Scope of "sequence" extension

[duglin]

The CE sequence extension:

... defines an attribute that can be included within a CloudEvent to describe the position of an event in the ordered sequence of events produced by a unique event source.

Notice the phrase "unique event source".

It also says:

Produces and consumers are free to define an out-of-band agreement on the semantic meaning, or valid values, for the attribute.

Which provides a lot of freedom for how this extension can be used.

Does the first quote imply that each sequence is scoped to the source attribute? Meaning source+sequence MUST be unique and the sequence comparisons are limited to matching source values? Or, in terms of a concrete example, does the 2nd quote above allow for implementations to compare (order) sequence values even when the source values differ? E.g. across all events received by a consumer, as long as the out-of-band agreement specifies such a rule.

If we really intended for sequence to be scoped by source then we should probably be more explicit about it in the spec.

@deissnerk let me know if I misrepresented your questions.

/cc @jskeet

[yordis]

The spec does specify that "unique event source". In case that people want to do Source + Subject + Sequence; then the solution is,

1. Move the Subject information into the Source
2. Create their own extension, example streamversion

{
  "specversion": "1.0",
  "id": "evt-1234",
  "source": "/streams/OrderAggregate/Order-123",
  "type": "com.myorg.order.ItemAdded",
  "sequence": "00000000000000001234" // unique per source
}

{
  "specversion": "1.0",
  "id": "evt-5678",
  "source": "/streams/OrderAggregate",
  "subject": "Order-456",
  "type": "com.myorg.order.ItemRemoved",
  "sequence": "00000000000000208618", // you can still have `sequence`
  "streamversion": "00000000000000005678" // unique per source + subject
}

[duglin]

On 8/28 call:

• we agreed to limit it to "unique event source", so "source" + "sequence" should be used for comparison
• if that combination doesn't work for certain use cases (e.g. perhaps someone needs a sequence per "source AND subject") then they are free to define a new extension
• additionally they could also add the "subject" to the "source" attribute if they don't want to define a new extension
• @deissnerk will take a first pass at a PR

@BenBeattieHood - any thoughts on this?

[embano1]

@duglin the proposal LGTM: provides clarity, makes it easy to get started with as a reference, and allows for customization which I think especially in Kafka environments is needed where the source typically doesn't carry partition information (which is the resource guaranteeing ordering unless auto/manual repartitioning occurs).

[yordis]

@embano1 check #1366 🙏🏻