it tells the framework where the messages will be coming from (and therefore how to subscribe to them).
The actual subscription gets created when the system starts up and NSB finds a handler for a particular type of message that matches a section in the UnicastBusConfig (assuming auto-subscribing is turned on).
This also works for sending Commands--the config section lets the framework know to which endpoint to Send() a Command.
Why can't you add the same message type more than once to UnicastBusConfig?
Because a Command can have only one (logical) endpoint that handles it, and an Event can have only one (logical) endpoint that publishes it.
Why can't I just subscribe to messages of a certain type without having to know which server they come from?
This question is a bit more difficult to answer definitively, as it gets into the philosophy of having a central broker (hub and spoke) vs. bus-style architecture.
But in a nutshell, something, somewhere needs to know how to find the publisher in order to subscribe to it. Because NServiceBus does not have a central broker or routing table, it is left to the client to be configured with knowledge of the endpoints it consumes.
You might want to check out the NServiceBus documentation at http://docs.particular.net/nservicebus/, it's quite comprehensive and should provide answers to most of your questions.

Is an OAuth 2.0 token forever unique to the provider?

When an OAuth 2.0 provider issues a token, is that token value forever unique to the provider? Or is it possible that sometime in the future, presumably after the token expires, another token, potentially for a different user, could be issued with the same value? In searching I found much information about tokens expiring, but no details about if that token value could potentially be re-used in the future.
There's nothing in the core OAuth 2 spec that guarantees this. It is implementation specific if there is a chance of collision or not. You should find out from your OAuth AS provider what the likelihood is. But agreed with Artem - this sounds odd if you are trying to uniquely identify users based on what is suppose to be just an API (access) token.
If you use something like UUID - it's time dependent and unique - so you should NOT make them reusable. Taking in account that you'll generate tokens in different instants of time - they all will be different.

Stateless vs Stateful

I'm interested in articles which have some concrete information about stateless and stateful design in programming. I'm interested because I want to learn more about it, but I really can't find any good articles about it. I've read dozens of articles on the web which vaguely discuss the subject, or they're talking about web servers and sessions - which are also 'bout stateful vs stateless, but I'm interested in stateless vs stateful design of attributes in coding. Example: I've heard that BL-classes are stateless by design, entity classes (or at least that's what I call them - like Person(id, name, ..)) are stateful, etc.
I think it's important to know because I believe if I can understand it, I can write better code (e.g. granularity in mind).
Anyways, really short, here's what I know 'bout stateful vs stateless:
Stateful (like WinForms): Stores the data for further use, but limits the scalability of an application, because it's limited by CPU or memory limits
Stateless (Like ASP.NET - although ASP tries to be stateful with ViewStates):
After actions are completed, the data gets transferred, and the instance gets handed back to the thread pool (Amorphous).
As you can see, it's pretty vague and limited information (and quite focussed on server interaction), so I'd be really grateful if you could provide me with some more tasty bits of information :)
Stateless means there is no memory of the past. Every transaction is performed as if it were being done for the very first time.
Stateful means that there is memory of the past. Previous transactions are remembered and may affect the current transaction.
Stateless:
// The state is derived by what is passed into the function
function int addOne(int number)
{
return number + 1;
}
Stateful:
// The state is maintained by the function
private int _number = 0; //initially zero
function int addOne()
{
_number++;
return _number;
}
Refer from: https://softwareengineering.stackexchange.com/questions/101337/whats-the-difference-between-stateful-and-stateless
A stateful app is one that stores information about what has happened or changed since it started running. Any public info about what "mode" it is in, or how many records is has processed, or whatever, makes it stateful.
Stateless apps don't expose any of that information. They give the same response to the same request, function or method call, every time. HTTP is stateless in its raw form - if you do a GET to a particular URL, you get (theoretically) the same response every time. The exception of course is when we start adding statefulness on top, e.g. with ASP.NET web apps :) But if you think of a static website with only HTML files and images, you'll know what I mean.
I suggest that you start from a question in StackOverflow that discusses the advantages of stateless programming. This is more in the context of functional programming, but what you will read also applies in other programming paradigms.
Stateless programming is related to the mathematical notion of a function, which when called with the same arguments, always return the same results. This is a key concept of the functional programming paradigm and I expect that you will be able to find many relevant articles in that area.
Another area that you could research in order to gain more understanding is RESTful web services. These are by design "stateless", in contrast to other web technologies that try to somehow keep state. (In fact what you say that ASP.NET is stateless isn't correct - ASP.NET tries hard to keep state using ViewState and are definitely to be characterized as stateful. ASP.NET MVC on the other hand is a stateless technology). There are many places that discuss "statelessness" of RESTful web services (like this blog spot), but you could again start from an SO question.
The adjective Stateful or Stateless refers only to the state of the conversation, it is not in connection with the concept of function which provides the same output for the same input. If so any dynamic web application (with a database behind it) would be a stateful service, which is obviously false.
With this in mind if I entrust the task to keep conversational state in the underlying technology (such as a coockie or http session) I'm implementing a stateful service, but if all the necessary information (the context) are passed as parameters I'm implementing a stateless service.
It should be noted that even if the passed parameter is an "identifier" of the conversational state (e.g. a ticket or a sessionId) we are still operating under a stateless service, because the conversation is stateless (the ticket is continually passed between client and server), and are the two endpoints to be, so to speak, "stateful".
Money transfered online form one account to another account is stateful, because the receving account has information about the sender.
Handing over cash from a person to another person, this transaction is statless, because after cash is recived the identity of the giver is not there with the cash.
Just to add on others' contributions....Another way is look at it from a web server and concurrency's point of view...
HTTP is stateless in nature for a reason...In the case of a web server, being stateful means that it would have to remember a user's 'state' for their last connection, and /or keep an open connection to a requester. That would be very expensive and 'stressful' in an application with thousands of concurrent connections...
Being stateless in this case has obvious efficient usage of resources...i.e support a connection in in a single instance of request and response...No overhead of keeping connections open and/or remember anything from the last request...
We make Webapps statefull by overriding HTTP stateless behaviour by using session objects.When we use session objets state is carried but we still use HTTP only.
I had the same doubt about stateful v/s stateless class design and did some research. Just completed and my findings has been posted in my blog
Entity classes needs to be stateful
The helper / worker classes should not be stateful.

Using messaging to do writes as well as reads

I come from a web background where I only have to deal with HTTP so please excuse my ignorance.
I have an app which where clients listen for changes in a message queue which uses stomp. Previously the client only needed to listen to the relevant channels for messages telling them about changes on the server and update themselves accordingly. Simple stuff.
There is now a requirement for the client to be able to edit data and push those changes back to the server. The data on the server is already exposed via restful resources so my first thought was just to make REST put requests to change the data on the server, but then I started to wonder whether I could find a solution using messaging. I could just open up another channel which the clients could publish changes to and the server could subscribe to that channel and update itself accordingly. Implementing this would obviously be simple but I would love to have some of the potential pitfalls pointed out to me ahead of time.
I am familiar with REST so I want to ask some questions in the context of REST:
Would I map a group of queues to REST/CRUD verbs for each resource i.e. itemPostQueue, itemPutQueue, itemDeleteQueue?
What about GET's how can I request data to read using a queue?
What do I use to replace my status code mechanism to catch problems or do I just fire and forget (gulp) or use error/receipt headers in Stomp somehow?
Any answers and advise will be much appreciated.
Regards,
Chris
While I am not clear on why you must use messaging here, a few thoughts:
You could map to REST on the wire like itemPostQueue, but this would likely feel unnatural to a message-oriented person. If you are using some kind of queue with a guaranteed semantic and deliver-once built in, then go ahead and use that mechanism. For a shopping-cart example, then you could put an AddItem message on the wire, and you trust the infrastructure to deliver it once to the server.
There is no direct GET like concept here in message queuing. You can simulate it with a pair of messages, I send you a request and you send me back a response. This is much like RPC, but even further decoupled. So I send you a PublishCart request and later on, the server sends a CartContents message on a channel that the client is listening to.
Status codes are more complex, and generally fall into two camps. First are the actual queue-library messages - deal with them just as you would any normal system message. Second you may have your own messages you want to put on the wire that signal failure at some place in the chain.
One thing that messaging does do is significantly decouple your app. Unlike HTTP, where you know that something happened, with a queue, you send a letter to somebody. It may get there. The postman might drop it in the snow. The dog might eat it. If you don't get a response in some period of time, you try other means to contact your relatives, or to pull back the analogy, to contact the server. Monitoring of the health of the queue infrastructure and depth of queues and the like take on added importance, as they are the plumbing that you are now depending upon.
Good Luck