Backfilling 2 years of history without melting the warehouse

Check whether AQE is disabled in your Spark conf, skew join handling helped us a lot here.

Any downside to this approach with incremental models?

Run the backfill on a separate compute pool or warehouse if your platform supports it, like Snowflake virtual warehouses or BigQuery reservations, so it doesn't contend with interactive queries at all.

In our case the root cause was an implicit cast preventing pushdown.

Any downside to this approach with incremental models?

We saw the same issue, fixing the partition filter dropped runtime 60%.