feat: get_obstore_store function for creating Obstore store with Open PC credentials

[Kyle Barron (kylebarron)]

Obstore is an alternative to fsspec that's cleaner and can be more performant.

We might also want to consider whether there's an API that would allow to construct a store with either open PC or PC pro with a single function signature?

Gustavo Hidalgo (@ghidalgo3)

[Gustavo Hidalgo (ghidalgo3)]

To make this work with MPCPro, can you:

1. Rename the inner function credential_provider to planetary_computer_credential_provider
2. Add an optional callable argument to get_obstore_store that returns an AzureSASToken. When it's None, use the planetary_computer_credential_provider function.
3. Write a function in _obstore.py that returns callable credential providers for MPCPro instances named planetary_computer_pro_credential_provider.

Such that an end-user can do this:

fromplanetary_computerimportget_obstore_store, planetary_computer_pro_credential_providerfromplanetary_computer.utilsimportparse_blob_urlstac_item : pystac.Item= ... # results of querying a geocatalog account, container=parse_blob_url(stac_item.assets["asset"].href)
store=get_obstore_store(account, container, credential_provider=planetary_computer_pro_credential_provider("https://geocatalog.azure.com")

Maybe the call to parse_blob_url is not necessary and get_obstore_store can just accept a string argument and parse it inside? So that callers can just pass the asset href to get_obstore_store and not have to think about parsing account names and containers.

[Kyle Barron (kylebarron)]

Thinking about this a bit more:

While we can have a credential_provider parameter that matches obstore's own API, it seems a little verbose, and I think it's exposing concepts that are unnecessary here because we know users will always be using a credential provider. It's just a question of which one: Open PC or PC Pro, and that's all we need to have in our signature here. And I think we can figure out a simpler user API here in the context of users who will definitely be using PC.

[Kyle Barron (kylebarron)]

Maybe the call to parse_blob_url is not necessary and get_obstore_store can just accept a string argument and parse it inside? So that callers can just pass the asset href to get_obstore_store and not have to think about parsing account names and containers.

This is why in obstore we have both AzureStore.__init__ and AzureStore.from_url so that we can customize the signature for the two different entry points.

[Kyle Barron (kylebarron)]

• rename to get_obstore
• Rename to credential provider to default to optional argument with default of open pc provider
• One day we can swap in the mpc pro credential provider there.

[Kyle Barron (kylebarron)]

Gustavo Hidalgo (@ghidalgo3) We can't have a default argument as an instance of the credential provider because the credential provider needs to know the account_name and container_name. So for now it defaults to None and that defaults to the open PC credential provider.

[Kyle Barron (kylebarron)]

@microsoft-github-policy-service agree [company="Development Seed"]

[Kyle Barron (kylebarron)]

@microsoft-github-policy-service agree company="Development Seed"

[Gustavo Hidalgo (ghidalgo3)]

Can you please remove Python 3.8 from the test matrix?

[Kyle Barron (kylebarron)]

Obstore doesn't have wheels for 3.8 so this CI is failing on version 3.8. So we should remove the CI tests for 3.8 as well.

[Kyle Barron (kylebarron)]

I removed Python 3.8 from CI Gustavo Hidalgo (@ghidalgo3)

Note that the minimum Python version for install is still 3.7:

planetary-computer-sdk-for-python/pyproject.toml

Line 9 in 4df68b5

requires-python = ">=3.7"

[Gustavo Hidalgo (ghidalgo3)]

Can you please also increase the minimum python version in the toml to 3.9?

[Gustavo Hidalgo (ghidalgo3)]

I have edited the settings to allow your commits to run CI/CD automatically.

[Kyle Barron (kylebarron)]

I'm not sure it worked

[Kyle Barron (kylebarron)]

Gustavo Hidalgo (@ghidalgo3) looks like CI is finally passing!