Open ryandstoughton opened 7 months ago
What happens if you define the partition definition once (StaticPartitionsDefinition(["partition1", "partition2"]
) and share it across both assets?
Unfortunately it appears to behave the same either way.
partitions = StaticPartitionsDefinition(["partition1", "partition2"])
...
@asset(partitions_def=partitions)
def some_asset2(context: OpExecutionContext, some_asset1: str) -> str:
...
@asset(partitions_def=partitions)
def some_asset4(context: OpExecutionContext, some_asset3: str) -> str:
...
yields
To add further insight, adding a some_asset5
after the partitioned asset that never materializes causes some_asset5
and some_asset1
to materialize at the same time. This appears to ignore the some_asset5
's dependency on the asset before it.
This is still an issue with the latest Dagster version at the time of writing (1.6.4).
This is still an issue for my team and is blocking us from using partitions at all.
Based on this minimum reproducible error, are we misunderstanding partitioning in Dagster or is this truly a bug? We would like to partition our assets by client, but not all assets are able to be partitioned this way. Our alternative is to just run every client sequentially (which is becoming prohibitively slow), or to implement concurrency on our own (would like to lean on Dagster wherever possible).
Any info would be greatly appreciated here.
Dagster version
dagster, version 1.6.1
What's the issue?
When materializing assets that go from non-partitioned to partitioned and then back, the 2nd reintroduction of partitioning seems to keep the asset from being materialized when selecting the asset group and clicking "Materialize selected".
What did you expect to happen?
I expect
some_asset4
to materialize the same assome_asset2
. They both rely on a non-partitioned asset.How to reproduce?
DAGSTER_VERSION = "1.6.1" DAGSTER_LIB_VERSION = "0.22.1"
My asset definitions are
My definitions are loaded in
__init__.py
asMy workspace is defined in
workspace.yaml
asDeployment type
Docker Compose
Deployment details
No response
Additional information
No response
Message from the maintainers
Impacted by this issue? Give it a 👍! We factor engagement into prioritization.