apache / apache/datafusion

Minimize the dependency on `SessionState`

Open
#11,420 2 comments 0 reactions 0 assignees View on GitHub
enhancement
Dominant language
Rust
Stars
9.3k
Forks
2.4k
Avg merge
3d 7h
Merged PRs (30d)
344

Description

### Is your feature request related to a problem or challenge?

There are many functions in `datafusion-core` that take `SessionState` as arguments but only actually rely on portion of them. This add the additional dependency that is not necessary, therefore blocking us from extracting module out of core #10782.

For example, If we want to pull `CatalogProvider` out of core, we need to pull out `TableProvider` first. But because it has `scan` function that takes `SessionState` which contains `CatalogProviderList` therefore there is a circular dependency. Similar issues are already mentioned in https://github.com/apache/datafusion/issues/11182

### Describe the solution you'd like

I think we need to redesign those functions that take `SessionState` and minimize the dependencies for them.

Given one of the `scan` function here, we can see that we only need `state.config_options().explain` and `state.execution_props()` instead of the whole `SessionState`
https://github.com/apache/datafusion/blob/4bed04e4e312a0b125306944aee94a93c2ff6c4f/datafusion/core/src/datasource/memory.rs#L207-L245

In this case, we can create `TableProviderConext` that encapsulates a subset of the information from `SessionState`.

```rust
pub struct TableProviderConext {
explain_options: ExplainOptions,
executionProps: ExecutionProps // maybe with reference to avoid copy
}
```

The same idea applies to `PhysicalPlanner`, we usually just need `PhysicalOptimizeRules` or other information about plan.
https://github.com/apache/datafusion/blob/4bed04e4e312a0b125306944aee94a93c2ff6c4f/datafusion/core/src/physical_planner.rs#L364-L385

And, `create_initial_plan` in `DefaultPhysicalPlanner`, `ExecutionOptions ` is all we need, nothing else.
https://github.com/apache/datafusion/blob/4bed04e4e312a0b125306944aee94a93c2ff6c4f/datafusion/core/src/physical_planner.rs#L582-L585

### Describe alternatives you've considered

_No response_

### Additional context

_No response_

Contributor guide

Open the contributing guide

Research direction

Start by reading the scan implementation in datafusion/core/src/datasource/memory.rs and the referenced functions in datafusion/core/src/physical_planner.rs: PhysicalPlanner, create_initial_plan, and DefaultPhysicalPlanner. Map which parts of SessionState each entry point actually uses and review the related dependency concerns in issue #11182. Done means the unnecessary SessionState dependencies are replaced with narrower interfaces without preserving the circular dependency.

Written by the indexing model from the issue text.

Assessment

Tech stack
rust
Domain
data-engineering
Issue type
Refactor
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.