The latest updates on your projects. Learn more about Vercel for Git ↗︎

Test deployment unnecessary, no dbt files have been changed.

These changes are good! I actually had a slightly different idea but I think we should merge this first as this will work and pragmatically it makes the most sense. I think this depends on some docs and clean up that I need to do.

Here's what I was hoping:

I was hoping though that we could just write a single metric model. These changes work but my vision for this would definitely require a deeper refactor of the factory. Here's the idea:

1. A single model for say "commits" is written like the current commits.sql
2. From this single model we can derive the following style of metrics:
   • Time Aggregations: commits_to_{entity}_daily, commits_to_{entity}_weekly, commits_to_{entity}_monthly, etc.
   • Rolling Window Aggregations: commits_to_{entity}_over_{window}_{unit}
   • Key Metrics: commits_to_{entity}_over_all_time

This would then instead look like this in metrics_factories.py:

        "commits": MetricQueryDef(
            ref="commits.sql",
            time_aggregations=["daily", "weekly", "monthly"],
            rolling=RollingConfig(
                windows=[30, 90, 180],
                unit="day",
                cron="@daily", 
            ),
            include_all_time=True,
        ),

If we do this we would need to change a bit of how the metrics_peer_ref works (I believe)

Sooo....

I will make a follow on issue for this :), but I think we can wrap up anything needed with the current PR and merge it!

Oh also, I'm going to start writing an issue as a design doc for semantic models in sqlmesh. If we did that, then all of this is moot as we could handle that automatically (I think)