Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Event Sourcing avec Azure, quelle base de donné...

Event Sourcing avec Azure, quelle base de données choisir

Global Azure Bootcamp 2023 (Rennes)

Sylvain PONTOREAU

May 12, 2023
Tweet

More Decks by Sylvain PONTOREAU

Other Decks in Programming

Transcript

  1. 🕺 Why I am talking about this topic ? I

    have been diving deep in it for a long time!
  2. 💁 The Event Sourcing pattern “Every change to the state

    of an application is captured in an event” https://martinfowler.com/eaaDev/EventSourcing.html
  3. 📨 Event Stream Sequence of events = Event Stream Event

    Stream Event Event Event Event Event Event Event Event Event Timeline
  4. 📨 Event Stream Load from stream Event Event Event Event

    Event Event Event Event Event Load Reduce State
  5. 📨 Event Stream Subscribe Event Event Event Event Event Event

    Event Event Event Append Event Subscribe
  6. 📰 Read operations Load Reduce State A little bit expensive

    for read operations… … and not really great for querying :(
  7. 📽 Projections Using subscribe to build a queryable representation of

    the application state (aka Projections) Subscribe Projections Projector ℹ A good database for event sourcing must have a data streaming mechanism
  8. ✋ Keep that in mind App DB Projections Append Subscribe

    App Projections Append Publish ✅ ❌ DB ❗ Even if you’re using a Message Broker or an Event Bus
  9. ⚠ Eventually consistent CAP Theorem (Eric Brewer) Consistency Availability Partition

    tolerance All system nodes needs to manipulate exactly the same data at the same time Requests must succeed Node failure must not create a complete system failure A distributed system can only guarantee 2 constraints at the same time, not 3…
  10. 🔃 CQRS pattern What is CQRS? • A pattern designed

    by Greg Young • Command Query Responsibility Segregation • Read != Write (different needs) Query A read operation Command Validation Business logic Persistence CQRS suit perfectly well with Event Sourcing!
  11. 🔃 CQRS pattern GetBalance Query Deposite Command GetAllOperation Query Withdraw

    Command … … Dispatcher Read DB Dispatcher QueryHandler CommandHandler Event Store Projection Projection Aggregate State Events Behaviors
  12. 🗃 PostgreSQL - Table CREATE TABLE "EVENT_STREAM" ( "streamId" VARCHAR

    NOT NULL, "createdAt" VARCHAR NOT NULL, "type" VARCHAR NOT NULL, "version" INTEGER NOT NULL, "payload" JSONB, PRIMARY KEY ("streamId", "createdAt", "version") ); CREATE INDEX ON "EVENT_STREAM" ("streamId");
  13. 🗃 PostgreSQL - Notify CREATE FUNCTION notify_new_event() RETURNS TRIGGER AS

    $trigger$ BEGIN PERFORM pg_notify('new_event', row_to_json(NEW)::text); RETURN NULL; END; $trigger$ LANGUAGE plpgsql VOLATILE COST 100;
  14. 🗃 PostgreSQL - Trigger CREATE TRIGGER notify_new_event AFTER INSERT ON

    "EVENT_STREAM" FOR EACH ROW EXECUTE PROCEDURE notify_new_event();
  15. 🗃 PostgreSQL - Rules CREATE FUNCTION THROW_WHEN_TYRING_TO_UPDATE_EVENT_STREAM() RETURNS VOID AS

    $$ BEGIN RAISE EXCEPTION 'Event Stream can not be updated'; END; $$ LANGUAGE plpgsql VOLATILE; CREATE RULE "PREVENT_DELETE_ON_EVENT_STREAM" AS ON DELETE TO "EVENT_STREAM" DO INSTEAD SELECT THROW_WHEN_TYRING_TO_UPDATE_EVENT_STREAM(); CREATE RULE "PREVENT_UPDATE_ON_EVENT_STREAM" AS ON UPDATE TO "EVENT_STREAM" DO INSTEAD SELECT THROW_WHEN_TYRING_TO_UPDATE_EVENT_STREAM();
  16. 🗃 PostgreSQL - Load events SELECT "type", "payload", "version" FROM

    "EVENT_STREAM" WHERE "streamId" = $1 ORDER BY "createdAt", "version"
  17. 🗃 PostgreSQL - Append events INSERT INTO "EVENT_STREAM" ( "streamId",

    "createdAt", "type", "version", "payload" ) VALUES ($1, $2, $3, ( SELECT COUNT("streamId") + 1 FROM "EVENT_STREAM" WHERE "streamId" = $5 ), $4 )
  18. 🗃 PostgreSQL - Conclusion ✅ ❌ • Simple to implement

    • One impl for all languages • On premise if needed • DB Tuning • You must know how to do it • Distributed • No acknowledgement
  19. 🌐 Cosmos DB - Load stream const endpoint = "https://your-account.documents.azure.com";

    const key = "<database account masterkey>"; const client = new CosmosClient({ endpoint, key }); const container = client.database("EventStore").container("EventStream"); const querySpec: SqlQuerySpec = { query: `SELECT eventStream.eventType, eventStream.payload, eventStream.version, FROM eventStream WHERE eventStream.streamId = @streamId ORDER BY eventStream.createdAt, eventStream.version`, parameters: [{ name: "@streamId", value: "<stream identifier>" } ] }; const { resources } = await container.items.query(querySpec).fetchAll();
  20. 🌐 Cosmos DB - Append events const endpoint = "https://your-account.documents.azure.com";

    const key = "<database account masterkey>"; const client = new CosmosClient({ endpoint, key }); const container = client.database("EventStore").container("EventStream"); await container.items.create({ streamId: "<stream identifier>", createdAt: "<event creation date (ISO 8601)", type: "<event type>", version: <stream version>, payload: { //… } });
  21. 🌐 Cosmos DB - Conclusion ✅ ❌ • Straightforward •

    Serverless • Schemaless • No rules • Version computed in code
  22. 🌀 Event Store DB - Load stream const client =

    new EventStoreDBClient({ endpoint: "<database endpoint>", }); const streamName = "<stream identifier>"; const events = client.readStream(streamName, { fromRevision: START, direction: FORWARDS, maxCount: 10, });
  23. 🌀 Event Store DB - Append events const client =

    new EventStoreDBClient({ endpoint: "<database endpoint>", }); const streamName = "<stream identifier>"; const event = jsonEvent({ type: "<event type>", data: { //… }, }); await client.appendToStream(streamName, [event]);
  24. 🌀 Event Store DB ✅ • Out of the box

    • State of the art • Built in projection • On premise if needed • Manage outside of Azure… • Not too much customization • Only gRPC client ❌