Eduardo Muñoz
04/12/2023, 9:48 AMneeds
) and we set when
to manual.
Is there a similar workflow in Dagster for ops
inside a job?
My use case is to train a model and output/log some validation results and choose to manually trigger an op
to deploy the model. This is a compliance requirement. Both ops
are called from the same job
.
The alternative would be load the output of an op
in train_job
in an op
called from deploy_job
. Is this possible? Using I/O managers maybe?
Any other suggestions?claire
04/14/2023, 6:04 PMdeploy_job
with an op that has an unconnected input (https://docs.dagster.io/concepts/io-management/unconnected-inputs#providing-an-input-manager-for-an-unconnected-input) that reads from the ML model.claire
04/14/2023, 6:05 PMtrain_model
asset and a downstream deploy_model
asset, then in the UI or within a schedule/sensor you can kick off the deploy_model
asset whenever desired. This would then automatically read from the latest persisted output for train_model
.Eduardo Muñoz
04/14/2023, 6:53 PM