Thank you @Eevans!

Makes sense @JAllemandou!

Merged and deployed the update, thanks!

Summary of JA3N data vetting:

After some troubleshooting I saw that, when we added the linktarget table as a datasource for Commons Impact Metrics, we forgot to add the corresponding sensor.
This made it so that the September DAG run started before the linktarget data was properly loaded to the data lake, and so the CIM job produced empty results.
The MR above adds the proper sensor to the DAG.

OK, I think this time it worked.

• We rolled back to the Iceberg snapshot previous to the corrupting of the data by the backfill process.
• Then we copied April (still not backfilled) into a temp table in the wmf_staging database.
• Then we rolled the tables forward to its present state
• And finally deleted April data and copied the old temporary uncorrupted data into the April gap.

We executed the snapshot revert commands with success:

spark-sql (default)> CALL spark_catalog.system.rollback_to_snapshot('wmf_readership.unique_devices_per_domain_monthly', 2325163397903337906);
25/10/08 18:50:56 WARN BaseTransaction: Failed to load metadata for a committed snapshot, skipping clean-up
previous_snapshot_id	current_snapshot_id
1441360873772091690	2325163397903337906
Time taken: 2.376 seconds, Fetched 1 row(s)

We've had an issue, where the April partition for monthly datasets:

• unique_devices_per_domain_monthly
• unique_devices_per_project_family_monthly
• clickstream_monthly

has been computed with incomplete source data and thus generated an incomplete April partition.

Backfilling plan updated