Consuming Kafka messages in Microservices - apache-kafka

We have different services based on the domain. All these services communicate via Rest(Sync) and Kafka (async).
However, the writers of this service have used a common library to write the logic to consume the records from Kafka and individual services use this common library as dependency.
I believe this is an anti pattern of Microservices.
This is because all services have dependency on this common library, however the the library is getting the consumer record value as string(stringdeserializer) and then based on the type of the message content it is delegated to respective handler.
The common library is origin of event consumption, and deserialization is happening to string and then using Gson the string is converted to specific event types.
The problem with above approach is that for schema evolution of the services is becoming bottleneck. As individual services are listening for certain events on some topic, but everything is deserialized to string, hence unable to use schema registry for schema evolution.
After many attempts i decided that common library is an evil for microservices as independency is killed.

The problem here almost certainly isn't the common library, because you'd hit the same problem without it of producers and consumers needing to agree on a schema. A schema registry makes it potentially easier to reach that agreement, but it doesn't really solve the problem (there are scenarios where the schema registry won't help you).
There are two deeper problems:
First (and this is almost certainly the bigger of the two), it sounds like you're using the same types as wire types (for interservice communication) and as internal model types. This is what actually leads to the coupling. By separating the wire types from the model types, you do incur the overhead of translation between them, but, "you don't get freedom for free" (Peart, 1976). What you gain is only having to agree on the wire type which will change a lot less often than the internal model types as their respective services evolve.
Second, approaches to serialization that try to do things by magic based on implementation details of what you're serializing are intrinsically fragile. This is perhaps less of a problem with a wire type, but defining the wire type in a "schema-first" manner might be useful.

Who says the common library is needed? You could repeat all string parsing logic over and over in any new consumer to that topic.
So, that approach definitely isn't any better.
Your thoughts aren't unique to Kafka, or microservices, either. For example, a REST API uses Openapi and publishes a schema and client dependency. Any HTTP "consumer" needs to depend on that API and client, plus it's pinned at a specific version at runtime. If the API "producer" changes the server "events/schema", your "consumer" will fail.
The Schema Registry also makes a shared dependency, plus the overhead of maintaining an external service separate from the broker that must have higher availability than the brokers themselves, otherwise your clients will completely drop events. Also, Schema Registry supports custom types, so evolution can still happen, even for strings, albeit with much custom code.
If you want to store multiple types in one topic to use with Schema Registry, you'd use Subject naming strategies. Before that feature existed, though, the only way to do so was to create some switch-case in the consumer and wrap string/bytes data, such as CloudEvents object and annotate it with a type field.
Also worth pointing out - Kafka includes Jackson, so you shouldn't need Gson as an extra dependency

Related

Axon or Kafka to support CQRS/ES

Consider the simple use case in which I want to store product ratings as events in an event store.
I could use two different approaches:
Using Axon: A Rating aggregate is responsible for handling the CreateRatingCommand and sending the RatingCreatedEvent. Sending the event would case the Rating to be stored in the event store. Other event handlers have the possibility to replay the event stream when connecting to the Axon server instance and doing whatever needed with the ratings. In this case, the event handler will be used as a stream processor.
Using Kafka: A KafkaProducer will be used to store a Rating POJO (after proper serialization) in a Kafka topic. Setting the topic's retention time to indefinite would cause no events to get lost in time. Kafka Streams would in this case be used to do the actual rating processing logic.
Some architectural questions appear to me for both approaches:
When using Axon:
Is there any added value to use Axon (or similar solutions) if there is no real state to be maintained or altered within the aggregate? The aggregate just serves as a "dumb" placeholder for the data, but does not provide any state changing logic.
How does Axon handle multiple event handlers of the same event type? Will they all handle the same event (same aggregate id) in parallel, or is the same event only handled once by one of the handlers?
Are events stored in the Axon event store kept until the end of time?
When using Kafka:
Kafka stores events/messages with the same key in the same partition. How does one select the best value for a key in the use case of user-product ratings? UserId, ProductId or a separate topic for both and publish each event in both topics.
Would it be wise to use a separate topic for each user and each product resulting in a massive amount of topics on the cluster? (Approximately <5k products and >10k users).
I don't know if SO is the preferred forum for this kind of questions... I was just wondering what you (would) recommend in this particular use case as the best practise. Looking forward to your feedback and feel free to point out other points of thought I missed in the previous questions.
EDIT#12/11/2020 : I just found a related discussion containing useful information related to my question.
As Jan Galinski already puts it, this hasn't got a fool proof answer to it really. This is worth a broader discussion on for example indeed AxonIQ's Discuss forum. Regardless, there are some questions in here I can definitely give an answer to, so let's get to it:
Axon Question 1 - Axon Framework is as you've noticed used a lot for DDD centric applications. Nothing however forces you to base yourself on that notion at all. You can strip the framework from Event Sourcing specifics, as well as modelling specifics entirely and purely go for the messaging idea of distinct commands, events and queries. It has been a conscious decision to segregate Axon Framework version 3 into these sub-part when version 4 (current) was released actually. Next to that, I think there is great value in not just basing yourself on event messages. Using distinct commands and queries only further decouples your components, making for a far richer and easier to extend application landscape.
Axon Question 2 - This depends on where the #EventHandler annotated methods are located actually. If they're in the same class only one will be invoked. If they're positioned into distinct classes, then both will receive the same event. Furthermore if they're segregated between distinct classes, it is important to note Axon uses an Event Processor as the technical solution to invoking your event handlers. If distinct classes are grouped under the same Event Processor, you can impose a certain ordering which handler is invoked first. Next to this if the event handling should occur in parallel, you will have to configure a so called TrackingEventProcessor (the default in Axon Framework), as it allows configuration of several threads to handle events concurrently. Well, to conclude this section, everything you're asking in question two is an option, neither a necessity. Just a matter of configuration really. Might be worth checking up on this documentation page of Axon Framework on the matter.
Axon Question 3 - As Axon Server serves the purpose of an Event Store, there is no retention period at all. So yes, they're by default kept until the end of time. There is nothing stopping your from dropping the events though, if you feel there's no value in storing the events to for example base all your models on (as you'd do when using Event Sourcing).
It's the Kafka question I'm personally less familiar with (figures as a contributor to Axon Framework I guess). I can give you my two cents on the matter here too though, although I'd recommend a second opinion here:
Kafka Question 1 - From my personal feeling of what such an application would require, I'd assume you'd want to be able to retrieve all data for a given product as efficient as possible. I'd wager it's important that all events are in the same partition to make this process as efficient as possible, is it wouldn't require any merging afterwards. With this in mind, I'd think using the ProductId will make most sense.
Kafka Question 2 - If you are anticipating only 5_000 products and 10_000 users, I'd guess it should be doable to have separate topics for these. Opinion incoming - It is here though were I personally feel that Kafka's intent to provide you direct power to decide on when to use topics over complicates from what you'd actually try to achieve, which business functionality. Giving the power to segregate streams feels more like an after thought from the perspective of application development. As soon as you'd require an enterprise grade/efficient message bus, that's when this option really shines I think, as then you can optimize for bulk.
Hoping all this helps you further #KDW!

CQRS Read Models & REST API

We are implementing a REST API over our CQRS services. We of course don't want to expose any of our domain to users of the REST APIs.
However, a key tenant of CQRS is that the read models generally correspond to a specific view or screen.
With that being the case, it seems logical that the resources in our REST API, will map virtually 1:1 with the read / view models from our queries (where the queries return a DTO containing all the data for the view). Technically this is exposing a part of our domain (the read models - although returned as DTOs). In this case, this seems to be what we want. Any potential downsides to being so closely coupled?
In terms of commands, I have been considering an approach like:
https://www.slideshare.net/fatmuemoo/cqrs-api-v2. There is a slide that indicates that commands are not first class citizens. (See slide 26). By extension, am I correct in assuming that the DTOs returned from my queries will always be the first class citizens, which will then expose the commands that can be executed for that screen?
Thanks
Any potential downsides to being so closely coupled?
You need to be a little bit careful in terms of understanding the direction of your dependencies.
Specifically, if you are trying to integrate with clients that you don't control, then you are going to want to agree upon a contract -- message semantics and schema -- that you cannot change unilaterally.
Which means that the representations are relatively fixed, but you have a lot of freedom about about how you implement the production of that representation. You make a promise to the client that they can get a representation of report 12345, and it will have some convenient layout of the information. But whether that representation is something you produce on demand, or something that you cache, and how you build it is entirely up to you.
At this level, you aren't really coupling your clients to your domain model; you are coupling them to your views/reports, which is to say to your data model. And, in the CQRS world, that coupling is to the read model, not the write model.
In terms of commands, I have been considering an approach like...
I'm going gently suggest that the author, in 2015, didn't have a particularly good understanding of REST by today's standards.
The basic problem here is that the author doesn't recognize that caching is a REST constraint; and the design of our HTTP protocols needs to consider how general purpose components understand cache invalidation.
Normally, for a command (meaning here "a message intended to change the representation of the resource"), you normally want the target-uri of the HTTP request to match the identifier of the primary resource that changes.
POST /foo/123/command
Isn't particularly useful, from the perspective of cache invalidation, if nobody ever sends a GET /foo/123/command request.

How to retrieve data from another bounded context in ddd?

Initially, There is an app runs in Desktop, however, the app will run in web platform in the future.
There are some bounded contexts in the app and some of them needs to retrieve data from another. In this case, I don't know which approach I have to use for this case.
I thought of using mediator pattern that a bound context "A" requests data "X" and then mediator call another bound context, like B" " and gets the correct data "X". Finally, The mediator brings data "X" to BC "A".
This scenario will be change when the app runs in web, then I've thought of using a microservice requests data from another microservice using meaditor pattern too.
Do the both approaches are interest or there is another better solution?
Could anyone help me, please?
Thanks a lot!
If you're retrieving data from other bounded contexts through either DB or API calls, your architecture might potentially fall to death star pattern because it introduces unwanted coupling and knowledge to the client context.
A better approach might be is looking at event-driven mechanisms like webhooks or message queues as a way of emitting data that you want to share to subscribing context(s). This is good because it reduces coupling of your bounded context(s) through data replication across contexts which results to higher bounded contexts independence.
This gives you the feeling of "Who cares if bounded context B is not available ATM, bounded context A and C have data they need inside them and I can resume syncing later since my data update related events are recorded on my queue"
The answer to this question breaks down into two distinct areas:
the logical challenge of communicating between different contexts, where the same data could be used in very different ways. How does one context interpret the meaning of the data?
and the technical challenge of synchronizing data between independent systems. How do we guarantee the correctness of each system's behavior when they both have independent copies of the "same" data?
Logically, a context map is used to define the relationship between any bounded contexts that need to communicate (share data) in any way. The domain models that control the data are only applicable with a single bounded context, so some method for interpreting data from another context is needed. That's where the patterns from Evan's book come in to play: customer/supplier, conformist, published language, open host, anti-corruption layer, or (the cop-out pattern) separate ways.
Using a mediator between services can be though of as an implementation of the anti-corruption layer pattern: the services don't need to speak the same language, because there's an independent buffer between them doing the translation. In a microservice architecture, this could be some kind of integration service between two very different contexts.
From a technical perspective, direct API calls between services in different bounded contexts introduce dependencies between those services, so an event-driven approach like what Allan mentioned is preferred, assuming your application is okay with the implications of that (eventual consistency of the data). Picking a messaging platforms that gives you the guarantees necessary to keep the data in sync is important. Most asynchronous messaging protocols guarantee "at least once" delivery, but ordering of messages and de-duplication of repeats is up to the application.
Sometimes it's simpler to use a synchronous API call, especially if you find yourself doing a lot of request/response type messaging (which can happen if you have services sending command-type messages to each other).
A composite UI is another pattern that allows you to do data integration in the presentation layer, by having each component pull data from the relevant service, then share/combine the data in the UI itself. This can be easier to manage than a tangled web of cross-service API calls in the backend, especially if you use something like an IT/Ops service, NGINX, or MuleSoft's Experience API approach to implement a "backend-for-frontend".
What you need is a ddd pattern for integration. BC "B" is upstream, "A" is downstream. You could go for an OHS PL in upstream, and ACL in downstream. In practice this is a REST API upstream and an adapter downstream. Every time A needs the data from B , the adapter calls the REST API and adapts the info returned to A domain model. This would be sync. If you wanna go for an async integration, B would publish events to MQ with the info, and A would listen for those events and get the info.
I want to add-on a comment about analysis in DDD. Exist e several approaches for sending data to analytic.
1) If you have a big enterprise application and you should collect a lot of statistic from all bounded context better move analytic in separate service and use a message queue for send data there.
2) If you have a simple application separate your Analytic from your App in other context and use an event or command to speak with there.

Defining gRPC RPCs

I'm looking for some suggestions here. The usecase is a networking device (like router) with networking operations performed over gRPC.
Let's say there are "n" model objects, like router, interfaces, routing configuration objects like OSPF etc. Every networking operation, like finally be a CRUD on on or many of the model objects.
Now, when defining this over a gRPC service, there seems to be 2 options:
Define generic gRPC RPCs, like "SET" and "GET". The parameter will be a list of objects and operations. Like SET((router, update), (interface, update)..
Define very specific RPCs. Like "setInterfaceProperty_x", "createOSPFInstance".. And there could be many many such RPCs.
With #2, we are building the application intelligence in the RPCs itself. Every new feature might need new RPCs from this service.
With #1, the RPCs are the means, but the intelligence reside with the application which uses the RPC in a context. The RPC list will be just a very few and doesn't change over time.
What is the preferred approach? Generic RPCs (and keep it very few) or have tens (or more) of operation driven RPCs? I see some opensource projects like P4Runtime take approach #1.
Thanks for your time. I can provide more information if required.
You should use option #2. This puts your interface contract in the proto, rather than in your application. You leave your self many open doors by picking option #2 that would be cumbersome or unsupportable otherwise:
If the API definition of an object doesn't match the internal representation, you need to define a mapping between the two. Suppose you update your internal code to not need InterfaceProperty any more, and it was instead moved to a new field called BetterInterfaceProperties. Option one would force you to keep the old field exposed, while option 2 would allow you to reinterpret the call and do the right thing.
Fine grained access controls are easier with specific methods. All users may be able to set publicProperty, but only admins can set dangerousProperty. By grouping all the fields into a single call (as in #1), your caller has to reinterpret error messages, while option #2 it's more clear why authorization failed.
Smaller return values. Having a method like getSpecificProperty will do much less work than getFullObject. As your data model gets more complex, you will have to include more and more data on return messages. Even if the caller only cares about one thing, they have to wait for all of them. Consider a Database application. The database might have to do several unnecessary queries to fill in fields the client will never read.
There are reason to use #1, but they aren't that valuable until you identify what properties go together and are logically a single RPC. (such as a Get)

Exposing protocol-buffers interfaces

I am building a distributed system which consist of modules/application with interfaces defined by protobuf messages.
Is it a good idea to expose those protobuf messages to a client directly? ... or maybe it's better to prepare a shared library which will be responsible for translation of (let's assume) method based interface to a protobuf based for each module and clients won't be aware about protobuf at all?
It's neither a "good idea" nor a bad one. It depends on whether or not you want to impose protocol buffers onto your consumers. A large part of that decision is, then:
Who are your consumers? Do you mind exposing the protobuf specifics to them?
Will the clients be written in languages which have protobuf support?
My $0.02 is that this is a perfect use case for Protocol Buffers, since they were specifically designed with cross-system, cross-language interchange in mind. The .proto file makes for a concise, language-independent, thorough description of the data format. Of course, there are other similar/competing formats & libraries out there to consider (see: Thrift, Cap'n Proto, etc.) if you decide to head down this path.
If you are planning to define interfaces that take Google Protobuf message classes as arguments than according to this and that section in Google's Protobuf documentation it is not a good idea to expose Protobuf messages to a client directly. In short, with every version of Protobuf the generated code is likely to be not binary compatible with older code. So don't do it!
However, if you are planning to define interfaces that take byte arrays containing serialized Protobuf messages as function/method parameters then I totally agree with Matt Ball's answer.