Skip to content

merge_reduction is incorrectly computed in recursive circuits #7561

Open

Description

@mihaibudiu
Use of len_upper_bound() Effect of the different counts
merge_reduction (spine_async.rs:607-608) negative for file-to-vector merges, too large for vector-to-file merges
size_to_level (spine_async.rs:2608) chooses the merge level of a batch A timed vector batch looks larger than a file batch with the same data, so it can go to a higher level. This changes merge scheduling inside nested circuits. Performance only
Negative-weight share (spine_async.rs:353-362) The denominator mixes the two kinds of count, so the share that triggers aggressive merging depends on the batch kind

Activity

  1. added
    DBSP coreRelated to the core DBSP library
    profilerIssues related to the profiler and its APIs
    on Oct 21, 2026
  2. added theissue type on Oct 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

DBSP coreRelated to the core DBSP libraryprofilerIssues related to the profiler and its APIs

Type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions