DDD DYSFUNCTIONAL...3/13/2019 reveal.js 1/133 DYSFUNCTIONAL DDD

3/13/2019 reveal.js

http://localhost:8000/?print-pdf#/ 1/133

DYSFUNCTIONALDYSFUNCTIONALDDDDDD

3/13/2019 reveal.js


JAREKJAREKWizard, Anarchitect, Coder

3/13/2019 reveal.js


JAREKJAREKWizard, Anarchitect, Coder

> 25 years of coding

3/13/2019 reveal.js


3/13/2019 reveal.js


DISCLAIMERDISCLAIMER

3/13/2019 reveal.js


DISCLAIMERDISCLAIMERI tried not to attack anyone personally, actually I dolike my colleagues doing DDD, and I have learned alot from some of them

3/13/2019 reveal.js


DISCLAIMERDISCLAIMERI tried not to attack anyone personally, actually I dolike my colleagues doing DDD, and I have learned alot from some of themnot an expert, (some experience wih CQRS/ES notDDD)

3/13/2019 reveal.js


WHY DDD?WHY DDD?People talk about itObject orientedworldLike mixing concepts

3/13/2019 reveal.js


My problem with DDD

3/13/2019 reveal.js


DDD story resembles me

UMLwhat happened toAgile

3/13/2019 reveal.js


3/13/2019 reveal.js


cool

3/13/2019 reveal.js


3/13/2019 reveal.js


What??

3/13/2019 reveal.js


AGILE COOLAGILE COOLTests, TDDCI,Code overcommentscontact with user

3/13/2019 reveal.js


3/13/2019 reveal.js


AGILE REALITYAGILE REALITYboringmeetingsagile coaches,certificatesagile toolsbooksconferencestrainingsvelocity :-)

3/13/2019 reveal.js


3/13/2019 reveal.js


There is nothing wrong about money.

I work for money!

3/13/2019 reveal.js


But who likes marketing b...it?!

3/13/2019 reveal.js


3/13/2019 reveal.js


For years I was scared of DDD marketing

3/13/2019 reveal.js


BOOKSBOOKS

3/13/2019 reveal.js


Make logic of the system visible in code

3/13/2019 reveal.js


DDD GOOD PARTSDDD GOOD PARTScommunityconstantimprovementlots of patterns /ideasinteresting stories

3/13/2019 reveal.js


3/13/2019 reveal.js


DDD bad(?) parts

marketingexample projectsover-engineeringpartly toxiccommunityhard to see the point

3/13/2019 reveal.js


If a tool, methodology, framework makes simple casecomplex

it will probably not make a real life, complex projectsimple

3/13/2019 reveal.js


3/13/2019 reveal.js


BUTBUT

3/13/2019 reveal.js


3/13/2019 reveal.js


DDDDDD

3/13/2019 reveal.js


HARD WAYHARD WAY

3/13/2019 reveal.js


Skipping some essential parts.

Like bounded context etc.

3/13/2019 reveal.js


DDD is mostly not about technology

3/13/2019 reveal.js


FINDING COMMONFINDING COMMONLANGUAGELANGUAGE

Ubiquitous Language

3/13/2019 reveal.js


Some concepts are useful, but have nothing to do withFP

3/13/2019 reveal.js


I just love to code more

3/13/2019 reveal.js


And you wil not pay me areal money (for a fancy car) :-(

3/13/2019 reveal.js


LETS PLAY SNAKELETS PLAY SNAKE(MULTIPLAYER)(MULTIPLAYER)

3/13/2019 reveal.js


3/13/2019 reveal.js


COMMANDSCOMMANDS

3/13/2019 reveal.js


data SnakeCommand = SetDirection { wantedDirection :: SnakeDirection } | MakeStep | Begin { initName :: String , initCell :: SnakeCell }

3/13/2019 reveal.js


COMMANDSCOMMANDSuser or subsytem wants to do somethingcommands may be associated withvalidation

3/13/2019 reveal.js


EVENTSEVENTSdata SnakeEvent = DirectionChanged { newDirection :: SnakeDirection } | StepMade | Killed | Born { bornName :: String , bornCell :: SnakeCell } deriving (Eq, Show, Generic)

3/13/2019 reveal.js


EVENTEVENTit has happenedno validation (it really happened)a single command is associated in 0..nevents

3/13/2019 reveal.js


Fact

You will not find anything about this in Eric Evans book(blue))

Even though nowadays DDD community seems to beall around those concepts

3/13/2019 reveal.js


EVENT SOURCINGEVENT SOURCINGWhat if just stored only events?

3/13/2019 reveal.js


[ { "event": { "tag": "Born", "bornName": "aa", "bornCell": { "cellY": 18, "cellX": 14 } }, "snakeId": "4d09ac06-0375-4cb0-ad08-c70d14968677" }, { "event": { "tag": "DirectionChanged",

3/13/2019 reveal.js


COMMANDCOMMANDSOURCING?SOURCING?

Also possible... but in fact harder - validation is aproblem

3/13/2019 reveal.js


VALUE OBJECTVALUE OBJECTimmutable...represents value (from reallife)has no identityproperties define equality

3/13/2019 reveal.js


type SnekeId = String type NickName = Text data SnakeDirection = SnakeUp | SnakeRight | SnakeDown | SnakeLeft

3/13/2019 reveal.js


ENTITYENTITYhas identityin OOP may be mutableID defines equalityentity may contain valueobjects

entity may contain entities

3/13/2019 reveal.js


data SnakeCell = SnakeCell { cellX :: Int , cellY :: Int } data SnakeState = Alive { direction :: SnakeDirection , cells :: [SnakeCell] , maxLength :: Int } | Dead | Init data Snake = Snake { name :: String

3/13/2019 reveal.js


Easy?

3/13/2019 reveal.js


SnakeCell(x,y) is value object orentity?

3/13/2019 reveal.js


To think

I have never introduced type SnakeEntity =(SnakeId, Snake)

Only have SnakeState which does not (physically)contain Id

3/13/2019 reveal.js


To think

I have never introduced type SnakeEntity =(SnakeId, Snake)

Only have SnakeState which does not (physically)contain Id

an Entity without id??

3/13/2019 reveal.js


Some DDD concepts may not be explicitly existing incode

(SnakeId, Snake)

3/13/2019 reveal.js


It gets worse

3/13/2019 reveal.js


AGGREGATEAGGREGATE

3/13/2019 reveal.js


Cluster of objects (entities, value objects, + )

3/13/2019 reveal.js


Aggregate remains consistent

3/13/2019 reveal.js


Keeps invariants

3/13/2019 reveal.js


Transactions should not cross aggregates

3/13/2019 reveal.js


AGGREGATE ROOTAGGREGATE ROOTSelected entity from Aggregate (root)outside world communicates with it (sendscommands)outside world only keeps reference to this rootobject

command handlerevent handler

3/13/2019 reveal.js


Typeclass source:

class Aggregate s where data Error s :: * data Command s :: * data Event s :: * execute :: s -> Command s -> Either (Error s) (Event s) apply :: s -> Event s -> s seed :: s

https://gist.github.com/Fristi/7327904

https://gist.github.com/Fristi/7327904

3/13/2019 reveal.js


COMMAND HANDLERCOMMAND HANDLERexecuteCommand :: Snake -> SnakeCommand -> [SnakeEvent]

3/13/2019 reveal.js


BETTER COMMANDBETTER COMMANDHANDLERHANDLER

executeCommand :: Snake -> SnakeCommand -> Either MyError [SnakeEvent]

3/13/2019 reveal.js


EVENT HANDLEREVENT HANDLERapplyEvent::Snake-> SnakeEvent -> Snake

3/13/2019 reveal.js


SUMMARYSUMMARYdefine commandsdefine eventsselect root Entitydefine commandshandlerdefine events handler

3/13/2019 reveal.js


Real Command handler

executeCommand :: SnakeData -> SnakeCommand -> [SnakeEvent] executeCommand SnakeData {state = Alive {}} MakeStep = [StepMaexecuteCommand _ MakeStep = [] executeCommand SnakeData {state = Alive {direction = od}} SetD | opposite = [] | otherwise = [DirectionChanged {newDirection = nd}] where opposite = V.dirIs0 $ V.dirPlus newVec currentVec newVec = V.dirVector nd currentVec = V.dirVector od executeCommand anySnake SetDirection {} = [] executeCommand SnakeData {state = Alive {}} Die = [Killed] executeCommand _ Die = [] executeCommand SnakeData {state = Init} Begin {initName = d, iexecuteCommand _ Begin {} = []

3/13/2019 reveal.js


Real event handler

applyEventX snake@(SnakeData {state = alive@Alive {}}) Directi makeRes $ snake { state = alive{direction = nd} } applyEventX snake@(SnakeData {state = Alive {}}) Killed = makeapplyEventX SnakeData {state = Init} Born {bornName = nm, born newSnake = SnakeData { name = nm , state = initial newCells = [cell], removedCells = [] } where initialState = Alive { direction = SnakeUp, cells = [applyEventX snake@(SnakeData {state = alive@Alive {maxLength = makeRes snake { state = alive { maxLength = n+3} } applyEventX _ _ = error "todo"

3/13/2019 reveal.js


Modelling with events, commands is not needed inDDD

It was not even considered in an original DDD book

behaviour first seem to be quite efficient

3/13/2019 reveal.js


Event storming

3/13/2019 reveal.js


Alternative to command handler / Aggregate

3/13/2019 reveal.js


Commands

data SnakeCommand = SetDirection { wantedDirection :: SnakeDirection } | MakeStep | Begin { initName :: String , initCell :: SnakeCell } | Die

3/13/2019 reveal.js


Have You seen that before?

data SnakeCommand next = SetDirection { wantedDirection :: SnakeDirection } | MakeStep next | Begin { initName :: String , initCell :: SnakeCell } next | Die

3/13/2019 reveal.js


Free monad DSL

3/13/2019 reveal.js


Seems to be more usable in sequencing

3/13/2019 reveal.js


In typical REST we have one http call -> onecommand. Sequencing is not that needed.

3/13/2019 reveal.js


HOW TO FINDHOW TO FINDAGGREGATES?AGGREGATES?

3/13/2019 reveal.js


Whole system as an aggregate?

(One Big Aggregate)

3/13/2019 reveal.js


REPOSITORYREPOSITORY

3/13/2019 reveal.js


Remember DAO?

3/13/2019 reveal.js


Magic...

3/13/2019 reveal.js


3/13/2019 reveal.js


loadEntity::Id->IO EntitysaveEntity::Id->Entity->IO()etc...

3/13/2019 reveal.js


Fact Lots of magic Java frameworks trace state ofobjects and automatically persist changes to database

This means that a sensible repository save methodmay look like:

3/13/2019 reveal.js


void save(MyObject t) { }

3/13/2019 reveal.js


Yep, this works. There are lot of such projects.

void save(MyObject t) { }

3/13/2019 reveal.js


But what in case of event sourcing?

3/13/2019 reveal.js


data SnakeAggregate = SnakeAggregate { state :: SnakeState, uncommittedEvents :: [SnakeEvent] }

3/13/2019 reveal.js


Common pattern in DDD style event sourcing is to savethose uncommited events

3/13/2019 reveal.js


I find it unnatural

3/13/2019 reveal.js


I started to send commands to a Repository

3/13/2019 reveal.js


applyCommand :: SnakesRepo-> SnakeId -> Snake.SnakeCommand -> IO SnakesRepo

3/13/2019 reveal.js


This is so unDDD

applyCommand :: SnakesRepo-> SnakeId -> Snake.SnakeCommand -> IO SnakesRepo

3/13/2019 reveal.js


Makes for more sense than repeating a code with saveevents

In my aggregates I do not have those uncommitedevents (is this a domain?)

3/13/2019 reveal.js


CQSCQSCommand Query separation

3/13/2019 reveal.js


If you ask (Query) do not change the state

3/13/2019 reveal.js


If You change state (Command) do not expect result

3/13/2019 reveal.js


A Stackvoid stack.push( T t); T stack.pop();

3/13/2019 reveal.js


A CQS Stackvoid stack.push( T t); T stack.top(); void stack.pop();

3/13/2019 reveal.js


Simple?

3/13/2019 reveal.js


what if called on empty stack?void stack.pop(); // boom

3/13/2019 reveal.js


OO world consensus:

commands may return exceptions, some status, etc.

3/13/2019 reveal.js


Is an error not a result?

3/13/2019 reveal.js


In FP world

3/13/2019 reveal.js


push::Stack a->a->Stack a top::Stack a->a pop::Stack a->Stack a

3/13/2019 reveal.js


In FP world each operation gives a result

3/13/2019 reveal.js


IO ()

3/13/2019 reveal.js


Is it really bad?

whatAPop::Stack a->(Stack a, a)

3/13/2019 reveal.js


Actually I do not see much sense in classical CQS

nice to have separated queriesError/Exception is an resultIt only makes API easier to use ... in some mediocrelanguagesasync?

3/13/2019 reveal.js


push::Stack a -> MonadAsync (Stack a)

3/13/2019 reveal.js


CQRSCQRSCommand Query Response Segregation

3/13/2019 reveal.js


CQRS ~ CQS on a higher level

3/13/2019 reveal.js


Write/Command model - Aggregates

3/13/2019 reveal.js


Read/Query model - Projections

3/13/2019 reveal.js


3/13/2019 reveal.js


Some CQRS principles

Critical: Events application cannot use any externaldata (projection)So�er: Commands should not use projection toproduce EventsYou can always recreate Aggregates using eventsYou can always recreate multiple projections usingevents

3/13/2019 reveal.js


Fact

In less impure languages it is easy to make amistake

LocalDate.now() someRandom.nextInt()

3/13/2019 reveal.js


Fact

In typical Event souring operations likefindMeIdsOfAllAggregates are performed

using projections

3/13/2019 reveal.js


My game field projection

data PlaneState = PlaneState { allSnakes :: Repo.SnakesMap , allCells :: CellsMap , changes :: Changes } deriving (Show, Generic) -- projection applyEvent :: PlaneState -> Repo.SnakeQualifiedEvent -> (Plane

3/13/2019 reveal.js


Used by browser

3/13/2019 reveal.js


Projections are great:

potentialperformancemake UI code simpler

3/13/2019 reveal.js


I use them to detect collisions

If I find collision on a gameField I send Die commandto snake

I read few times this is wrong...

But I do not see better solution (other than One BigAggregate)

3/13/2019 reveal.js


In CQRS some operations are quite hardnextId::Sequence->Int

3/13/2019 reveal.js


CODECODEdsnake - haskell rest server, (yesod)

Work in progress

Please, do not use it as a sensible DDD exampleresource (yet)

https://github.com/jarekratajski/dysfunctional_ddd

https://github.com/jarekratajski/dysfunctional_ddd

3/13/2019 reveal.js


RESOURCES ONRESOURCES ONDDD/CQRSDDD/CQRS

Bottega presentations

"The" books

http://CQRS.nu (FAQ) page

https://github.com/ddd-by-

examples/event-source-cqrs-sample byKuba Nabrdalik

3/13/2019 reveal.js


https://github.com/vlingo vlingo platform

3/13/2019 reveal.js


Q?Q?

DDD DYSFUNCTIONAL...3/13/2019 reveal.js 1/133 DYSFUNCTIONAL DDD

Documents

Transcript of DDD DYSFUNCTIONAL...3/13/2019 reveal.js 1/133 DYSFUNCTIONAL DDD