Haruno Knowledge base

Health

Check whether the AI service is working, and run a live check when you need to be sure.

  • Operator consoleAvailable

Health answers one question: "is the AI service working?". It also has a Live check button that tests the service for real.

Three made-up models. One is degraded.
The Health page: the status strip with three made-up models, the time buttons, the Live check button, three count cards and the table of made-up models with status, recent calls, last call and error rate

The status strip at the top of every page is a short form of this page. See Signing in and roles.

Is it working?

The time buttons run from 6h to 1y. The page opens on 24 hours. This page reads at most 30 days, so a longer button still shows 30.

The cards count how many AI settings are Healthy, Degraded and Unhealthy.

The table lists each AI setting, worst first:

ColumnWhat it shows
ModelThe setting's name
StatusHealthy, Degraded, Unhealthy or Unknown
Recent callsHow many times it was used in the period
Last callHow long ago it was last used. A setting nobody used in the period takes its status from that last call, so it reads as it last was
Error rateThe share of calls that failed, and how many

Unknown is not a fault. It means there is no reading. The table also has columns about the service's inner workings. They are for engineers.

All of this comes from what real traffic already did. It cannot tell you whether a setting works right now, or at all if nobody used it recently. That is what a live check is for.

Live check

Live check sends a real request to every AI setting and shows which ones answered.

  1. Press Live check.
  2. Read the confirmation. Press Run it — this costs money only if you mean it.
  3. A row of chips shows each setting with the time it took, or the error.
It costs real money

Every run is billed. It sends a tiny request to each setting. Everything else on the page is free. After a check, the published status is replaced at once, and another check cannot be run for a minute. Only a super admin sees the button.

Who can use it

Every operator can read the page. The Live check button is for super admins, so a viewer does not see it.

What a person in support or sales can ask an operator

  • "Customers say the AI is down. Is it?" Look at the status strip and Status here. A super admin can run a live check.

Last updated · PR #2944 · View this version on GitHub

History (1 version)
  1. Internal knowledge base for staff: site, reader keys, assistant, 197 pages and the PR rule (#2944)
    PR #2944Commit a2a7836The page at this version
    What changed
    @@ -0,0 +1,55 @@
+---
+title: Health
+summary: Check whether the AI service is working, and run a live check when you need to be sure.
+feature: ops-health
+covers:
+  - web/monitoring/components/HealthPage.tsx
+  - web/monitoring/components/ModelStatusChips.tsx
+  - server-dotnet/src/Haruno.Server/Controllers/AdminHealthController.cs
+---
+
+**Health** answers one question: "is the AI service working?". It also has a **Live check** button that tests the service for real.
+
+<Shot src="/m/1801bf7d4a8a-ops-health-dark.png" alt="The Health page: the status strip with three made-up models, the time buttons, the Live check button, three count cards and the table of made-up models with status, recent calls, last call and error rate" caption="Three made-up models. One is degraded." />
+
+The status strip at the top of every page is a short form of this page. See [Signing in and roles](/ops/console/signing-in-and-roles/).
+
+## Is it working?
+
+The time buttons run from **6h** to **1y**. The page opens on 24 hours. This page reads at most 30 days, so a longer button still shows 30.
+
+The cards count how many AI settings are **Healthy**, **Degraded** and **Unhealthy**.
+
+The table lists each AI setting, worst first:
+
+| Column | What it shows |
+|---|---|
+| Model | The setting's name |
+| Status | **Healthy**, **Degraded**, **Unhealthy** or **Unknown** |
+| Recent calls | How many times it was used in the period |
+| Last call | How long ago it was last used. A setting nobody used in the period takes its status from that last call, so it reads as it last was |
+| Error rate | The share of calls that failed, and how many |
+
+**Unknown** is not a fault. It means there is no reading. The table also has columns about the service's inner workings. They are for engineers.
+
+All of this comes from what real traffic already did. It cannot tell you whether a setting works right now, or at all if nobody used it recently. That is what a live check is for.
+
+## Live check
+
+**Live check** sends a real request to every AI setting and shows which ones answered.
+
+1. Press **Live check**.
+2. Read the confirmation. Press **Run it — this costs money** only if you mean it.
+3. A row of chips shows each setting with the time it took, or the error.
+
+<Callout variant="warning" title="It costs real money">
+Every run is billed. It sends a tiny request to each setting. Everything else on the page is free. After a check, the published status is replaced at once, and another check cannot be run for a minute. Only a super admin sees the button.
+</Callout>
+
+## Who can use it
+
+Every operator can read the page. The **Live check** button is for super admins, so a viewer does not see it.
+
+## What a person in support or sales can ask an operator
+
+- "Customers say the AI is down. Is it?" Look at the status strip and **Status** here. A super admin can run a live check.