Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

SOA and Event Driven Architecture (SOA 2.0)


Published on

A historic view on software integration on how we've failed with our current approach to SOA (also know as SOA 1.0).
We look at the problem, what to do about them using Events and the changes that we need to apply to our design philosophy and architecture to really obtain loose coupling, reuse (or replaceability) and a reactive system.

Published in: Technology, Business

SOA and Event Driven Architecture (SOA 2.0)

  1. 1. Jeppe Cramon – Partner in TigerTeam ApS AANUG Conference September 2013 SOA & EDA
  2. 2. There are problems with SOA • Systems are more fragile • Development and Maintenance costs are higher • Your services are not being reused • You thought SOA would solve your integration problems, but they’ve gotten worse • No one wants to build or maintain services • System performance is worse
  3. 3. If you have arrived at the realization that SOA is a pain and costs too much without enabling business agility or reducing your costs, you are not the only one. You will have to change your approach to achieve the benefits of SOA
  4. 4. Let’s take a trip back in time
  5. 5. And look at how we could do business
  6. 6. Sales Department Old style SOA Design Department Customer Department Finance Department Fabric Department The Customer Create Sales Order
  7. 7. This is slow
  8. 8. We are not selling enough!!
  9. 9. Sales Department Old style SOA II Design Department Customer Department Finance Department Fabric Department Create Sales Order
  10. 10. We need to speed things up! Hackel walks between desks? Way too slow.
  11. 11. We need to speed things up! He needs to call So we make sure that he can Actually, call a lot!
  12. 12. We need to speed things up! But Hackel still have to wait Even if we get creative
  13. 13. We need to speed things up! Something needs to be done!!
  14. 14. Speed up even more! Lets get an automatic switchboard!
  15. 15. Speed up even more! New technology, same problem Even if we get creative again
  16. 16. Or do things Async So Hackel can send request to Ingeborg and Kresten which can respond to Hackel when they’ve got an answer But even that doesn’t scale well, there’s still only one Ingeborg and Kresten
  17. 17. Speed up even more! So Kresten and Ingeborg adds more staff
  18. 18. Scalability is expensive Sales Department Design Department Customer Department Finance Department Fabric Department Coordinator Hackel
  19. 19. Same problem different technology Data Storage Data Storage Data Storage Finance Service Data Service Design Service Hackel Coordinator Service Hackel Coordinator Service Sales Order Service Sales Order Service Customer Customer Fabric Service Customer Service Data Storage Client Process Service layer Activity Service layer Data Service layer
  20. 20. Resemble much?
  21. 21. Now back to the story… • Scaling up Sales requires scaling up other departments relatively much • People cost money and why have 9 Finance people for every 10 sales people just to answer questions about pricing • There’s got to be a better way to handle this • And there is 
  22. 22. Sales Department Event Driven SOA and local caching Design Department Customer Department Finance Department Fabric Department The Customer Create Sales Order
  23. 23. Scalability is cheaper Sales Department Design Department Customer Department Finance Department Fabric Department
  24. 24. This is just the beginning
  25. 25. Have our mindset changed over the last 40 years?
  26. 26. Paper work, file cabinets and manual workflows Inventory ShippingPurchase Manual workflow Manual workflow
  27. 27. Next step was automating paper forms (aka. The birth of CRUD) Inventory ShippingPurchase Manual workflow Manual workflow
  28. 28. Database integration Inventory ShippingPurchase Database Integration Database Integration
  29. 29. Enterprise Application Integration (EAI) Inventory ShippingPurchase Application Integration Application Integration
  30. 30. WebServices to the Rescue? SOA Web of Mess Inventory Purchase ShippingPortal
  31. 31. ESB’s the the Rescue? SOA Web of Mess hidden by ESB Inventory Purchase ShippingPortal ESB
  32. 32. ESB Reality SOA Web of Mess hidden by ESB Inventory Purchase ShippingPortal ESB
  33. 33. I’ve got a cheaper solution
  34. 34. To be clear – none of the examples represent in my opinion SOA The 4 tenets of SOA 1. Services are autonomous – Encapsulation & Cohesion at a bigger scale. – A service is autonomous if it doesn’t rely on other services to complete its job 2. Services have explicit boundaries – Services should not be too familiar with each other. There’s a clear boundary to where the responsibilities of each services starts and ends – Boundaries are made unclear when a service rely on other services to complete its job 3. Services share contract and schema, not class or type – Encapsulation improved. We need not worry about how things are implemented (languages or platform dependent information) 4. Service interaction is controlled by a policy – Controls under which rules/form of technical communication between services take place (e.g. using encryption incl. algorithm and keys)
  35. 35. What we’ve seen is just old EAI wine on new SOA bottles
  36. 36. WebServices and in general synchronous integration has nothing to do with real SOA Some call this pattern for SOA 1.0 to distinguish them selves from the old approach and the mess it causes
  37. 37. Layered Architectures typically leaves all orchestration to a central manager (ESB) where business processes are coordinated through spaghetti code (BPEL)
  38. 38. These BPEL processes typically break all Service encapsulation as they are data and feature envious This hurts our coupling an autonomy even further
  39. 39. What we have with classic layered SOA is a HARD COUPLED architecture
  40. 40. But what about reuse? • Layered SOA has typically been recommended because it increases reuse (at the expense of a hard coupled architecture) • The thesis is that the more we reuse, the faster we will be done • But the thesis rest on the false assumption that writing code is the most expensive part of a project – Not all code takes the same amount of time to write. Some code is very trivial while other code is very hard to write • The real money and time consumers on any project are: – Figuring out what the customer needs – The time it takes for the customer to figure what they really needed and the resulting rework – Meetings – Integration work, databases, webservices – Fix bugs – Debugging – Deployment – Test – Ship Faile d? Yes No The more dependencies we have, the worse and more expensive this becomes
  41. 41. Service reuse multiplies our direct and especially indirect dependencies which creates high coupling My new Service that wants to reuse other services Service that is going to be reused Reusable Service Reusable Service Reusable Service Reusable Service Reusable Service DB service Another Service that is going to be reused
  42. 42. Can’t we just avoid Services and make one big application?
  43. 43. Retail Domain Invoicing Subdomain Product Catalog Subdomain Shipping Subdomain Sales (Order) Subdomain Inventory Subdomain Pricing
  44. 44. One Domain Model to Rule them All? Combining subdomains into one big domain model increases complexity as everything tends to be connected to and depend on everything else
  45. 45. One big model to capture it ALL? …. ….. ….. …. ….. ….. …. ….. ….. …. ….. …..
  46. 46. Result: Our queries get messy
  48. 48. The database determines our scalability
  49. 49. And databases often scale poorly To solve the performance problems for a few, you have to upgrade it all
  50. 50. Do not design for locality and try to distribute Design for distribution, take advantage of locality
  51. 51. Alternatively we can use Read replicas Master Slave SlaveSlaveSlave You’re now eventually consistent
  52. 52. Or introduce caching… What’s your synchronization mechanism? ?
  53. 53. Many perspectives on data Customer Retail System Pricing Inventory Sales Product Unit Price Promotional Price Promotion End Date Stock Keeping Unit (SKU) Quantity On Hand (QOH) Location Code Price Quantity Ordered Name The lifecycle for Data is VERY important! Forecasting?
  54. 54. Smaller models & clear data ownership Sales Product ProductID Name Description Quantity Ordered … Sales Inventory Product ProductID SKU QOH Location Code … Inventory Pricing Product ProductID Unit Price Promotional Price … Pricing DDD: Bounded Context SOA: Service Retail System Shared Entity identity
  55. 55. A Service is • A technical authority for a specific business capability – very similar to Bounded Contexts • The owner of all the data and business rules that support this business capability – like Bounded Contexts A Service is equivalent to a Bounded Context
  56. 56. What a Service is NOT • A service with only functionality (and no data) is a FUNCTION – Like: check if order is valid • A service that only has data is a DATABASE – Like: CRUD of an entity This is typically seen in a lot of layered SOA usages where a function calls a function that calls a function that calls a database
  57. 57. Autonomy becomes the next (and biggest) problem Can you rely on others?
  58. 58. Data Layer Service Orchestration issue Data Access Layer Service Layer Interface/Display Layer System CSystem BSystem A UI Service Service DAL DAL DAL Schema Schema Schema Local transaction spanning all modules Service
  59. 59. Service autonomy Service B Service C Service A System X Service A System Y Service B Service C System X Slow/unreliable network Different SLA Slow system
  60. 60. Orchestration using synchronous communication and local transactions System CSystem BSystem A UI Service Service DAL DAL DAL Schema Schema Schema B:Service() call C:Service() call B:DAL() call A:Service() commit() Service
  61. 61. Solution for when going distributed? Can’t we just distributed transactions? (XA / 2 phase commit)
  62. 62. Distributed Transactions…. Oh my Sales system Sale Delivery system Deliveries Customer/CR M system Custome r SAP Bookkeeping Complete Purchase Transaction Coordinator Transactional Resource Prepare Phase Commit Phase 2 Phase Commit
  63. 63. What’s wrong with distributed transactions? • Transactions lock resources while active • Services are autonomous – Can’t be expected to finish within a certain time interval • Locking keeps other transactions from completing their job • Locking doesn’t scale • X Phase Commit is fragile by design
  64. 64. Orchestration using synchronous communication and without distributed transactions System CSystem BSystem A UI Service Service DAL DAL DAL Schema Schema Schema ServiceB:Service() call C:Service() call B:DAL() call A:Service() if (A:Call-Failed?) call A:Service() if (A:Call-Failed?) Wait-A-While() call A:Service() if (A:Call-Failed-Due-To-IO?) Save-We-Need-Check-If-Call-A-Succeded-After-All AND We-Need-To-Retry call C:Service and call B:DAL AND Tell-Customer-That-This-Operation-Perhaps-Went-Well if (A:Call-Went-Well) commit()
  65. 65. More of the same clearly can’t be The answer…
  66. 66. The more autonomous services are, the more loosely coupled they are. Loose coupling is what drives business agility – which is why we wanted SOA in the first place
  67. 67. Loose coupling Requires good encapsulation and high cohesion
  68. 68. Return of the SILOS!!!!
  69. 69. We need SMALL SILOS that SHARE very LITTLE with each other Sales Pricing InventoryShipping ✕ ✕ ✕
  70. 70. A silo/bounded-context contains all parts related to it DB Schema Domain Services Application Services Domain Model Aggregates , Entities, Value Objects, Events Integration Endpoints (REST, SOAP, Pub/Sub) User Interface Silo
  71. 71. A silos parts are not limited to deployment at a single location Subdomain User Interface Integration Endpoints (REST, SOAP, Pub/Sub) Domain Services Application Services Aggregates, Entities, Value Objects, Events DB Schema Units of individual deployment
  72. 72. What should our silos share? Bounded Context
  73. 73. Degrees of coupling UI UI Service ServiceData Data Events Events
  74. 74. Composite UI’s Retail System Sales Pricing Inventory Web/Application tier Background server tier Storage tier Composite UI UI Layered ApplicationData Access Layer Readmodel Writemodel ✕ ✕
  75. 75. Composite UI - example Context: Book ISBN-10 0-321-12521-5
  76. 76. Composite UI - HTML #DIV – BookReviews #DIV – BookImage #DIV – BookTitleAndAuthor #DIV – BookPricing #DIV – BookAvailability #DIV - Header #DIV - Body
  77. 77. Composite UI
  78. 78. The Service owns it UI in all Contexts and for all Composite UI’s Not just for HTML clients
  79. 79. Invoice Composite UI example InvoiceHeader Order:ShippingI nfo Invoice: NextInvoiceNumb er Invoice: Data and Due date Order: RelationInformation Order:Item- Qty Product:Ite m Product: Description Order: Item-Unit-Price Order: Item- Total- Price Order:Total Billing:Balance All Services participate at the UI level for each individual Item in the Order
  80. 80. When won’t a Composite UI make sense? For certain types of UI’s that built outside of the organization, e.g. customer specific netbank apps and mobile banking application we need to provide a synchronous integration API that will allow them to integrate with the business domains without having to coordinate with the business domains UI team. The integration API could be built using both a REST* and/or a SOAP approach. * It’s not such a distant thought to see a Composite UI part as a REST resource with a rendering matching the requested MimeType encoding.
  81. 81. That covers data presentation What about transactions?
  82. 82. Next Step – Event Driven Architecture
  83. 83. Why Events?
  84. 84. Synchronous calls lowers our Fault tolerance • When servers crashes • When databases are down • When deadlocks occurs in the database – Do you retry? With synchronous RPC style Services interaction we loose business data. There’s no automatic retry and idempotence is often an after though.
  85. 85. Cure: Messaging • Message based communication is asynchronous and thereby breaks temporal coupling • Messages can be exchanged between services over a message channel • The Message channel is responsible for delivering the message(s) to the relevant parties (consumers). If something goes wrong, the message will be put back on the Message channel (rollback) and will be resent later
  86. 86. Let’s make the implicit explicit! Old wisdom seems to have been forgotten. Let’s introduce: Business Events Which: • Signal that something has happened • Closely aligned to the Domain Model • Are handled by a messaging system • They are in the past tense: – CustomerBilled – ParcelShipped – CustomerCreated – ReviewCreated – CommentAdded – CommentDeleted
  87. 87. Business Events as Messages A Business Event can be transferred over a message channel, such as a Queue or a Topic, in many different formats
  88. 88. Business Event as XML Message <OrderWasAccepted> <CustomerId>50D1F244-ABBC-4EC7-BDCA-E4934C124A89</CustomerId> <OrderId>C199322A-01F1-4E56-918E-7A63529F8FA3</OrderId> <ShippingAddress> ... </ShippingAddress> <BillingAddress> ... </BillingAddress> <Items> <Item ProductId="4CD22C4B-600C-4477-B5BF-48ABDEE4DA61" Amount="100" AmountUnit="Pieces" UnitPrice="100,10" UnitCurrency="EUR"/> <Item ProductId="56E6BD19-660C-464A-9120-100DAF579855" Amount="10" AmountUnit="Litres" UnitPrice="56,95" UnitCurrency="CHF"/> </Items> </OrderWasAccepted>
  89. 89. Business Event as JSON Message { EventType: "OrderWasAccepted", CustomerId: "50D1F244-ABBC-4EC7-BDCA-E4934C124A89", OrderId: "C199322A-01F1-4E56-918E-7A63529F8FA3", ShippingAddress: { ... } BillingAddress: { ... } Items: [ { ProductId: "4CD22C4B-600C-4477-B5BF-48ABDEE4DA61", Amount: "100", AmountUnit: "Pieces", UnitPrice: "100,10", UnitCurrency: "EUR" }, { ProductId: "56E6BD19-660C-464A-9120-100DAF579855", Amount: "10", AmountUnit: "Litres", UnitPrice: "56,95", UnitCurrency: "CHF" } ] }
  90. 90. Business Events help us achieve autonomy, loose coupling and encapsulation • Encapsulation - because we don’t need to supply our services internal data (because no on else needs to know them) – internal data will only be accessible through the Services UI (which takes part in a Composite UI) • Events only need to explain WHAT happened and what it relates to and very few business data (which are typically only needed for replication scenarios)
  91. 91. Business Events Messages and Business Processes By publishing Events messages from our Services we can communicate with each other and also drive Business Processes
  92. 92. Using Business Events to drive Business Processes Sales Service Shipping Invoicing Sales Customers MessageChannel The sales fulfillment processing can now begin… Online Ordering System Web Shop (Composite UI) Invoicing Service Shipping Service Order Accepted Event
  93. 93. Business Events example Sales Service Order Accepted Invoicing Service Retail System Order Accepted Customer Billed MessageChannel We use the Order Accepted event message published from the Sales Service to drive the Invoicing/Billing of the customer. The billing part of the process also use Business Events, in this case Customer Billed Event, to indicate that its part of the process is completed. Because we use asynchronous messaging we can still accept orders in the sales service even though the invoicing services is down. The Order Accepted event message will remain in the Message Channel until the Invoicing Service is ready to process it.
  94. 94. Introducing Choreographed Processes Sales Service Order Accepted Invoicing Service Shipping Process Manager (Saga) Shipping Service Retail System Message Channel Order Accepted Order Accepted Customer Billed Customer Billed Order Ready for Shipment Order Ready for Shipment Works as a Finite State Machine (WorkFlow) handling the life cycle of Shipping and thereby forms a very central new Aggregate in the System
  95. 95. Process Managers • Process Managers are essential to the coordination and monitoring of long running processes. • They work as a Finite State Machines (WorkFlow) which handling the life cycle of Process (e.g. Shipping an Order) and thereby forms a very central new Aggregate in the System • They can include manual steps/person intervention • Sometimes these Process Managers belong naturally within a specific Subdomain and other times they are truly and thing by themselves and often form new Subdomains that way Companies often derive their competitive advantages from their Processes. A Process Manager allows you coordinate Business Processes on the basis of Events
  96. 96. Domain Events can also be used for data replication This is typically used for read-mostly master data that is needed in many business- domains. This way we avoid having to call a synchronous services to query the latest value(s)
  97. 97. Event Based Replication Subdomain Z Repositor y Repositor y Z Data Subdomain Y Service Repository Message channel Event Receiver Y Event Receiver and Local Cache Builder Y Data Cache Y Events Y Events
  98. 98. Questions? @jeppec on @TigerTeamDK on Twitter