Add filter field to source table definitions (#1495) #1776
Conversation
There was a problem hiding this comment.
This technically look very good to me - the mypy additions here are great, and I like the approach you implemented for making empty source freshness results appear as "stale".
My one holdup is that a top-level filter: config on the source probably isn't the right way to specify this field. I'd expect a field like that to work globally, but it really only contributes to source freshness queries at the moment. In #1495, there's also a comment about how schema tests don't work properly for tables that require filters. I don't think we should try to address that problem right now - that's going to involve changing schema test logic, and it can be worked around in user-space.
Can we change the filter field to live inside of the freshness: block? I think that will assuage my concerns around the placement of the filter: field and will better clarify to users what it actually does. I'll follow up in #1494 with some approaches (outside the scope of this PR) for adding filters in schema tests on BigQuery.
| {% macro default__collect_freshness(source, loaded_at_field) %} | ||
| {% call statement('check_schema_exists', fetch_result=True, auto_begin=False) -%} | ||
| {% macro default__collect_freshness(source, loaded_at_field, filter) %} | ||
| {% call statement('collect_freshness', fetch_result=True, auto_begin=False) -%} |
Added some type annotations Clean up some mypy issues around the "available" decorators
beckjake
force-pushed
the
feature/partition-filters-sources
branch
from
489f631
to
55a6b9a
Compare
beckjake
force-pushed
the
feature/partition-filters-sources
branch
from
55a6b9a
to
0522535
Compare
Fixes #1495
Added a
filterparameter to source table definitions. It defaults toNone. If provided, awhere {{ filter }}clause is added to the source freshness calculation select statement.I also added some tests, did some mypy-related cleanup. While writing the tests, I needed to define some sort of reasonable behavior for when the source freshness query returns no data - I decided that we'll treat your data as if the latest update was on January 1, year 1 at 00:00:00 UTC. Hopefully that's outside your source freshness range. We could alternatively raise an exception about there being no data, but I thought the idea of your data being considered very out of date was more coherent.