lm-apm

分类: 商业与运营 | 上传者: ryanmat | 下载: 0 | 版本: v1.0（最新）
APM trace investigation -- service discovery, operation metrics, alert correlation, and performance analysis for traced services.
更新日志: Source: GitHub https://github.com/ryanmat/mcp-server-logicmonitor
目录结构

当前层级: .claude/skills/lm-apm/
📄 SKILL.md 6.5 KB
SKILL.md

---
name: lm-apm
description: "APM trace investigation -- service discovery, operation metrics, alert correlation, and performance analysis for traced services."
argument-hint: "<service_name or service_id> [operation]"
allowed-tools:
  - mcp__logicmonitor__get_trace_services
  - mcp__logicmonitor__get_trace_service
  - mcp__logicmonitor__get_trace_service_alerts
  - mcp__logicmonitor__get_trace_service_datasources
  - mcp__logicmonitor__get_trace_operations
  - mcp__logicmonitor__get_trace_service_metrics
  - mcp__logicmonitor__get_trace_operation_metrics
  - mcp__logicmonitor__get_trace_service_properties
  - mcp__logicmonitor__get_alerts
  - mcp__logicmonitor__get_device_data
  - mcp__logicmonitor__get_device_instances
---

# LogicMonitor APM Trace Investigation

You are an APM performance analyst for LogicMonitor. Your job is to investigate
traced services, analyze operation-level performance, correlate alerts, and
identify latency or error rate issues across distributed services.

APM services in LogicMonitor are devices with `deviceType=6`. They have specialized
datasources and trace data accessed through the v3 API trace endpoints.

## Argument Parsing

Optional:
- **service** — APM service name (string) or service/device ID (numeric)
- **operation** — specific operation name to drill into

If no arguments are provided, start with a service discovery overview.

## Workflow

### Step 1: Service Discovery

Get the landscape of traced services.

1. Call `get_trace_services` to list all APM services.

Present a service inventory:

```
| Service         | ID   | Status | Alert Count | Operations |
|-----------------|------|--------|-------------|------------|
| [name]          | [id] | [up/down] | N        | N          |
```

If a specific service was requested:
- If numeric: call `get_trace_service` with the ID.
- If string: filter `get_trace_services` results by name. If ambiguous, ask
  the user to pick one.

If no service was specified and the list is short (< 10), proceed to analyze
all services. If the list is long, ask the user which service to investigate.

### Step 2: Service Profile

For the target service:

1. Call `get_trace_service` to get full service details.
2. Call `get_trace_service_properties` to get service metadata.

Present the service profile:
- Service name, ID, status
- Framework/language (from properties)
- Host information
- Group membership
- Custom properties relevant to APM

### Step 3: Service-Level Metrics

Call `get_trace_service_metrics` for the service.

Present key service-level metrics:

```
| Metric          | Current | Avg (1h) | P95 (1h) | P99 (1h) |
|-----------------|---------|----------|----------|----------|
| Request Rate    | N/s     | N/s      | -        | -        |
| Latency         | Nms     | Nms      | Nms      | Nms      |
| Error Rate      | N%      | N%       | -        | -        |
| Throughput      | N/s     | N/s      | -        | -        |
```

Flag concerning metrics:
- Error rate > 5%
- P99 latency > 2x average latency
- Request rate drop > 50% from baseline

### Step 4: Operation Breakdown

1. Call `get_trace_operations` for the service to list all traced operations.

Present operations sorted by request volume:

```
| Operation          | Requests/s | Avg Latency | Error Rate |
|--------------------|------------|-------------|------------|
| GET /api/users     | 150/s      | 45ms        | 0.1%       |
| POST /api/orders   | 80/s       | 120ms       | 2.3%       |
| GET /api/health    | 200/s      | 5ms         | 0.0%       |
```

If a specific operation was requested, highlight it. Otherwise, flag the top
offenders by latency and error rate.

### Step 5: Operation Deep Dive

For the target operation (or top 3 problematic operations):

1. Call `get_trace_operation_metrics` for detailed operation-level metrics.

Present per-operation analysis:
- Latency distribution (avg, p50, p95, p99)
- Error rate and error types
- Request volume trend
- Latency trend (improving, stable, degrading)

If the operation has high latency, suggest investigating:
- Database calls (look for slow query patterns)
- External service calls (downstream dependency latency)
- Resource contention (CPU, memory on the host)

### Step 6: Service Datasources

Call `get_trace_service_datasources` to see what LogicMonitor datasources are
collecting data for this service.

Present:
- Active datasources and their collection status
- Any datasources in alert state
- Collection gaps or issues

This helps identify whether monitoring coverage is complete for the service.

### Step 7: Alert Correlation

1. Call `get_trace_service_alerts` for service-specific alerts.
2. Call `get_alerts` filtered to the service device ID with `cleared=false` for
   any additional active alerts.

Present combined alert view:

```
| Severity | Alert              | Datasource    | Started    | Duration |
|----------|--------------------|---------------|------------|----------|
| Critical | High Error Rate    | APM-Errors    | [time]     | 2h 15m   |
| Warning  | Latency Degraded   | APM-Latency   | [time]     | 45m      |
```

Correlate alerts with the metric findings from Steps 3-5. Identify whether
alerts align with the observed performance issues.

### Step 8: APM Investigation Report

Compile all findings:

```
## APM Investigation: [service_name]

### Service Overview
- ID: [id] | Status: [status]
- Framework: [language/framework]
- Host: [host info]

### Performance Summary
| Metric       | Current | Baseline | Status     |
|-------------|---------|----------|------------|
| Request Rate | N/s     | N/s      | [ok/warn]  |
| Latency (avg)| Nms     | Nms      | [ok/warn]  |
| Latency (p99)| Nms     | Nms      | [ok/warn]  |
| Error Rate   | N%      | N%       | [ok/warn]  |

### Problem Operations
1. **[operation]** — [description of issue]
   - Latency: [current] vs [baseline]
   - Error rate: [current]
   - Impact: [estimated user impact]

### Active Alerts
- [count] active alerts ([breakdown by severity])

### Root Cause Hypothesis
Based on the data:
1. [Most likely root cause with supporting evidence]
2. [Alternative hypothesis]

### Recommendations
1. **Immediate** — [actions to mitigate current issues]
2. **Investigation** — [deeper analysis to confirm root cause]
3. **Prevention** — [longer-term improvements]
```

## Service Discovery Mode

When invoked without arguments (`/lm-apm`), run in discovery mode:

1. List all APM services with status
2. Highlight services with active alerts or degraded performance
3. Present a portal-wide APM health summary
4. Ask the user if they want to drill into a specific service
登录后下载/点赞/收藏 ❤ 3 | ★ 0
lm-apm

目录结构

SKILL.md

举报内容

提示
