Skip to content

Track index-type latency metrics in Prometheus #1579

@sparknack

Description

@sparknack

Background

Knowhere latency metrics currently do not distinguish between different index types, which makes it hard to compare search/build/load/range-search latency across FLAT, HNSW, IVF, DISKANN, sparse indexes, and other index families.

Goal

Add Prometheus latency metrics that can be broken down by index_type while preserving the existing Cardinal metric behavior.

Scope

  • expose index_type as a label on Knowhere latency metrics
  • keep the metric names as build_latency, load_latency, search_latency, and range_search_latency
  • retain Cardinals existing metric compatibility
  • ensure the metric lookup path is concurrency-safe

Notes

This is being addressed by PR #1577.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions