Hello everyone, I have a small issue with the IO_manager used by my assets, basically I create an asset mentioning the IO_manager_key and the required resource key for this IO_manager, but when I run the materialization, it just uses the default key "io_manager" and takes the default FilesystemIOManager. Asset definition :

def build_output_asset (asset_name: str, key_prefix: list, ins: dict) -> asset:

@asset(

name = asset_name,

key_prefix = key_prefix,

required_resource_keys = {"custom_io_manager"},

io_manager_key = "custom_io_manager",

group_name = "outputs",

ins = {"upstream": AssetIn(**ins)}

)

def asset_def(context, upstream) -> None:

print("WRITING IN GCS")

return asset_def

Resources definition:

resources_definitions = {

"dbt": dbt_cli_resource.configured({

"project_dir": DBT_PROJECT_PATH,

"profiles_dir": DBT_PROFILES

}),

"custom_io_manager": bigquery_pandas_io_manager.configured({

"project": GCP_PROJECT

}),

"io_manager": FilesystemIOManager(),

}

Error when I run materialization : dagster._core.errors.DagsterExecutionLoadInputError: Error occurred while loading input "upstream" of step "dbt_mf__weather_gcs":: FileNotFoundError: [Errno 2] No such file or directory: '/path/to/a/directory/storage/another_path/path/path' I confirm that when I replace the custom_io_manager as the default IO manager my pipeline works but it messes my code a little bit . Troubleshot a bit further, basically the loaded input is using the key "io_manager" and the output is using the "custom_io_manager", isn't there a way to force using the same "custom_io_manager" in both operations ? Thanks in advance

When loading the input, it will use the IO manager that is set on the upstream asset. Are you setting "custom_io_manager" on the upstream asset? Would this make sense to do?

Hi @sandy, the upstream asset is a dbt asset so I assume it just uses the default io manager. This can be a workaround yes since dbt runs are independent from the io_manager

I just wrote up a long-form answer on this: https://github.com/dagster-io/dagster/discussions/14648 I think the second option might be relevant to what you're looking for