# Feishu Template Hints

This file persists the inspection-template structure that was previously loaded
from Feishu, so the skill can reuse the same layout without re-fetching the
template on every run.

## Template Header

Keep these header fields in each cluster section:

- inspection window
- cluster ID
- Clinic overview link
- provider and region
- `tikv-details` dashboard link
- `dedicated-dataplane-node-info` dashboard link
- `tikv` Loki explore link

## Section Order

Use this fixed order:

1. `Summary`
2. `Key Findings`
3. `Error Signals`
4. `gRPC QPS and Latency`
5. `CPU and Memory`
6. `Critical Thread Pool CPU`
7. `Disk Usage and Region Count`
8. `Disk Load`
9. `ERROR/WARN Logs`
10. `Follow-up Investigation`

## Panel Mapping

Map the template sections to these dashboard panels:

| Section | Dashboard Target | Focus Panels |
|---|---|---|
| Error Signals | `tikv-details` | `Critical Error`, `Server is busy`, `Server report failures` |
| gRPC QPS and Latency | `tikv-details` | `gRPC message count`, `gRPC message failed`, `gRPC message duration`, `Average gRPC message duration` |
| CPU and Memory | `tikv-details` | `CPU`, `Memory` |
| Critical Thread Pool CPU | `tikv-details` | `Raft store CPU`, `Async apply CPU`, `Raft async writer CPU`, `gRPC poll CPU`, `Scheduler pool CPU`, `Unified read pool CPU` |
| Disk Usage and Region Count | `tikv-details` | `Store size`, `Available size`, `Region` |
| Disk Load | `tikv-details` and `dedicated-dataplane-node-info` | `Read/Write Duration 0.99`, `Write Duration Breakdown 0.99`, `Disk I/O Util`, `Read Disk iops`, `Write Disk iops`, `Read Disk Wait Latency`, `Write Disk Wait Latency`, `Read Disk Bandwidth`, `Write Disk Bandwidth` |
| ERROR/WARN Logs | Loki explore | `ERROR|FATAL` lines scoped to `container="tikv"` after removing known daily-noise patterns such as `cdc initialize fail` |

## Interpretation Style

The original template uses short reviewer-oriented conclusions such as:

- the cluster looks healthy and can remain under routine watch
- the cluster has warning-level pressure and should stay under watch
- the cluster has error-level findings and should be checked immediately

The automated skill should keep the same style, but:

- generate English output only
- omit screenshots
- print explicit time windows for sustained warnings
- keep direct dashboard and log links in the cluster header
- keep the health verdict and action items inside a dedicated `Summary` section
- use Feishu-supported colored text tags for verdict, action, and severity labels