Using Dagster Serverless, is there a way to downlo...
# ask-community
j
Using Dagster Serverless, is there a way to download a materialized asset partition to inspect & debug locally? When I’m running Dagster locally, Dagit tells me where the pickle file is on disk, but I’m not seeing an analogous way to get the asset on Dagster Cloud.
t
Not that I know of. May I ask how your partition is stored? If you use an I/O Manager, you can dump it out to blob storage, download it from there, and then unpickle it as you would locally.
j
I’m not sure — I didn’t configure a custom IO manager, so I’m using whatever the default is for Serverless
I see logs in the run suggesting S3 is involved, but I believe they’re Dagster Cloud owned buckets. I’m mostly curious about the possibility of debugging production assets which “succeed” but because of bugs don’t contain the data we think it would. For now it seems its best to rerun the job locally to reproduce, or else use logging or Observations to get the needed debugging info
t
Aaah, okay, by default, it'll use
fs_io_manager
, which is made for local development and not recommended in prod (because of the point you mention). Downstream users (ex. BI tools) won't have access to those assets. I'd recommend overriding it with a storage location so your data is stored somewhere accessible to you, either in your own S3 buckets, databases, etc. This way you'll own your data completely and not be dependent on Dagster Cloud hosting it.
j
Oh okay, that makes sense! I will read up on doing that and store our stuff in our own bucket(s), that sounds like a great idea. Thanks for the help! Might I suggest adding a note about this in the Deployment > Serverless guide in the docs (https://docs.dagster.io/dagster-cloud/deployment/serverless)? Maybe I could even draft it & open a PR after I get a chance to read more and setup our own IO Manager configuration.
t
Thanks for the feedback! and yes, feel free to @ me if you raise a PR and I'll get the right eyes on it (if they're not my own 😬)!
🙌 1