Provide better safety for configuration

[colmsnowplow]

Some of our transformations seem easy to misconfigure, in ways that we can't really validate for. Eg. if you provide a typo in an event_name, the enriched filters will simply treat all those events as not matched.

We also can't log the values due to user data privacy concerns in the intended deployment model...

It would be great to think of some creative solutions for this. One thing that I think could work well (and would be v useful to scripting transformations) is to provide a harness to test your config against certain values locally/without processing all the way through. Like a 'transformation config test harness'... But I'm not tied to that solution, I'd love to hear ideas on the topic.

(relevant discussion: https://github.com/snowplow-devops/stream-replicator/pull/125#discussion_r886556282)

[jbeemster]

How about a way to observe whether the filters do anything? So if you event_name = x but after a configured time the filter is never activated to log a warning?

[colmsnowplow]

I guess the case I'm worried about here is a filter configured to event_name = pagee_view -> The filter will be active, but the typo will cause it to filter out all legitimate page views, but this will be invisible until someone notices the problem later.

[jbeemster]

I mean at some level the user has to be responsible for configuration input right? You also cannot control what a user calls the event (they could make one called pagee_view that actually works).

Could you maybe add logging periodically of what the filter inputs did so a user can more easily figure out that they messed up?

[colmsnowplow]

Yeah that's fair - could be over-egging. For custom scripting transformations I think providing tests would be nice anyway. But yes for this, perhaps periodic logs are indeed good enough