Publishing_Events_Using_Local_Transactions.md

Publishing Events Using Local Transactions

• Motivation
• Solution
  • Concepts
  • Implementation
• Pros & Cons
  • Pros
  • Cons
• Consideration
• When To Use
• References

Motivation

• How to update the database and publish a message (event) atomically (either all success, or all fail).
• Cannot use distributed transaction pattens to solve this problem.
  • Many modern message brokers don’t support distributed transactions.

Solution

Concepts

• Uses the event table to store the events need to be published.
• Bundles the operation of inserting new record into a business entity table and the operation of inserting a new event to the event table as a single local database transaction.
• Event publisher publishes events by polling the event table periodically.

Implementation

Pros & Cons

Pros

• The atomicity of updating the database and publishing a message (event) is guaranteed.

Cons

• Frequently polling the database can be expensive.
• Tricky to publish events in order.
• Not all NoSQL databases support this pattern.

Consideration

Topic	Consideration	Possible Solution Options
Error handling	The event publisher may publish a event more than once (For example, crash after publishing an event but before marking the corresponding event record as published).	Consider to let the event consumer be idempotent by recording the IDs of the events that it has already processed.

When To Use

References

• Book: Chris R.(2018). Chapter 3. Interprocess communication in a microservice architecture, Microservices Patterns (pp. 65-109). Manning Publications
• Book: Kasun I., Prabath S.(2018). Chapter 5: Data Management, Microservices for the Enterprise (pp. 125-150). Apress
• Web Article: Pattern: Transactional outbox | https://microservices.io/patterns/data/transactional-outbox.html
• Web Article: Event-Driven Data Management for Microservices | https://www.nginx.com/blog/event-driven-data-management-microservices/