Page MenuHomePhabricator

Switch experiment_assignment_v1 to use experiment_exposure events
Closed, ResolvedPublic1 Estimated Story Points

Description

The query which populates wmf_experiments.experiment_assignment_v1 includes a commented out filter in its WHERE clause:

WHERE timestamp >= to_timestamp('${date_hour}')
  AND timestamp < to_timestamp('${date_hour}') + INTERVAL 1 HOUR
  -- AND action = 'experiment_exposure' -- Uncomment after experiments start logging exposure (cf. T411677)
  AND action != 'tick' -- Remove once we're filtering to experiment exposure events only

Until we make the switchover, we will continue to treat all events collected from experiments as exposure events, which negatively impacts GrowthBook's performance. Switching over will reduce the volume of data that is scanned any time an experiment is re-analyzed.

Acceptance criteria

  • Query is updated to use action = 'experiment_exposure' events only.
  • Airflow DAG is updated to use the new query URI

Details

Due Date
Thu, Apr 30, 4:00 AM
Related Changes in GitLab:
TitleReferenceAuthorSource BranchDest Branch
analytics_product/growthbook: update query URIrepos/data-engineering/airflow-dags!2164bearlogause-experiment-exposuresmain
Update experiment_assignment_query_iceberg.hqlrepos/product-analytics/data-pipelines!47bearlogause-experiment-exposuresmain
Customize query in GitLab

Event Timeline

mpopov triaged this task as Medium priority.
mpopov updated the task description. (Show Details)