Practical Change Data Streaming Use Cases With Apache Kafka and Debezium (QCon San Francisco 2019)

Debezium (noun | de·be·zi·um | /dɪ:ˈbɪ:ziːəm/) - Secret Sauce for Change Data Capture

Apache Kafka is a highly popular option for asynchronous event propagation between microservices. Things get challenging though when adding a service’s database to the picture: How can you avoid inconsistencies between Kafka and the database?

Enter change data capture (CDC) and Debezium. By capturing changes from the log files of the database, Debezium gives you both reliable and consistent inter-service messaging via Kafka and instant read-your-own-write semantics for services themselves.

In this session you’ll see how to leverage CDC for reliable microservices integration, e.g. using the outbox pattern, as well as many other CDC applications, such as maintaining audit logs, automatically keeping your full-text search index in sync, and driving streaming queries. We’ll also discuss practical matters, e.g. HA set-ups, best practices for running Debezium in production on and off Kubernetes, and the many use cases enabled by Kafka Connect's single message transformations.

Transcript

1. Practical Change Data Streaming Use Cases Practical Change Data Streaming Use Cases With Apache Kafka and Debezium With Apache Kafka and Debezium Gunnar Morling Gunnar Morling Software Engineer @gunnarmorling 1
2. DATA 2
5. The Issue with Dual Writes What's the problem? Change data capture to the rescue! CDC Use Cases & Patterns Replication Audit Logs Microservices Practical Matters Deployment Topologies Running on Kubernetes Single Message Transforms 1 2 3 5
6. Gunnar Morling Gunnar Morling Open source software engineer at Red Hat Debezium Hibernate Spec Lead for Bean Validation 2.0 Other projects: Deptective, MapStruct Java Champion #CDCUseCases @gunnarmorling 6
7. A Common Problem A Common Problem Updating Multiple Resources Updating Multiple Resources @gunnarmorling Database Order Service #CDCUseCases 7
8. A Common Problem A Common Problem Updating Multiple Resources Updating Multiple Resources @gunnarmorling Cache Database Order Service #CDCUseCases 8
9. A Common Problem A Common Problem Updating Multiple Resources Updating Multiple Resources @gunnarmorling Cache Database Order Service Search Index #CDCUseCases 9
10. A Common Problem A Common Problem Updating Multiple Resources Updating Multiple Resources @gunnarmorling Order Service Cache Database Search Index 10 “ Friends Don't Let Friends Do Dual Writes #CDCUseCases
11. A Better Solution A Better Solution Streaming Change Events From the Database Streaming Change Events From the Database @gunnarmorling Order Service #CDCUseCases 11
12. A Better Solution A Better Solution Streaming Change Events From the Database Streaming Change Events From the Database @gunnarmorling Order Service C C U C U U D C C - Create U - Update D - Delete 12 Change Data Capture #CDCUseCases
13. A Better Solution A Better Solution Streaming Change Events From the Database Streaming Change Events From the Database @gunnarmorling Order Service 13 C C U C U U D C C - Create U - Update D - Delete Change Data Capture #CDCUseCases
14. Change Data Capture Change Data Capture With Debezium With Debezium 14
15. Debezium Debezium Change Data Capture Platform Change Data Capture Platform CDC for multiple databases Based on transaction logs Snapshotting, Filtering etc. Fully open-source, very active community Via Apache Kafka or embedded Many production deployments (e.g. WePay, Convoy, JW Player, Usabilla, BlaBlaCar etc.) @gunnarmorling #CDCUseCases 15
16. Debezium Connectors Debezium Connectors MySQL Postgres MongoDB SQL Server Cassandra (Incubating) Oracle (Incubating, based on XStream) Possible future additions DB2? MariaDB? @gunnarmorling #CDCUseCases 16
17. Meme idea: Robin Moffatt 17
18. Log- vs. Query-Based CDC Log- vs. Query-Based CDC @gunnarmorling Query-Based Log-Based All data changes are captured - + No polling delay or overhead - + Transparent to writing applications and models - + Can capture deletes and old record state - + Installation/Configuration + - #CDCUseCases 18
19. { "before": null, "after": { "id": 1004, "first_name": "Anne", "last_name": "Kretchmar", "email": "annek@noanswer.org" }, "source": { "name": "dbserver1", "server_id": 0, "ts_sec": 0, "file": "mysql­bin.000003", "pos": 154, "row": 0, "snapshot": true, "db": "inventory", "table": "customers" }, "op": "c", "ts_ms": 1486500577691 } Change Event Structure Change Event Structure Key: Primary key of table Value: Describing the change event Old row state New row state Metadata Serialization formats: JSON Avro @gunnarmorling #CDCUseCases 19
20. The Issue with Dual Writes What's the problem? Change data capture to the rescue! 1 2 3 CDC Use Cases & Patterns Replication Audit Logs Microservices Practical Matters Deployment Topologies Running on Kubernetes Single Message Transforms 20
21. @gunnarmorling CDC – "Liberation for Your Data" CDC – "Liberation for Your Data" #CDCUseCases 21
22. @gunnarmorling Postgres MySQL Apache Kafka Data Replication Data Replication Zero-Code Streaming Pipelines Zero-Code Streaming Pipelines #CDCUseCases 22
23. @gunnarmorling Postgres MySQL Apache Kafka Kafka Connect Kafka Connect Data Replication Data Replication Zero-Code Streaming Pipelines Zero-Code Streaming Pipelines #CDCUseCases 23
24. @gunnarmorling Postgres MySQL Apache Kafka Kafka Connect Kafka Connect DBZ PG DBZ MySQL Data Replication Data Replication Zero-Code Streaming Pipelines Zero-Code Streaming Pipelines #CDCUseCases 24
25. @gunnarmorling Postgres MySQL Kafka Connect Kafka Connect Apache Kafka DBZ PG DBZ MySQL Elasticsearch ES Connector 25 Data Replication Data Replication Zero-Code Streaming Pipelines Zero-Code Streaming Pipelines #CDCUseCases
26. @gunnarmorling Postgres MySQL Kafka Connect Kafka Connect Apache Kafka DBZ PG DBZ MySQL Elasticsearch ES Connector 26 JDBC Connector ES Connector Data Warehouse Data Replication Data Replication Zero-Code Streaming Pipelines Zero-Code Streaming Pipelines #CDCUseCases
27. @gunnarmorling Postgres MySQL Kafka Connect Kafka Connect Apache Kafka DBZ PG DBZ MySQL Elasticsearch ES Connector 27 JDBC Connector ES Connector ISPN Connector Infinispan Data Replication Data Replication Zero-Code Streaming Pipelines Zero-Code Streaming Pipelines Data Warehouse #CDCUseCases
28. @gunnarmorling Data Replication Data Replication Low-Latency Streaming Pipelines Low-Latency Streaming Pipelines #CDCUseCases https://medium.com/convoy-tech/ 28
29. @gunnarmorling Auditing Auditing Source DB Kafka Connect Apache Kafka DBZ Customer Events CRM Service #CDCUseCases 29
30. @gunnarmorling Auditing Auditing Source DB Kafka Connect Apache Kafka DBZ Customer Events CRM Service Id User Use Case tx-1 Bob Create Customer tx-2 Sarah Delete Customer tx-3 Rebecca Update Customer "Transactions" table 30 #CDCUseCases
31. @gunnarmorling Auditing Auditing Source DB Kafka Connect Apache Kafka DBZ Customer Events Transactions CRM Service Id User Use Case tx-1 Bob Create Customer tx-2 Sarah Delete Customer tx-3 Rebecca Update Customer "Transactions" table 31 #CDCUseCases
32. @gunnarmorling Auditing Auditing Source DB Kafka Connect Apache Kafka DBZ Customer Events Transactions CRM Service Kafka Streams 32 Id User Use Case tx-1 Bob Create Customer tx-2 Sarah Delete Customer tx-3 Rebecca Update Customer "Transactions" table #CDCUseCases
33. @gunnarmorling Auditing Auditing Source DB Kafka Connect Apache Kafka DBZ Customer Events Transactions CRM Service Kafka Streams 33 Id User Use Case tx-1 Bob Create Customer tx-2 Sarah Delete Customer tx-3 Rebecca Update Customer "Transactions" table Enriched Customer Events #CDCUseCases
34. @gunnarmorling Auditing Auditing { "before": { "id": 1004, "last_name": "Kretchmar", "email": "annek@example.com" }, "after": { "id": 1004, "last_name": "Kretchmar", "email": "annek@noanswer.org" }, "source": { "name": "dbserver1", "table": "customers", "txId": "tx­3" }, "op": "u", "ts_ms": 1486500577691 } Customers #CDCUseCases 34
35. @gunnarmorling { "before": { "id": 1004, "last_name": "Kretchmar", "email": "annek@example.com" }, "after": { "id": 1004, "last_name": "Kretchmar", "email": "annek@noanswer.org" }, "source": { "name": "dbserver1", "table": "customers", "txId": "tx­3" }, "op": "u", "ts_ms": 1486500577691 } { "before": null, "after": { "id": "tx­3", "user": "Rebecca", "use_case": "Update customer" }, "source": { "name": "dbserver1", "table": "transactions", "txId": "tx­3" }, "op": "c", "ts_ms": 1486500577691 } Transactions Customers { "id": "tx­3" } #CDCUseCases 35
36. { "id": "tx­3" } { "before": { "id": 1004, "last_name": "Kretchmar", "email": "annek@example.com" }, "after": { "id": 1004, "last_name": "Kretchmar", "email": "annek@noanswer.org" }, "source": { "name": "dbserver1", "table": "customers", "txId": "tx­3" }, "op": "u", "ts_ms": 1486500577691 } Transactions Customers @gunnarmorling #CDCUseCases { "before": null, "after": { "id": "tx­3", "user": "Rebecca", "use_case": "Update customer" }, "source": { "name": "dbserver1", "table": "transactions", "txId": "tx­3" }, "op": "c", "ts_ms": 1486500577691 } 36
37. @gunnarmorling { "before": { "id": 1004, "last_name": "Kretchmar", "email": "annek@example.com" }, "after": { "id": 1004, "last_name": "Kretchmar", "email": "annek@noanswer.org" }, "source": { "name": "dbserver1", "table": "customers", "txId": "tx­3", "user": "Rebecca", "use_case": "Update customer" }, "op": "u", "ts_ms": 1486500577691 } Enriched Customers Auditing Auditing #CDCUseCases 37
38. @gunnarmorling @Override public KeyValue<JsonObject, JsonObject> transform(JsonObject key, JsonObject value) { boolean enrichedAllBufferedEvents = enrichAndEmitBufferedEvents(); if (!enrichedAllBufferedEvents) { bufferChangeEvent(key, value); return null; } KeyValue<JsonObject, JsonObject> enriched = enrichWithTxMetaData(key, value); if (enriched == null) { bufferChangeEvent(key, value); } return enriched; } Auditing Auditing Non-trivial join implementation no ordering across topics need to buffer change events until TX data available bit.ly/debezium-auditlogs #CDCUseCases 38
39. Microservice Microservice CDC Patterns CDC Patterns 39
40. @gunnarmorling Order Item Stock App Local DB Local DB Local DB App App 40 Item Changes Stock Changes Microservice Architectures Microservice Architectures Data Synchronization Data Synchronization Propagate data between different services without coupling Each service keeps optimised views locally #CDCUseCases
41. Source DB Kafka Connect Apache Kafka DBZ Order Events Credit Worthiness Check Events Outbox Pattern Outbox Pattern Separate Events Table Separate Events Table @gunnarmorling Order Service Shipment Service 41 Customer Service Orders Outbox #CDCUseCases
42. Source DB Kafka Connect Apache Kafka DBZ Order Events Credit Worthiness Check Events Outbox Pattern Outbox Pattern Separate Events Table Separate Events Table @gunnarmorling Order Service Shipment Service Customer Service 42 Id AggregateType AggregateId Type Payload ec6e Order 123 OrderCreated { "id" : 123, ... } 8af8 Order 456 OrderDetailCanceled { "id" : 456, ... } 890b Customer 789 InvoiceCreated { "id" : 789, ... } "Outbox" table bit.ly/debezium-outbox-pattern Orders Outbox #CDCUseCases
43. Strangler Pattern Strangler Pattern Migrating from Monoliths to Microservices Migrating from Monoliths to Microservices https://martinfowler.com/bliki/StranglerFigApplication.html @gunnarmorling #CDCUseCases 43
44. @gunnarmorling Customer Strangler Strangler Pattern Pattern #CDCUseCases 44
45. @gunnarmorling Router CDC Customer Customer' 45 Reads/ Writes Reads Strangler Strangler Pattern Pattern Transformation #CDCUseCases
46. @gunnarmorling Router CDC Customer 46 Reads/ Writes Reads/ Writes CDC Strangler Strangler Pattern Pattern #CDCUseCases
47. The Issue with Dual Writes What's the problem? Change data capture to the rescue! 1 3 2 CDC Use Cases & Patterns Replication Audit Logs Microservices Practical Matters Deployment Topologies Running on Kubernetes Single Message Transforms 47
48. @gunnarmorling Deployment Topologies Deployment Topologies Basic Set-Up Basic Set-Up CDC #CDCUseCases 48
49. Deployment Topologies Deployment Topologies Database High Availability Database High Availability @gunnarmorling CDC #CDCUseCases 49
50. Deployment Topologies Deployment Topologies Database High Availability Database High Availability @gunnarmorling CDC #CDCUseCases 50
51. Deployment Topologies Deployment Topologies Automatic Fail-over Automatic Fail-over @gunnarmorling HA Proxy CDC #CDCUseCases 51
52. Deployment Topologies Deployment Topologies Automatic Fail-over Automatic Fail-over @gunnarmorling HA Proxy CDC #CDCUseCases 52
53. Deployment Topologies Deployment Topologies Can't Change Binlog Mode? Can't Change Binlog Mode? @gunnarmorling CDC Primary Secondary #CDCUseCases 53
54. Deployment Topologies Deployment Topologies High Availability for Connectors High Availability for Connectors @gunnarmorling CDC Deduplicator CDC 54 #CDCUseCases
55. apiVersion: "kafka.strimzi.io/v1alpha1" kind: "KafkaConnector" metadata: name: "inventory­connector" labels: connect­cluster: my­connect­cluster spec: class: i.d.c.p.PostgresConnector tasksMax: 1 config: database.hostname: "postgres", database.port: "5432", database.user: "bob", database.password: "secret", database.dbname : "prod", database.server.name: "dbserver1", schema.whitelist: "inventory" Running on Kubernetes Running on Kubernetes Deployment via Operators Deployment via Operators YAML-based custom resource definitions for Kafka/Connect clusters, topics etc. Operator applies configuration Advantages Automated deployment and scaling Simplified upgrading Portability across clouds @gunnarmorling #CDCUseCases 55
56. Running on Kubernetes Running on Kubernetes Operating Kafka Connect Operating Kafka Connect Distributed mode Offsets stored in Kafka Configuration via REST Single node: no re-balancing issues (< Apache Kafka 2.3) Single connector: health checks based on REST API Fight duplication: Jsonnet templates @gunnarmorling // a database + connector per tenant { "name": "inventory­connector", "config": { "connector.class": "i.d.c.p.PostgresConnector", "tasks.max": "1", "database.hostname": "postgres", "database.port": "5432", "database.user": "bob", "database.password": "secret", "database.dbname" : std.extVar('tenant'), "database.server.name": std.extVar('tenant'), "schema.whitelist": "inventory" } } #CDCUseCases 56
57. Single Message Transformations Single Message Transformations The Swiss Army Knife of Kafka Connect The Swiss Army Knife of Kafka Connect Format conversions Time/date fields Extract new row state Aggregate sharded tables to single topic Keep compatibility with existing consumers @gunnarmorling #CDCUseCases © Emilian Robert Vicol https://flic.kr/p/c8s6Y3 57
58. Single Message Transformations Single Message Transformations Externalizing Large Column Values Externalizing Large Column Values @gunnarmorling DBZ Amazon S3 #CDCUseCases 58
59. Single Message Transformations Single Message Transformations Externalizing Large Column Values Externalizing Large Column Values @gunnarmorling DBZ Amazon S3 { "before": { ... }, "after": { "id": 1004, "last_name": "Kretchmar", "email": "annek@noanswer.org", "image": "imgs­<offset>­after" }, ... } #CDCUseCases 59
60. Takeaways Takeaways Change Data Capture – Liberation for your data! Enabling use cases such as replication, streaming queries, maintaining CQRS read models etc. Microservices: outbox and strangler patterns Debezium: open-source CDC for a growing number of databases @gunnarmorling #CDCUseCases “ Friends Don't Let Friends Do Dual-Writes 60
61. DATA 61
62. Resources Resources Website: Strimzi (Kafka on Kubernetes) Latest news: @debezium debezium.io debezium.io/documentation/online-resources debezium.io/blog strimzi.io @gunnarmorling #CDCUseCases 62
63. gunnar@hibernate.org @gunnarmorling @gunnarmorling Q&A #CDCUseCases 63
64. Outlook: View Materialization Outlook: View Materialization Awareness of Transaction Boundaries Awareness of Transaction Boundaries Topic with BEGIN/END markers Enable consumers to buffer all events of one transaction @gunnarmorling { "transactionId" : "tx­123", "eventType" : "begin transaction", "ts_ms": 1486500577125 } { "transactionId" : "tx­123", "ts_ms": 1486500577691, "eventType" : "end transaction", "eventCount" : [ { "name" : "dbserver1.inventory.Order", "count" : 1 }, { "name" : "dbserver1.inventory.OrderLine", "count" : 5 } ] } #CDCUseCases BEGIN END 64