Skip to main content

Crate oxidelake_planner

Crate oxidelake_planner 

Source
Expand description

Planning (docs/SPEC.md §2.5): the HardwarePlacementRule — a DataFusion PhysicalOptimizerRule that rewrites eligible FilterExec / ProjectionExec / HashJoinExec / AggregateExec nodes into Gpu*Exec nodes — and the OxidePhysicalCodec, a PhysicalExtensionCodec that lets Ballista ship those nodes to executors.

Placement is decided against a target backend: the local hardware in embedded mode, the declared cluster capability (OXIDE_CLUSTER_BACKEND) on the Ballista scheduler. EXPLAIN shows the target as a tag, e.g. GpuFilterExec[cuda]; every node still selects its real backend at execute time and falls back to the CPU path when the device is absent.

Dependency direction: depends on oxidelake-core, oxidelake-compute and datafusion-proto (the single datafusion-* sub-crate exception, ADR-0009).

Re-exports§

pub use codec::OxidePhysicalCodec;
pub use rule::HardwarePlacementRule;
pub use rule::VectorDistancePlacementRule;
pub use rule::physical_optimizer_rules;
pub use target::CLUSTER_TARGET_ENV;
pub use target::cluster_target_from_env;
pub use target::resolve_cluster_target;

Modules§

codec
OxidePhysicalCodec: (de)serializes Gpu*Exec nodes for Ballista.
rule
HardwarePlacementRule: rewrites eligible physical nodes into Gpu*Exec.
target
Resolution of the placement target backend.