Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Document parallelism and thread scheduling in the architecture guide #8986

Merged
merged 2 commits into from
Jan 26, 2024
Merged

Document parallelism and thread scheduling in the architecture guide #8986

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ee4ebc1
Document parallelism and thread scheduling in the architecture guid
alamb Jan 24, 2024
8bb038a
Apply suggestions from code review
alamb Jan 25, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
40 changes: 32 additions & 8 deletions datafusion/core/src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -342,16 +342,20 @@
//!
//! [`ExecutionPlan`]s process data using the [Apache Arrow] memory
//! format, making heavy use of functions from the [arrow]
//! crate. Calling [`execute`] produces 1 or more partitions of data,
//! consisting an operator that implements
//! [`SendableRecordBatchStream`].
//!
//! Values are represented with [`ColumnarValue`], which are either
//! crate. Values are represented with [`ColumnarValue`], which are either
//! [`ScalarValue`] (single constant values) or [`ArrayRef`] (Arrow
//! Arrays).
//!
//! Balanced parallelism is achieved using [`RepartitionExec`], which
//! implements a [Volcano style] "Exchange".
//! Calling [`execute`] produces 1 or more partitions of data,
//! as a [`SendableRecordBatchStream`], which implements a pull based execution
//! API. Calling `.next().await` will incrementally compute and return the next
//! [`RecordBatch`]. Balanced parallelism is achieved using [Volcano style]
//! "Exchange" operations implemented by [`RepartitionExec`].
alamb marked this conversation as resolved.
Show resolved Hide resolved
//!
//! While some recent research such as [Morsel-Driven Parallelism] describes challenges
//! with the pull style Volcano execution model on NUMA architectures, in practice DataFusion achieves
//! similar scalability as systems that use morsel driven approach such as DuckDB.
//! See the [DataFusion paper submitted SIGMOD] for more details.
alamb marked this conversation as resolved.
Show resolved Hide resolved
//!
//! [`execute`]: physical_plan::ExecutionPlan::execute
//! [`SendableRecordBatchStream`]: crate::physical_plan::SendableRecordBatchStream
Expand All @@ -364,8 +368,26 @@
//!
//! [`RepartitionExec`]: https://docs.rs/datafusion/latest/datafusion/physical_plan/repartition/struct.RepartitionExec.html
//! [Volcano style]: https://w6113.github.io/files/papers/volcanoparallelism-89.pdf
//! [Morsel-Driven Parallelism]: https://db.in.tum.de/~leis/papers/morsels.pdf
//! [DataFusion paper submitted SIGMOD]: https://github.com/apache/arrow-datafusion/files/13874720/DataFusion_Query_Engine___SIGMOD_2024.pdf
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
//! [DataFusion paper submitted SIGMOD]: https://github.com/apache/arrow-datafusion/files/13874720/DataFusion_Query_Engine___SIGMOD_2024.pdf
//! [DataFusion paper submitted to SIGMOD]: https://github.com/apache/arrow-datafusion/files/13874720/DataFusion_Query_Engine___SIGMOD_2024.pdf

//! [implementors of `ExecutionPlan`]: https://docs.rs/datafusion/latest/datafusion/physical_plan/trait.ExecutionPlan.html#implementors
//!
//! ## Thread Scheduling
//!
//! DataFusion incrementally computes output from a [`SendableRecordBatchStream`]
//! with `target_partitions` threads. Parallelism is implementing using multiple
//! [Tokio] [`task`]s, which are executed by threads managed by a tokio Runtime.
//! While tokio is most commonly used
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I should probably also point out that this phrasing came from the SIGMOD paper and I think at least @yjshen and @ozankabak had a part in authoring it

//! for asynchronous network I/O, its combination of an efficient, work-stealing
//! scheduler, first class compiler support for automatic continuation generation,
//! and exceptional performance makes it a compelling choice for CPU intensive
//! applications as well, as explained in more detail in [Using Rustlang’s Async Tokio
alamb marked this conversation as resolved.
Show resolved Hide resolved
//! Runtime for CPU-Bound Tasks].
//!
//! [Tokio]: https://tokio.rs
//! [`task`]: tokio::task
//! [Using Rustlang’s Async Tokio Runtime for CPU-Bound Tasks]: https://thenewstack.io/using-rustlangs-async-tokio-runtime-for-cpu-bound-tasks/
//!
//! ## State Management and Configuration
//!
//! [`ConfigOptions`] contain options to control DataFusion's
Expand Down Expand Up @@ -393,10 +415,12 @@
//!
//! The amount of memory and temporary local disk space used by
//! DataFusion when running a plan can be controlled using the
//! [`MemoryPool`] and [`DiskManager`].
//! [`MemoryPool`] and [`DiskManager`]. Other runtime options can be
//! found on [`RuntimeEnv`].
//!
//! [`DiskManager`]: crate::execution::DiskManager
//! [`MemoryPool`]: crate::execution::memory_pool::MemoryPool
//! [`RuntimeEnv`]: crate::execution::runtime_env::RuntimeEnv
//! [`ObjectStoreRegistry`]: crate::datasource::object_store::ObjectStoreRegistry
//!
//! ## Crate Organization
Expand Down
Loading