Building Event-Driven Systems at Scale with Azure Cosmos DB

The most prominent shift we've observed in the field of distributed systems over the last decade is probably the transition from traditional synchronous integrated systems to message-driven means. This change is a logical response to the high demands for scalability, elasticity, and resilience within today's software systems. Regardless of how overwhelming the array of message-based options may seem, Azure Cosmos DB serves as an incredible solid backbone for most of them, whether utilized as a transactional outbox or a fully-fledged event store in the context of event-sourcing. By diving deeper into real-world experiences, we will discover the mechanics of how Azure Cosmos DB can guarantee nearly infinite scalability if you follow a few basic principles and patterns to efficiently model your data. While Azure Cosmos DB should not be considered as another silver bullet, this talk will teach you why you should certainly keep it as a prominent tool in your toolkit when it comes to building event-driven systems at scale.

Transcript

1. Building Event Driven Systems at Scale with Azure Cosmos DB Azure Cosmos DB Global User Group
2. We‘re all good! I N TR ODU C TI O N !
3. We‘re all good! I N TR ODU C TI O N !
4. We‘re all good! I N TR ODU C TI O N !
5. We‘re all good! I N TR ODU C TI O N !
6. > whoami Shahab Ganji Software Architecture Software Transformation .NET and C# enthusiast Embracing Change Telling dad jokes (Proudly) Code Artisan Lead Coding Architect MA IN F OC US ON T RI VI A @shahab-ganji @shahabganji @shahabganji
7. What is an Event Driven Architecture? I N TR ODU C TI O N Has three main components Software components execute in response to events Uses events to communicate Promotes loose coupling
8. Type of Events W H A T I S A N EV EN T D RI VE N A R C HI T EC T U RE ? ENTITY EVENT KEYED EVENT U NK EYED EV ENT Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. An entity is a unique thing and is keyed on the unique id of the thing Describes the properties and state of the entity at a given point in time Contains a key, but does not represent an entity Used for partitioning the stream of events to guarantee data locality within a single partition of an event stream Describes an event as a singular statement of a fact
9. EVENT SOURCING Captures every change to the state Provides full audit trail Easier handling of complex transactions Replay what has happened in the system Related Patterns W H A T I S A N EV EN T D RI VE N A R C HI T EC T U RE ? CQR S Separate read and write models Enables optimized performance and scalability https://lostechies.com/jimmybogard/2012/08/22/busting-some-cqrs-myths
10. Event Streams W H A T I S A N EV EN T D RI VE N A R C HI T EC T U RE Forward to EH Snapshot Email Service Inventory Service
11. Schema free, NoSQL Cloud Solution A Z U R E C OS MO S DB Globally Distributed Horizontally Scalable Provisioned throughput Multi model database
12. 3 Dimensions of scaling S C A L A BI LI TY https://microservices.io/articles/scalecube.html DATABASE PER APPLICATION Y axis – Functional Decomposition Scale by splitting different things 1 REPLICATION x axis – horizonal decomposition Scale by cloning 2 Sharding z axis – data partitioning Scale by splitting similar things 3
13. S C A L A BI LI TY Sharding Application A single logical database Nodes have different data Cluster of databases Application Application
14. S C A L A BI LI TY Sharding – Database Level PHYSICAL INSTANCES LOGICAL DATABASE
15. S C A L A BI LI TY Sharding – Database Level PHYSICAL INSTANCES LOGICAL DATABASE Application READ/WRITE READ/WRITE
16. Advantages S C A L A BI LI TY Each Server deals with a subset of data Improves transaction scalability Fault Isolation Cache Utilization Reduces Memory & I/O usage
17. Disadvantages S C A L A BI LI TY Increased application complexity Design a Partition Schema Re-partitioning Improper Traffic Distribution Performance Issues with Queries – Cross Partition
18. Containers, Partitions, Request Units A Z U R E C OS MO S DB 2 MB 20 GB 50 GB 50 GB LOGICAL PARTITION PHYSICAL PARTITION 10K RUs 10K RUs
19. Transaction Scope A Z U R E C OS MO S DB LOGICAL PARTITION TRANSACTION BATCH /partition-key: 123 LOGICAL PARTITION
20. Transaction Scope A Z U R E C OS MO S DB LOGICAL PARTITION TRANSACTION BATCH /partition-key: 123 LOGICAL PARTITION
21. Dig deeper! A Z U R E C OS MO S DB WEST EUROPE LEADER FOLLOWER FOLLOWER FOLLOWER
22. CAP theorem A Z U R E C OS MO S DB CONSISTENCY AVIALABILTY PARTITION TOLERANCE
23. Globally Distributed A Z U R E C OS MO S DB
24. Globally Distributed A Z U R E C OS MO S DB WEST EUROPE LEADER FOLLOWER FOLLOWER FORWARDER SOUTH CENTRAL US LEADER FORWARDER FOLLOWER FOLLOWER
25. Strong A Z U R E C OS MO S DB CONSISTENCY LEVELS
26. Bounded Staleness A Z U R E C OS MO S DB CONSISTENCY LEVELS
27. Session A Z U R E C OS MO S DB CONSISTENCY LEVELS
28. Consistent Prefix A Z U R E C OS MO S DB CONSISTENCY LEVELS
29. Eventual A Z U R E C OS MO S DB CONSISTENCY LEVELS
30. Change Feed A Z U R E C OS MO S DB
31. Change Feed A Z U R E C OS MO S DB PHYSICAL PARTITION 1 LOGICAL DATABASE 1 1 1 1 1 1 2 2 2 2 1 1 3 3 2 2 2 2 PHYSICAL PARTITION 2
32. Change Feed A Z U R E C OS MO S DB FeedIterator<Customer> iteratorForPartitionKey = _container.GetChangeFeedIterator<Customer>( ChangeFeedStartFrom.Beginning(FeedRange.FromPartitionKey(new PartitionKey("stream-id"))), ChangeFeedMode.LatestVersion);
33. Change Feed A Z U R E C OS MO S DB
34. SHOW ME THE CODE
35. Summary BU I LD I N G EV EN T D RI VE N A R C HI T EC T U RE W IT H A Z U RE C O S MOS DB E DA AZURE COS MOS DB Global distribution Guaranteed Performance and SLAs Change Feed Automatic Indexing Multi-model and Multi-API support High volume of events Real-time processing Scalability is a primary concern Requires immediate reaction Separate read and write models Audit log Tracking state changes are critical C QR S E V E N T S OU R CI N G
36. Get in touch Shahab aka. Saeed Ganji Code Artisan!
37. Q&A