Skip to content
P.
05in production2025 — now

Okiyumi Status

Public status and observability for production services

Reliability · infrastructure

ORIGINPUBLIC90D UPTIME≠ TRUTH

Monitoring that tells the truth: public health and origin health are different things.

Context

A production status and observability system built on Uptime Kuma — but not the stock Kuma page: a custom public UI, an API gateway, a history layer and real monitoring semantics.

Problem

Green internal health doesn't mean the product is reachable for a user. A real incident exposed the gap: the system was alive inside but unreachable outside — and stock monitoring saw nothing.

Solution

  • 01

    Separating PUBLIC HEALTH from ORIGIN HEALTH: two independent views of one system, born from a real incident.

  • 02

    A custom public UI: overall status, groups, 90-day uptime history, latency, SSL, incidents, maintenance and stale-data detection.

  • 03

    Long memory: 730 days of daily history, 90 days of raw latency, Telegram alerts over a durable retry queue, backup monitoring and a historical store.

Architecture

Public UI

RU / EN · dark / light

API gateway

history layer · semantics

Uptime Kuma

monitoring 17 services

Infrastructure probes

backups · restore drill

Telegram delivery

durable retry queue

FLOW

  1. 01Origin health
    → Public probes
  2. 02Public probes
    truth for the user
    → Public status

Numbers

17

active monitors

90

days of UI history

730

days of daily history

Stack

  • Uptime Kuma
  • custom gateway
  • Telegram bot
  • nginx
  • backups
  • probes

Outcome

  • A real isolated restore drill of the backup was performed — recovery is proven in practice, not assumed.
  • A product insight from an incident became an architectural principle: public vs origin drives the whole system.