Hi, I'm running into an issue in which I've created a dependency structure between a few composite_solids by passing Nothing between them (since I don't have real values to pass between them), however, I expected the dependency structure to be enforced for their child solids as well, but this doesn't seem to be the case. For instance, I have two composite_solids. I expect all the child solids within my first composite_solid to finish before any of the child solids in my second composite_solid begin. However, the dependency structure is only being enforced for the child solid who's return value is returned by the composite. The other child solids in my second composite_solid start at the same time as the child solids in my first composite_solid. Please let me know if what I've described makes sense sorry if that is a bit verbose/confusing. Thanks so much for any help!

@alex I think we have an open issue for this

ya - you have to treat you

Nothing

dependency like it was real data and flow it all the way to all the solids you want to block

Oh I have to pass it to all the child solids as well?

ya - I think the right mental model is treat as a piece of data that solids need to have before they can execute

Ok that makes sense, good to know. Thanks for the help. Do you know the issue # so I can follow. This will work for a short term solution but yeah the Nothing stuff w/ composites doesn't feel ideal. Would love to see something a bit cleaner later on, and happy to help contribute if I can.

Awesome thanks!

ya it would be great if you could add a comment on that issue with a simplified version of what you tried to do and why you found it unpleasant

Definitely will do

just so we have more data points to factor in to our eventual solution

May actually have something blocking me from flowing my

Nothing

dependency to my child solid in a specific case—I'm using dagster-dbt and I can't change the input defs to map

Nothing

for the dbt_solid I'm using within my composite solid (I left a comment on that issue with much more detail if anything is unclear). Let me know if you have any suggested workarounds to this. Thanks again!

Thanks for that! I'm not actually familiar with Phabricator—is this a change I should just look for in the next release? In the meantime, I'll just run a local copy of Dagster and make this change myself. Thanks again.

phabricator is just what we use for code review - it will land to github shortly and likely make it out in a release today

Awesome. Thanks again