when using <https://github.com/Netflix/metaflow-se...
# ask-metaflow
f
when using https://github.com/Netflix/metaflow-service/ How can I configure and change the docker-compose file to use azure blob storage as artifact store? I am able to run it locally, but would like to see if I can use the artifact store in azure.
✅ 1
a
the metadata service has no dependency on blob storage
f
What do you mean? azure-storage-blob==12.13.1 is one dependency added there. How can I accomplish this then?
a
ah that's for the ui service and not the metadata service. as long as you have the right creds to read from blob storage in azure, you don't need to make any changes
the ui service caches some data from blob storage locally so that the ui is snappier. the urls to cache come from the database and depending on the url scheme, the ui service knows which libs to use to pull down that data
nothing ever gets written to blob storage from either the ui service or the metadata service
f
How do I make it point to an azure blob?
I could not find the env variables to set. I suspct a URL for the blob and some path
a
you don't need to. the ui service gets urls for blobs from the database. if those blobs are from azure blob storage, then it automatically picks the right libs to download those blobs
f
And how does the urls get to the database?
I tried running it locally. There were some issues. For example the DAG could not be shown. However, I was able to run the code fine and watch it run through one of the views
a
basically, when you are running metaflow, it will directly write to blob storage and record the url in the database through the metadata service
the ui service then scans the database and fetches logs etc. using the urls in the database
correct - the ui assumes that you are viewing runs that executed against blob storage to paint many of the views
f
so basically, when as the client runs the metaflow script. it sets the METAFLOW_DEFAULT_DATASTORE to azure. METAFLOW_AZURE_STORAGE_BLOB_SERVICE_ENDPOINT to the endpoint and then it will be sent to the mlflow-service and I can view the dag in the website? As long as it has access, I can view the dag
a
correct
and metadata-service and not mlflow-service 🙂
f
hehe, yeah 😉
okay, thank you for clearing things. I will give it a try tomorrow