Module streaming 

Support for Streaming Operations.

Streaming operations are defined by sequences of insertions, deletions, and replacements with the goal to help study how an algorithm performs under dynamic workloads.

Unlike the components defined in crate::build and crate::search, which usually define a single operation to benchmark, the streaming interface is a little more free-form.

Index algorithms should implement Stream to define how to process different operations. The Stream trait is designed to be layered, with various adaptors that modify the function argument types as needed.

Execution of streaming workloads is handled by the Executor trait.

Built-in Executors

• [executors::bigann::RunBook] - An executor for the BigANN style runbooks.

Stream Adaptors

• [executors::bigann::WithData] - Adapt the raw ranges in a BigANN runbook to data points.

Built-in Utilities

• graph: Tools for working with diskann::graph::DiskANNIndex.

  • graph::InplaceDelete: An implementation of crate::build::Build for invoking the inplace delete method. This is meant to be used in a higher-level Stream implementation.

  • graph::DropDeleted: A tool for cleaning up deleted neighbors after deletions. Like graph::InplaceDelete, this is a building block for Stream implementations.

Modules

executors

Provided Executors.

graph

Built-in methods for invoking streaming operations on a diskann::graph::DiskANNIndex.

Structs

AnyStream

A type-erased Stream implementation that wraps stream (outputs)Stream::Output in Box<dyn Any>.

Traits

Arguments

Operation arguments to Stream.

Executor

A sequential executor for Streams.

Stream

A streaming interface for performing dynamic (streaming) operations on an index.