CQRS for Great Good

Transcript

1. (Neue Wege mit CQRS) Oliver Wolf CQRS for Great Good
2. Oliver Wolf @owolf www.innoQ.com @innoQ
3. CQRS Command Query Responsibility Segregation
4. The default architecture for distributed business apps Domain Model User Interface Remote Facade Application Services DTO DTO ORM DB findCustomers() getCustomer() updateCustomer() <customer> <name>John Doe</name> <address>...</address> ... </customer>
5. The default architecture for distributed business apps Domain Model User Interface Remote Facade Application Services DTO DTO ORM DB GET /customers?filter=... GET /customer/{id} PUT /customer/{id} { "name": “John Doe”, "address": { ... } }
6. The default architecture for distributed business apps Domain Model User Interface Remote Facade Application Services DTO DTO ORM DB GET /customers?filter=... GET /customer/{id} PUT /customer/{id} { "name": “John Doe”, "address": { ... } } Anything wrong with this?
7. Maybe not.
8. Scalability?
9. Domain Model?
10. “The features that characterize a class are divided into commands and queries. A command serves to modify objects, a query to return information about objects.” Bertrand Meyer ETH Zurich cited from: Object-Oriented Software Construction, second edition, 1997
11. CQS Command Query Separation
12. Part of the Design-by-Contract methodology First demonstrated in the object-oriented EIFFEL programming language
13. class Foo { void command(); Result query(); } Mutates state Returns a value without causing side effects
14. CQRS = CQS in the large Scope is a single class Scope is a Bounded Context
15. interface CustomerService { void updateCustomer(Customer); CustomerList findCustomers(CustomerQuery); Customer getCustomer(ID); void deleteCustomer(ID); }
16. interface CustomerQueryService { CustomerList findCustomers(CustomerQuery); Customer getCustomer(ID); } interface CustomerCommandService { void updateCustomer(Customer); void deleteCustomer(ID); }
17. The default architecture for distributed business apps Domain Model User Interface Remote Facade Application Services DTO DTO ORM DB
18. The default architecture for distributed business apps Domain Model User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services CQRS PATTERN APPLIED
19. That’s it? You serious???
20. Yes, sorry.
21. The interesting thing about CQRS is not the pattern itself.
22. It’s damn simple, actually.
23. But it encourages you to challenge established assumptions and opens up new architectural options!
24. Unlearn what you have learned you must.
25. assumptions we often take for granted
26. Assumption 1: Reads and writes are strongly cohesive, so they must be part of the same Bounded Context.
27. Assumption 1: Reads and writes are strongly cohesive, so they must be part of the same Bounded Context. FALSE
28. The default architecture for distributed business apps Domain Model User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services CQRSified
29. The default architecture for distributed business apps CQRSified User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services Domain Model ORM Domain Model Query Facade Query Services ORM Domain Model Query Facade Query Services ORM Domain Model Command and query parts can scale independently, e.g. to accommodate highly asymmetric load.
30. Assumption 2: Reads and writes use the same data, so they must be served from and applied to the same domain model.
31. Assumption 2: Reads and writes use the same data, so they must be served from and applied to the same domain model. FALSE
32. The default architecture for distributed business apps Queries can benefit from a specialized query model, optimized for quick data retrieval (de-normalized, pre-aggregated,...) User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services Command Model ORM Domain Model Query Facade Query Services ORM Domain Model Query Facade Query Services ORM Query Model CQRSified
33. The default architecture for distributed business apps Queries can benefit from a specialized query model, optimized for quick data retrieval (de-normalized, pre-aggregated,...) User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services Command Model ORM Domain Model Query Facade Query Services ORM Domain Model Query Facade Query Services ORM Query Model ‣validate and process commands ‣keep data consistent ‣guarantee ACID properties ‣behaviour part of domain model ‣relatively difficult to scale out ‣rich query capabilities ‣short response times ‣different views on data ‣potentially denormalized ‣relatively easy to scale out CQRSified
34. Assumption 3: Even for queries, we have to go through the domain model to abstract from the underlying database model.
35. Assumption 3: Even for queries, we have to go through the domain model to abstract from the underlying database model. FALSE
36. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... The default architecture for distributed business apps Queries are just dealing with data, not with behaviour. What do we need objects for? CQRSified
37. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... The default architecture for distributed business apps Queries are just dealing with data, not with behaviour. What do we need objects for? CQRSified Introduce thin read layer that makes optimized use of the database’s query capabilities No ORM, no fuss – just plain SQL (SQL happens to be really good at queries, you know...)
38. Assumption 4: We must use the same database for queries and commands to make sure that data is consistent.
39. Assumption 4: We must use the same database for queries and commands to make sure that data is consistent. FALSE
40. The default architecture for distributed business apps In many cases, eventual consistency is sufficient. The data users are looking at in the UI is always stale to some extent. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... Query DB Query DB Query DB Event Handler Event Handler Event Handler Events Write CQRSified
41. The default architecture for distributed business apps In many cases, eventual consistency is sufficient. The data users are looking at in the UI is always stale to some extent. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services DTO DTO ORM DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... Query DB Query DB Query DB Event Handler Event Handler Event Handler Events Write CQRSified Separate query DB, not necessarily relational Command model emits events when data changes Event handlers process events and update query DB asynchronously
42. Assumption 5: Commands must be processed immediately to ensure data consistency.
43. Assumption 5: Commands must be processed immediately to ensure data consistency. FALSE
44. The default architecture for distributed business apps In many cases, users don’t care if their actions have immediate effect – as long as they eventually get feedback. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services CMD DTO ORM DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... Query DB Query DB Query DB Event Handler Event Handler Event Handler Events Write Command Queue Commands CQRSified
45. The default architecture for distributed business apps In many cases, users don’t care if their actions have immediate effect – as long as they eventually get feedback. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services CMD DTO ORM DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... Query DB Query DB Query DB Event Handler Event Handler Event Handler Events Write Command Queue Commands CQRSified Decouple user interaction and command processing Give visual feedback and allow users to check processing progress (and result)
46. Assumption 6: The current state of domain objects must be persistent.
47. Assumption 6: The current state of domain objects must be persistent. FALSE
48. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services CMD DTO DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... Query DB Query DB Query DB Event Handler Event Handler Event Handler Events Write Command Queue Commands Processor The default architecture for distributed business apps CQRS plays well with an Event Sourcing architecture – you just store events and re-create the state of domain objects as needed. CQRSified
49. Query Services Query Services Thin Read Layer User Interface Query Facade Query Services CMD DTO DB Command Facade Command Services Command Model Query Facade Query Services Query Facade Query Services SELECT … FROM … WHERE ... Query DB Query DB Query DB Event Handler Event Handler Event Handler Events Write Command Queue Commands Processor The default architecture for distributed business apps CQRS plays well with an Event Sourcing architecture – you just store events and re-create the state of domain objects as needed. Events can always be replayed from event store CQRSified Model is strictly in- memory DB holds events, not state
50. ‣ capture each update to application state in an event ‣ changes to domain objects are the result of applying events ‣ events are immutable ‣ events are a representation of what has happened at a specific time Event Sourcing in a nutshell
51. ‣ allows you to rebuild application state at any point in time just by replaying events from t0 up to that point in time ‣ allows you to analyse historic data based on detailed events that would otherwise have been lost ‣ gives you an audit log “for free” ‣ you can add new, optimized read models (potentially in-memory) later without migration hassle and such What’s so great about Event Sourcing?
52. http://www.flickr.com/photos/kaptainkobold/5170454747/ How do I convince my boss that we’ll have to rewrite our application with CQRS???
53. Chances are you don’t.
54. CQRS is not a silver bullet and doesn’t apply everywhere. Beware of the added complexity!
55. Don’t do CQRS… …if your application is just a simple CRUD-style app. …if you don’t have scaling issues. …if it doesn’t help improve your domain models.
56. Consider doing CQRS… …if your write/read load ratio is highly asymmetrical. …if scaling your application is difficult. …if your domain model is bloated by complex domain logic, making queries inefficient. …if you can benefit from event sourcing.
57. Frameworks, any?
58. Axon Framework (Java) www.axonframework.org Piotr Wyczesany Lokad (.NET) http://lokad.github.com/lokad-cqrs/ Piotr Wyczesany …and some more Piotr Wyczesany
59. www.axonframework.org
60. Apache 2 license Version 2.0 released earlier this year Maintained by Trifork Netherlands Commercial support offering available
62. Distributed Event Bus, based on ‣JGroup ‣Disruptor ‣AMQP ‣Spring Integration ‣File system ‣JPA ‣MongoDB ‣Cassandra ‣Redis ‣Google App Engine DataStore Supports both state persistence (via JPA) and event sourcing
63. ‣ relatively unobtrusive, commands and events are just Java objects ‣ Event sourcing is not mandatory, state persistence is still supported ‣ Support for multiple EventBus and EventStore implementations ‣ Integrates nicely with Spring ‣ supports complex business transactions (“sagas” in DDD parlance) Things I like about Axon
64. ‣ requires some familiarity with DDD terms (you should at least know what an Aggregate is) ‣ Axon doesn’t try to hide CQRS from developers – whether this good or bad depends on experience and knowledge about CQRS Things I’m not quite sure about
65. That’s all I have. Feel free to ask me anything! @owolf