Description
Basically since DependentResource now has getResource() method, what by definition should return the actual state after reconciliation (so reconcile() method is called ) . Currently the resource is cached now usually in an event source, that most of the cases also serves as a cache.
The question is if it make sense to force this pattern in general, thus to have always an even source or even a caching event source for every dependent resource? See ResourceEventSource .
For Kubernetes resources we already have this cache in Informer, for external resources we have the CachingInboundEventSource for a resource oriented event coming from an external system. And two flavors of polling event source: PollingEventSource and PerResourcePollingEventSource.
What we don't have covered is the situation for an external resource when it is capable to notify about change, but the notification not contains the resource it self. Typically such external events are comming from AWS resources, like from RDS:
see: https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/rds-cloudwatch-events.sample.html
A sample event does not contains the resource itself.
( see also: Event Notification https://martinfowler.com/articles/201701-event-driven.html )
We have two options here either the dependent resource fetches the resource on reconcile and puts it into local cache. Or we prepare an even source that when receives an inbound event also fetches and caches the resource (then triggers and event).
This would be much more optimal, since we just don't fetch resource on every reconcile. So the event source logic should look like this:
sequenceDiagram
participant InboundFetchingEventSource
participant ExternalService
participant EventHandler
ExternalService->>InboundFetchingEventSource: Send Resource Changed Event
InboundFetchingEventSource->>ExternalService: Fetch The Whole Resource
InboundFetchingEventSource->>InboundFetchingEventSource: Cache The Resource
InboundFetchingEventSource->>EventHandler: Propagate Event About The Change
Notes
see also: #1134
(note that this issue does not have that much impact as it seems, since even if there is separate caching layer, the event sources are propagating the resources to the cache.)