Closed jayckaiser closed 2 months ago
I believe that our bug where dynamically-mapped sideloading-to-Stadium task logs are inaccessible in Grid view, but not in Graph view is related to Issue 34535, and resolved in PR 34587. This may have been resolved in Airflow 2.7.2.
Feature: Dynamic Task Mapping for EarthbeamDAG (using Airflow Taskflow API)
Description & motivation
This PR refactors the
EarthbeamDAG
to include a newbuild_dynamic_tenant_year_taskgroup()
method that lists files inraw_dir
(or the python preprocess callable) and dynamically maps Earthbeam task groups across them. The same taskflow API approach is used for dynamic and default taskgroups. Additionally, this method standardizes input file argument passing into Earthmover.This PR is a draft for two reasons:
raw_dir
argument inbuild_tenant_year_taskgroup()
is unused (and I don't know if this is correct).PR Merge Priority:
I need to prioritize finishing this as soon as possible. It may be worth merging into the RC branch early and cleaning up rough edges afterwards.
Changes to existing files:
New files created:
Tests and QC done:
This has been successfully tested in SC dev and Texas (although more analysis of outputs and logs is needed to verify all pathing is working as expected).