Best practices for querying metadata - with context.instance or GraphQL API?

[dagsir[bot]]

Hi Dagster Community, is there a convention/preference of:
• Using GraphQL directly to query the metadata, OR
• Using Dagster-Instance (or OpExecutionContext.instance) to run Python commands similar to GraphQL query?
I believe that some of the GraphQL query has a Python implementation, but not all:

graphQL -> Python equivalent for (instance: DagsterInstance)

runsOrError -> instance.get_runs()
assetOrError -> instance.get_asset_records()
wipeAsset -> instance.wipe_assets()

Is there a documentation which GraphQL query/mutation correspond to which Python function?
Is there a convention we should stick to (e.g. Will the Python equivalent be well-supported? Or should we stick to directly querying GraphQL for stability purposes?)

---

The question was originally asked in Dagster Slack.

[smackesey]

In general you should use context.instance if you are inside an op/asset/schedule/sensor execution function. The GQL API is typically for use in non-Dagster environments.