Example of application built using Akka.net. Please Expect WIP code-quality.
Still missing
C#
Topshelf
Akka.Net
Serilog
F#
Chessie
Still missing
Still missing
Domain-Driven Design: Tackling Complexity in the Heart of Software
https://www.amazon.com/Domain-Driven-Design-Tackling-Complexity-Software/dp/0321125215
Implementing Domain-Driven Design [IDDD]
https://www.amazon.com/gp/product/0321834577/
Reactive Messaging Patterns with the Actor Model: Applications and Integration in Scala and Akka
https://www.amazon.com/Reactive-Messaging-Patterns-Actor-Model/dp/0133846830/
Effective Akka
https://www.amazon.com/Effective-Akka-Jamie-Allen/dp/1449360076
Enterprise Integration Patterns: Designing, Building, and Deploying Messaging Solutions
https://www.amazon.com/Enterprise-Integration-Patterns-Designing-Deploying/dp/0321200683
Pattern-Oriented Software Architecture Volume 1: A System of Patterns
https://www.amazon.com/Pattern-Oriented-Software-Architecture-System-Patterns/dp/0471958697
From IDDD:
Shared Kernel: Sharing part of the model [...] can leverage design work or undermine it. [...]
Keep the kernel small. [...] Define a continuous integration process [...]
Customer-Supplier Development: When two teams are in an upstream-downstream relationship,
where the upstream team may succeed interdependently of the fate of the downstream team,
[...] Negotiate and budget tasks for downstream requirements so that everyone understands
the commitment and schedule.
Conformist: When two development teams have an upstream/downstream relationship in which the
upstream team has no motivation to provide for the downstream team’s needs, the downstream
team is helpless. [...] The downstream [...] slavishly adhering to the model of the upstream team.
Anticorruption Layer: [...] when control or communication is not adequate to pull off a
shared kernel, partner, or customer-supplier relationship [...] As a downstream client,
create an isolating layer to provide your system with functionality of the upstream system in
terms of your own domain model.
[...]
A Domain Service (7) can be defined in the downstream Context for each type of Anticorruption Layer. You may also put an Anticorruption Layer behind a Repository (12) interface. If using REST, a client Domain Service
implementation accesses a remote Open Host Service. Server responses produce representations
as a Published Language. The downstream Anticorruption Layer translates representations into
domain objects of its local Context.
Open Host Service: Define a protocol that gives access to your subsystem as a set of services.
[...]
This pattern can be implemented as REST-based resources [...] We generally think of
Open Host Service as a remote procedure call (RPC) API, but it can be implemented
using message exchange.
Published Language: The translation between the models of two Bounded Contexts requires a
common language. Use a well-documented shared language that can express the necessary domain
information as a common medium of communication [...]
[...]
This can be implemented [...] as an XML schema. When expressed with REST-based services,
the Published Language is rendered as representations of domain concepts. [...]
It is also possible to render representations as Google Protocol Buffers. If you are
publishing Web user interfaces, it might also include HTML representations. One advantage
to using REST is that each client can specify its preferred Published Language,
and the resources render representations in the requested content type. REST also has the
advantage of producing hypermedia representations, which facilitates HATEOAS.
Hypermedia makes a Published Language very dynamic and interactive, enabling clients
to navigate to sets of linked resources. The Language may be published using standard
and/or custom media types. A Published Language is also used in an Event-Driven
Architecture (4), where Domain Events (8) are delivered as messages to subscribing interested parties.
Separate Ways: [...] If two sets of functionality have no significant relationship, they can
be completely cut loose from each other.
Big Ball of Mud: [...] models are mixed and boundaries are inconsistent. Draw a boundary around
the entire mess and designate it a Big Ball of Mud.
From [IDDD]
For example, what would happen if in the [Bounded Context A] user mistakenly unassigns
[User A1] from the [Role A1 in Bounded Context A]? Well, we receive an Event-carrying
notification that indicates that fact, so we use [Service B in Bounded Context B] to disable
the [Role B in Bounded Context B] corresponding to [User A1]. Wait. Seconds later the user
realizes that she has unassigned the wrong user from the [Role A1 in Bounded Context A],
and that she should have unassigned [User A2] instead. So she quickly assigns [user A1]
back to the role and unassigns [User A2]. [...] is everything actually OK?
We could be making a bad assumption about this use case. We are assuming that we receive
the notifications in the order in which they actually occurred in the [Bounded Context A].
Yet, things might not always work out so well. What would happen if, for whatever reason,
the notifications [were received in the reverse order]. [User A1] will be stuck in a disabled state,
and at best someone will have to patch the data in the [...] database, or the user will have to
play some tricks to get the right user reenabled.
This can happen, and ironically, it seems to always happen when we overlook the fact
that it could happen. So, how do we prevent this?
for a solution see Life beyond Distributed Transactions: an Apostate’s Opinion http://www.ics.uci.edu/~cs223/papers/cidr07p15.pdf
In the author opininon the best way to scale an app is:
Scalable Apps Use Uniquely Identified “Entities”
Atomic Transactions Cannot Span Entities
Messages Are Addressed to Entities
Entities Manage Per-Partner State (“Activities”)
From IDDD
[...] model true invariants in consistency boundaries according to real business rules.
Consider the advantages of designing small Aggregates.
See why you should design Aggregates to reference other Aggregates by identity.
Discover the importance of using eventual consistency outside the Aggregate boundary.
Learn Aggregate implementation techniques, including Tell, Don’t Ask and Law of Demeter.
From IDDD
Consider why Entities have their proper place when we need to model unique things.
See how unique identities may be generated for Entities.
Look in on a design session as a team captures its Ubiquitous Language (1) in Entity design.
Learn how you can express Entity roles and responsibilities.
See examples of how Entities can be validated and how to persist them to storage.
From IDDD
Learn how to understand the characteristics of a domain concept to model as a Value.
See how to leverage Value Objects to minimize integration complexity.
Examine the use of domain Standard Types expressed as Values.
Consider how SaaSOvation learned the importance of Values.
Learn how the SaaSOvation teams tested, implemented, and persisted their Value types.
//TODO replace the chapter roadmap with better guidelines
Why CQRS? From [IDDD]
It can be difficult to query from Repositories all the data users need to view.
This is especially so when user experience design creates views of data that
cuts across a number of Aggregate types and instances. The more sophisticated
your domain, the more this situation tends to occur.
Using only Repositories to solve this can be less than desirable. We could require
clients to use multiple Repositories to get all the necessary Aggregate instances,
then assemble just what’s needed into a Data Transfer Object (DTO) [Fowler, P of EAA].
Or we could design specialized finders on various Repositories to gather the disjointed
data using a single query. If these solutions seem unsuitable, perhaps we should instead
compromise on user experience design, making views rigidly adhere to the model’
s Aggregate boundaries. Most would agree that in the long run a mechanical and spartan user
interface won’t suffice.
Is there an altogether different way to map domain data to views? The answer lies in the
oddly named architecture pattern CQRS [Dahan, CQRS; Nijof, CQRS]. It is the result of pushing
a stringent object (or component) design principle, command-query separation (CQS), up to
an architecture pattern.
This principle, devised by Bertrand Meyer, asserts the following:
Every method should be either a command that performs an
action, or a query that returns data to the caller, but
not both. In other words, asking a question should not change
the answer. More formally, methods should return a value only
if they are referentially transparent and hence possess no
side effects. [Wikipedia, CQS]
From IDDD:
There are varying definitions of Event Sourcing, so some clarification is fitting.
We are discussing the use where every operational command executed on any given
Aggregate instance in the domain model will publish at least one Domain Event that
describes the execution outcome. Each of the events is saved to an Event Store (8)
in the order in which it occurred. When each Aggregate is retrieved from its
Repository, the instance is reconstituted by playing back the Events in the order
in which they previously occurred
From IDDD:
Long-Running Processes, aka Sagas
A Long-Running Process is sometimes called a Saga, but depending on your background that
name may collide with a preexisting pattern. An early description of Sagas is presented
in [Garcia-Molina & Salem].
//IMPROVE description
Effective Akka (https://www.amazon.com/Effective-Akka-Jamie-Allen/dp/1449360076)
worker actors are meant for parallelization or separation of dangerous tasks into
actors built specifically for that purpose, and the data upon which they will act is always
provided to them.
Petabridge also have a post about this pattern, but have called it https://petabridge.com/blog/top-akkadotnet-design-patterns/
The Character Actor Pattern is used when an application has some risky but critical operation to execute,
but needs to protect critical state contained in other actors and ensure that there are no negative side effects.
It’s often cheaper, faster, and more reliable to simply delegate these risky operations to a purpose-built,
but trivially disposable actor whose only job is to carry out the operation successfully or die trying.
These brave, disposable actors are Character Actors.
Domain actors, introduced in the previous section, represent a live
cache where the existence of the actors and the state they encapsulate are a view of the
current state of the application.
One of the most difficult tasks in asynchronous programming is trying to capture con‐
text so that the state of the world at the time the task was started can be accurately
represented at the time the task finishes. However, creating anonymous instances of
Akka actors is a very simple and lightweight solution for capturing the context at the
time the message was handled to be utilized when the tasks are successfully completed.
Those are good reasons for pulling the type you’re creating with the Extra Pattern into
a pre-defined type of actor, where you create an instance of that type for every message
handled.
Also See: https://www.javacodegeeks.com/2014/01/three-flavours-of-request-response-pattern-in-akka.html
introducing a layer of supervision between the
[one hierarchy] and [another hierarchy of] children, as we see [below] [...] I can tailor
supervision to be specific to failure that can occur with accounts and that which may
occur with [a hierarchy].
+---------------+
+---------+ Actor +-------+
| +---------------+ |
| |
+--------+------+ +---+---------------+
+-+ Supervisor 1 +--+ +-+ Supervisor 2 +--+
| +---------------+ | | +-------------------+ |
| | | |
+--+------+ +-------+--+ +----+-----+ +--------+--+
| child 1 | | child 2 | | child 3 | | child 4 |
+---------+ +----------+ +----------+ +-----------+
Backoff Superviser Pattern
Provided as a built-in pattern the akka.pattern.BackoffSupervisor implements the so-called
exponential backoff supervision strategy, starting a child actor again when it fails, each
time with a growing time delay between restarts.
also see http://getakka.net/docs/concepts/supervision
At this point it is vital to understand that supervision is about forming a recursive fault
handling structure. If you try to do too much at one level, it will become hard to reason about,
hence the recommended way in this case is to add a level of supervision.
I call these kinds of actors
“sentinels,” as they guard the system from falling out of synchrony with a known
expectation of what state should exist in the system.
https://petabridge.com/blog/how-to-stop-an-actor-akkadotnet/
Use Immutable Messages with Immutable Data
The reasons for this are the same as stabilizing a variable before using it in a future—you
don’t know when the other actor will actually handle the message or what the value of the
variable will be at that time. The same goes for when you want to send an immutable value
that contains mutable attributes.
Erlang does this for you with copy on write (COW) semantics, but we don’t get that for free
in the JVM. Assuming your application has the heap space, take advantage of it.
Hopefully, these copies will be short-lived allocations that never leave the Eden space of
your garbage collection generations, and the penalty for having duplicated the value will be minimal.
also see:
http://hackerboss.com/i-have-seen-the-future-and-it-is-copy-on-write/
https://www.nuget.org/packages/System.Collections.Immutable/
Immutability in c# is Hard
Comparison between c# and f# for Immutability http://www.slideshare.net/ScottWlaschin/domain-driven-design-with-the-f-type-system-functional-londoners-2014
http://getakka.net/docs/working-with-actors/Stashing%20Messages TODO
When DDD is applied with Actor based, it is common to model Aggregate Roots with Actors see http://pkaczor.blogspot.co.uk/2014/04/reactive-ddd-with-akka.html
To allow horizontal scalability, the ideal is that each Aggregate Root could be running in a different machine. Hence, there is no distributed transaction mechanism to allow modification of two Actors, for example the famous "transfer money between two accounts".
To allow such operation without distributed transaction, one can use the following workflow.
1 - With just one operation on the Source AccountActor:
1.1 - Decrease the account balance (the desired operation)
1.2 - Insert that there is an operation in progress (pending operation storage)
2 - Persist Source AccountActor;
2.1 - From this point it is guaranteed that the operation will finish;
3 - Create a TransferActor to communicate with Destination AccountActor;
3.1 - AccountActor must have a mechanism to guarantee that for every operation in progress exist one associated TransferActor (Cameo Pattern);
4 - TransferActor must send the message to the Destination AccountActor "n" times following some kind of pattern (every x seconds, for example)(Stubborn Link);
4.1 - If after n transient errors or a non-transient error, the actor can enter in a human-attention-needed state (maybe the destination account does not exist);
5 - TransferActor enter in Wait-for-confirmation state;
6 - Destination AccountActor will receive the message and do something: or will accept the tranfer or will ask the Account Manager if the money must be accepted, for example;
7 - After the transfer is accepted, Destination AccountActor will send the confirmation that the Transfer was accepted to the TransferActor;
8 - TransferActor will send a message to its parent telling that the transfer was accepted and will kill itself;
9 - Source AccountActor will remove the operation from the in-progress-list;
10 - Operation finished!
This complex workflow guarantees that the operation will be completed even with the following errors:
1 - Source AccountActor modified its data but could not start the TransferActor;
1.1 - The Source AccountActor have a auto-healing mechanism to always have a TransferActor to each operation-in-progress. So eventually the TransferActor will be started;
1.2 - If for some reason the application died, when the application start again, the AccountActor will create the TransferActor for each operation-in-progress;
2 - TransferActor started but somehow died;
2.1 - Transient errors will be handled by the retry mechanism;
2.2 - Non-transient errors will scalate to human intervention;
2.3 - If some reason the TransferActor vanished, the AccountActor-auto-healing-mechanism will recreate it.
3 - Source TransferActor cannot communicate with the Destination AccountActor
3.1 - Transient errors will be handled by the retry mechanism;
1.2 - Non-transient errors will scalate to human intervention;
4 - Destination AccountActor accepted the transfer, persisted its data, but died after that;
4.1 - The Source TransferActor will resend the message after some time;
4.1 - Destination AccountActor must be able to recognize that a tranfer was already accepted (must idempotently process this message). So if it died before sending the confirmation the Source TransferActor will resend the request and the Destination AccountActor will confirm the transfer without doing any data modification.
5 - The TransferActor died between waiting the confirmation and sending the confirmation to its parent
5.1 - The Source AccountActor will recreat it, the TransferActor will send the request again, the destination AccountActor will just confirm it;
6 - The TransferActor have sent the confirmation to the AccountActor and killed itself but the AccountActor have not persisted the operation in progress updated
6.1 - The AccountActor auto-healing-mechanism will start the whole process again until everything works.
see: Guaranteed Sender Actor Guaranteed Receiver Actor
how to use dispatchers
to create failure zones and prevent failure in one part of the application from affecting
another. This is sometimes called the Bulkhead Pattern. And once I create the failure
zones, how do I size the thread pools so that we get the best performance from the least
amount of system resources used?
also see: http://skife.org/architecture/fault-tolerance/2009/12/31/bulkheads.html
https://www.amazon.de/Reactive-Enterprise-Actor-Model-Applications/dp/0133846830
The domainModel actor serves as parent and supervisor for categories of
domain model of concepts [IDDD]
By definition, when using Actor Model, messages are delivered to an actor at most once.
For many uses this all works out just fine, especially when considering that any given
message is almost always successfully delivered. As well, when a given message is not
actually delivered, there are ways to make sure that it is eventually. For example, the
sending actor can listen for a response to its sent message, and also set a timeout. If the
timeout fires before the response is received, the client actor will simply resend the
message. In the case of a crash, possibly restart the receiving actor before it sends the
repeat message.
also see: http://getakka.net/docs/persistence/at-least-once-delivery
The BlackHoleActor is a special-purpose actor that will not respond to anything. No matter
what message you tell the BlackHoleActor, it will not respond. BlackHoleActor is perfect
for testing failure conditions. Common failure conditions are when an actor or
service dies, or an operation times out.
https://petabridge.com/blog/how-to-unit-test-akkadotnet-actors-akka-testkit/