Skip to content

[Performance] sessions.list slow - N+1 transcript fallback + full row build before limit #57715

Description

@loyur

Problem

sessions.list RPC takes 5-8s per call in our environment, causing Gateway backpressure under normal polling load.

Root Cause

  1. Full row build before filter/limit: listSessionsFromStore() builds the full gateway session row for all 253 sessions before applying limit. Even limit=20 doesn't reduce work.

  2. N+1 transcript fallback: buildGatewaySessionRow() calls resolveTranscriptUsageFallback() for ~122/253 sessions that lack totalTokens/contextTokens/estimatedCostUsd fields. Each fallback reads the transcript file from disk.

  3. O(N²) child session resolution: resolveChildSessionKeys() scans the full store for every row.

Evidence

  • sessions.json deserialization is fast (~15ms total)
  • 44% of sessions.list calls exceed 8s under concurrent polling
  • Peak: 234 slow calls at 18:00 when multiple cron jobs trigger
  • limit parameter has near-zero effect on latency

Suggested Fix

In src/gateway/session-utils.ts - listSessionsFromStore():

  1. Reorder: filter/sort/limit using cheap entry fields BEFORE buildGatewaySessionRow()
  2. Pre-compute child index at list level instead of per-row scan
  3. Consider TTL cache for repeated calls with no store changes

Environment

  • OpenClaw: v2026.3.x
  • Sessions: ~253 total across 5 agents
  • OS: Darwin/arm64

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    Priority

    None yet

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions