Feature: support embedded page_view_id

[jtcohen6]

Background

Many of our recent Snowplow installations have resulted in a single event stream table, with a schema matching Snowplow's canonical event model.

In these cases, we do not need to look up the page_view_id in a separate table containing web page context; it just needs to be un-arrayed and un-nested from the contexts object on the main events table. This change also enables a more fully incremental build, since page_view_id and collector_tstamp are united from the start.

N.B. OTF = "On The Fence" = I considered multiple approaches and picked one without being sure it's the best. Open to input.

Changelog

• Add functionality whereby a package installer can set 'snowplow:context:web_page': false. The package will expect to see a column called page_view_id directly within the event model.
• Add snowplow_web_events_tmp model/macro. IFF the web page context is disabled, this model performs the deduplication of snowplow_web_page_context—throwing away all events that have multiple page view IDs—directly on top of base events. All subsequent models (snowplow_web_events, snowplow_web_events_time, snowplow_web_events_scroll_depth) build on top of this one.
• Add canonical_event and canonical_event_update seeds that are exact replicas of event/event_update merged with web_page/web_page_update. OTF: Whether to include brand-new seeds or to just make the join happen in base_event. Adding more seeds is definitely duplicate code, but it's also in accordance with our integration test practice (?) of having seeds represent the expected format of raw data.
• Add new tests on the three covered adapters where 'snowplow:context:web_page': false.

Comments

• This PR assumes that the package installer has already plucked the page_view_id from contexts and added it to 'snowplow:events' as a column by the same name.
• OTF: Whether to include a cross-db macro to grab values from Snowplow contexts, or to include page-view plucking by default in the snowplow_web_events_tmp, or to do neither and leave it up to the installer (status quo).

[drewbanin]

OTF Add canonical_event and canonical_event_update seeds that are exact replicas of event/event_update merged with web_page/web_page_update

Yep, I think this is appropriate. Good call.

OTF Whether to include a cross-db macro to grab values from Snowplow contexts, or to include page-view plucking by default in the snowplow_web_events_tmp, or to do neither and leave it up to the installer (status quo).

I think we should leave this up to the user, but conceivably it will make sense to provide helper models / macros for very typical use cases (like Snowflake or Spectrum nested fields).

[drewbanin]

@jtcohen6 do you want me to re-review this one?

[jtcohen6]

@drewbanin Yessir. I've made a few more changes—related though likely beyond the initial scope of this PR—in order to support my experimentation with external tables. Namely:

• Allow the definition of a partition column + datepart precision. This is collector_tstamp + second by default.
• Use get_most_recent_record() macro instead of where ts > (select max(ts) from {{this}}) paradigm, since BQ + Spectrum + Snowflake external cannot prune scanning based on dynamic partition filters.

I believe these changes are relevant. To my mind, the primary use case for this PR's functionality is when Snowplow data is loaded or queried, in its canonical event structure, directly from external storage.

Failing tests

I would also appreciate your eye on the failing CircleCI tests, whose operative error appears to be:

ERROR: google-api-core 1.14.2 has requirement setuptools>=34.0.0, but you'll have setuptools 28.8.0 which is incompatible.

All tests are passing for me locally.