[docs-only] Events adr

[butonic]

writing down the decision process about events in ocis/reva

[kobergj]

Uh, I have 5 cents for this topic 😄

1. Eventstore
   Do we want to store all events in a common event store? That way it would be possible to "replay" events to get the system to any state it ever was in. But this has also downsides. Most notably GDPR compliance. (Events in the store must not be deleted)

2. "At-least-once" vs "At-most-once" QoS
   I would opt for "at-least-once". Events should be idempotent if possible. If a service goes down the events should stay in the queue until they are consumed. Otherwise a service can never be sure his data is correct

3. Eventual consistency
   Just to mention this here: Eventual consistency brings extra complexity to the software. I personally think it is worth it but it will make it more complicated for the community to contribute.

4. Events and multiple service instances
   Having multiple instances of the same service can lead to problems when dealing with events. For example if two services do the same action twice this could lead to (eg) multiple emails being sent. On the other hand if events are used to update in-memory caches, all service instances would need to get the event.

[dragotin]

Good from my POV