Skip to content

Look up CSR neighbours with async, batched calls #74

Closed as not planned
Closed as not planned

Description

@prrao87

Problem

Expansion reads neighbours with a synchronous call on an in-memory slice (neighbors_from). A disk-backed or partially loaded adjacency can't sit behind that interface without blocking a runtime thread on every lookup.

Idea

Look up the neighbours for a chunk of input rows with one async, batched call ("neighbours for these N keys"), then run today's loop over the result, so reads can be grouped. The review of lance-format/lance-graph#162 asks for async, version-aware adjacency and materialization interfaces for the same reason.

Status

Not needed by #72, which keeps the CSR loaded whole in memory. It becomes a prerequisite only if partial or disk-backed loading becomes a goal.

Activity

  1. prrao87 commented on Nov 3, 2026

    @prrao87
    MemberAuthor

    Closing as not planned. Every adjacency is loaded whole into memory: #72 ruled out topology larger than memory, and Stage 3 (#81) loads the CSR whole too. So batched async lookups have nothing to group and would only add overhead to every hop. Reopen if partial or disk-backed loading becomes a goal.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions