Open mmutso-boku opened 10 months ago
@mmutso-boku I ran into this issue as well -- as a workaround you can change the names of your partitions which will swap which one is considered the inner partition because the keys get sorted alphabetically here -> https://github.com/dagster-io/dagster/blob/master/python_modules/dagster/dagster/_core/definitions/multi_dimensional_partitions.py#L211
I am getting a similar issue.
partitions_def = MultiPartitionsDefinition(
{
"date": DailyPartitionsDefinition(start_date="2024-08-20"),
"static": StaticPartitionsDefinition(
["partition1", "partition2", "partition3"]
),
}
)
@asset(
partitions_def=partitions_def,
backfill_policy=BackfillPolicy.single_run(),
metadata={"partition_expr": "mmm"},
io_manager_key="duckdb_io_manager",
)
def test_multi_part_asset(context):
context.log.info("hi")
If I run 2 date x 2 static - it will create a new job run for each date partition (2 job runs)
However, it only seems to create two job runs for 2x2, I've tried a bunch of other combinations and it properly creates a single run.
Dagster version
1.5.7
What's the issue?
I have a MultiPartitioned asset (outer Hourly, inner Static) with single run backfill policy. When starting a backfill for X amount of outer partitions and all of the inner partitions, then the backfill runs in a single run, as expected. When starting a backfill for X amount of outer partitions and NOT all of the inner partitions, but just some of them, then the backfill runs as a multi-run backfill - each partition is materialized in a separate run.
What did you expect to happen?
Having defined
backfill_policy=BackfillPolicy.single_run()
on the asset, I expect the backfill to start in a single run for the selected partition range. Instead, each single partitions is executed in a separate run.How to reproduce?
From the UI select multiple outer partitions, and from the inner partitions, select one or two partitions.
Deployment type
Dagster Helm chart
Deployment details
No response
Additional information
No response
Message from the maintainers
Impacted by this issue? Give it a 👍! We factor engagement into prioritization.