Selectable latency percentile and ECS stopped tasks

Feature Improvements Fixes

TL;DR

Pick p50/p95/p99 or MAX on performance history and filter by degradation severity, and see killed ECS tasks alongside running ones.

Selectable latency percentile and ECS stopped tasks

Selectable percentile and degradation filter on Performance History

Performance history on service overview now supports a selectable percentile — p50, p95, p99, or MAX — in the response-time column header, instead of being fixed to p95. A new degradation filter (All / Degraded <50% / Degraded ≥50%) lets you focus on operations where the tail is actually regressing against the baseline, with cell coloring reflecting the same severity tiers.

Improvements

  • AI Assistant
    • Did-you-mean suggestions on MCP tool calls — when a service or endpoint name doesn't exactly match, the hosted MCP server suggests close matches instead of returning an empty result
  • Discover
    • ECS view now shows killed and stopped tasks alongside running ones, so you can see failures without losing historical context
  • Control Plane
    • Rehydration GET API returns created_by, so the "Created By" column no longer shows a dash for rehydration rules

Fixes

  • Control Plane
    • Drop rules could be overwritten and deleted by stale UI state when multiple edits happened in sequence
    • Rehydration was incorrectly blocked for editors with "Unauthorized user for this action" — now allowed for both editors and admins
  • General
    • Raw </details> tags appeared in the MCP docs section instead of rendering as collapsible content