Do not emit an event if it will cause a loop

[Pchelolo]

After an outage we've discovered, that when change-prop is introduced, some bugs playing together might cause rerender loops in RESTBase, when change-prop serenaders some content, and it creates the same event that caused a rerender to happen.

This PR adds some protection against it - change-prop will add the x-rerender-reason header (name subject to a bike shed) to a RESTBase request, containing an original event URI (it's enough as RB only emits resource_change events). If we are about to emit the same event causing a cycle - log an error and drop the event.

Questions:

1. What to use as an event hash. I think URI is enough, but suggestions are welcome
2. This doesn't prevent multiple-step loops (event1 causes rerender1 causes event2 causes rerender2 causes event1). Do we want to support cases like that? As a generic way of preventing multi-step loops, we've had an idea to add something like an XFF header to the event meta to indicate the whole sequence of events that caused some event to fire. This could be useful for debugging as well as to prevent multi-step loops

PS: need to add this header in change-propagation.

cc @wikimedia/services

[gwicke]

1. What to use as an event hash. I think URI is enough, but suggestions are welcome

Perhaps {topic}:{url}, as in resource_change://en.wikipedia.org/v1/page/html/Foobar ?

1. This doesn't prevent multiple-step loops (event1 causes rerender1 causes event2 causes rerender2 causes event1). Do we want to support cases like that? As a generic way of preventing multi-step loops, we've had an idea to add something like an XFF header to the event meta to indicate the whole sequence of events that caused some event to fire. This could be useful for debugging as well as to prevent multi-step loops

Yeah, I think this would be really useful. Even in the absence of loops, it's really useful to see where an event came from.

[gwicke]

[minor, bikeshed, alternative]: x-triggers, x-triggered-by, x-event-path, or x-events?

[d00rman]

👍 for x-triggered-by

[Pchelolo]

x-restbase-triggered-by?

[d00rman]

I think we may want to make it independent of RESTBase. We don't want to re-emit an event for the same URI/topic combo regardless of the original source.

[Pchelolo]

Updated to match the changes in https://github.com/wikimedia/change-propagation/pull/32

[coveralls]

Coverage increased (+0.05%) to 93.589% when pulling e24b6e5 on Pchelolo:rerender_cause into e9b661f on wikimedia:master.

[d00rman]

x-rerender-reason?

[d00rman]

Minor nits, LGTM otherwise

[Pchelolo]

@d00rman updated

[coveralls]

Changes Unknown when pulling acffff4 on Pchelolo:rerender_cause into * on wikimedia:master*.