-
Notifications
You must be signed in to change notification settings - Fork 21
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
docs: [FC-0074] add ADR with event design suggestions #438
base: main
Are you sure you want to change the base?
Changes from all commits
684283f
724e87f
81ea9f3
66846f4
dc51ca3
0f86b3d
c71bb7f
608bd43
e8c16da
361d478
63435f7
41cda0f
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,58 @@ | ||
16. Event Design Practices | ||
########################### | ||
|
||
Status | ||
------ | ||
|
||
Draft | ||
|
||
Context | ||
------- | ||
|
||
It is important to follow standards to ensure that the events are consistent, maintainable, and reusable. The design of the events should be self-descriptive, self-contained, and provide enough information for consumers to understand the message. This ADR aims to provide a set of suggested practices for designing Open edX events. | ||
|
||
Decision | ||
-------- | ||
|
||
We have compiled a list of suggested practices taken from the following sources: | ||
|
||
- `Event-Driven Microservices`_ | ||
- `Event-Driven article`_ | ||
- `Thin Events - The lean muscle of event-driven architecture`_ | ||
|
||
These are the practices that we recommend reviewing and following when designing an Open edX Event and contributing to the library. The goal is to implement events that are consistent with the architecture, reusable, and maintainable over time. | ||
|
||
Event Purpose and Content | ||
~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
||
- An event should describe as accurately as possible what happened (what) and why it happened (why). It must contain enough information for consumers to understand the message. For instance, if an event is about a user enrollment, it should contain the user's data, the course data, and the enrollment status and the event should be named accordingly. | ||
- Avoid immediately contacting the source service to retrieve additional information. Instead, consider adding the necessary information to the event payload by managing the granularity of the event. If the event requires additional information, consider adding a field to the event that contains the necessary information. This will reduce the number of dependencies between services and make the event more self-contained. | ||
- Keep the event size small. Avoid adding unnecessary information to the event. If the information is not necessary for consumers to react to the event, consider removing it. | ||
- Avoid adding flow-control information or business logic to events. Events should be solely a representation of what took place. If a field is necessary to control the behavior of the consumer, consider moving it to the consumer side. If adding additional data to the event is absolutely necessary document the reasoning behind it and carefully study the use case and implications. | ||
|
||
Responsibility and Granularity | ||
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
||
- Design events with a single responsibility in mind. Each event should represent a single action or fact that happened in the system. If an event contains multiple actions, consider splitting it into multiple events. For instance, if the course grade is updated to pass or fail, there should be two events: one for the pass action and another for the fail action. | ||
- Manage the granularity of the event so it is not too coarse (generic with too much information) or too fine-grained (specific with too little information). When making a decision on the granularity of the event, start with the minimum required information for consumers to react to the event and add more information as needed with enough justification. If necessary, leverage API calls from the consumer side to retrieve additional information but always consider the trade-offs of adding dependencies with other services. | ||
- Ensure that the triggering logic is consistent and narrow. For instance, if an event is triggered when a user enrolls in a course, it should be trigger only when the user enrolls in a course in all ways possible to enroll in a course. If the event is triggered when a user enrolls in a course through the API, it should also be triggered when the user enrolls in a course through the UI. | ||
|
||
Event Structure and Clarity | ||
~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
||
- Use appropriate data types and formats for the event fields. Don't use generic data types like strings for all fields. Use specific data types like integers, floats, dates, or custom types when necessary. | ||
- Avoid ambiguous data fields or fields with multiple meaning. For instance, if an event contains a field called ``status`` it should be clear what the status represents. If the status can have multiple meanings, consider splitting the event into multiple events or adding a new field to clarify the status. | ||
|
||
Consumer-Centric Design | ||
~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
||
- When designing an event, consider the consumers that will be using it. What information do they need to react to the event? What data is necessary for them to process the event? | ||
- Design events carefully from the start to minimize breaking changes for consumers, although it is not always possible to avoid breaking changes. | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. will it be necessary to version events to handle event changes, or what is the plan for handling event changes? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Events are versioned by definition, see this ADR for more info. As for the evolution of the events schema, this other ADR describes what the behavior is supposed to be: https://github.com/openedx/openedx-events/blob/main/docs/decisions/0006-event-schema-serialization-and-evolution.rst#decision, although according to this comment the ADR might be outdated -- I'll be working on updating it. The reality is that we haven't needed to change an event definition in any way that's breaking. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. @robrap: Can you help us figure out what needs to change from the ADR-0006? I could do it, but I need more context to do it effectively. |
||
|
||
Some of these practices might not be applicable to all events, but they are a good starting point to ensure that the events are consistent and maintainable over time. So, design the event so it is small, well-defined and only contain relevant information. | ||
|
||
In addition to these practices, review the Architectural Decision Records (ADRs) related to events to understand the naming, versioning, payload, and other practices that are specific to Open edX events. | ||
|
||
.. _Event-Driven Microservices: https://www.oreilly.com/library/view/building-event-driven-microservices/9781492057888/ | ||
.. _Event-Driven article: https://martinfowler.com/articles/201701-event-driven.html | ||
.. _Thin Events - The lean muscle of event-driven architecture: https://www.thoughtworks.com/insights/blog/architecture/thin-events-the-lean-muscle-of-event-driven-architecture |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I agree with the idea of avoid generic events, however, we should also avoid split events like course_passed or course_failed when we could use just course_completed with an status.
could we include a practical example on an appropiate level of granularity?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm interested in understanding why having a
status
is better. I think that in this case, it'd be easier to consume more granular events than leave the responsibility to the consumer to figure out whether the student passed or failed based on the status.There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I understand your perspective, and I think both approaches have valid use cases depending on the context, let me elaborate on why I suggested using course_completed with a status field and we can discuss further to find a balance
If the event represents a single conceptual action for example: completing a course, having one event like course_completed with a clear status passed, failed, etc. could simplify the producer's logic and reduce event proliferation and for consumers, interpreting the status field is relatively straightforward if it's well-documented and includes only a few well-defined values, however if different consumers need to handle passed and failed cases in significantly different ways, separate those events might reduce complexity for them but emitting separate events like course_passed and course_failed could also create challenges such as needing to ensure mutual exclusivty.
So in cases like this I think We could start with course_completed and a status field, ensuring it is well-documented and strictly validted and if in the future we observe a clear need for distinct event flows We could introduce the more granular events without breaking existing consumers.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
In the example we're using, I still think it'd be more straightforward to send smaller and more specific events. As I see it, course completion and grade passing would be two different critical facts happening in the system; therefore, they should be independent. This is more of a question of what consumers would want with a course completion event or a course passing status change.
In any case, these are the only suggestions that should be evaluated for each case. We currently have an event called
COURSE_PASSING_STATUS_UPDATED
which uses a similar flow control status field calledis_passing
. In that case, given that the status is directly related to the event, using a status field is acceptable. What do you think? Should we maybe use a rule of thumb indicating that an event can be split into more events, given that they could communicate more facts of the system?