AgentSkillsCN

monitoring-and-alerting

为Solana应用制定监控方案——从RPC健康状况、交易成功率，到程序运行错误与流动性信号的全面监测。借助这些指标，您可以快速搭建仪表盘与预警系统，为应用的稳定运行保驾护航。

SKILL.md
--- frontmatter
name: monitoring-and-alerting
description: Monitoring plan for Solana apps: RPC health, tx success, program errors, and liquidity signals. Use to set up dashboards and alerts.

Monitoring and Alerting

Role framing: You are an observability lead. Your goal is to ensure early detection of issues across RPC, programs, and markets.

Initial Assessment

  • •What components exist? (frontend, backend, programs, bots, LPs)
  • •SLOs for latency/success? On-call structure?
  • •Tools available (Grafana, Datadog, Helius webhooks, Sentry)?

Core Principles

  • •Measure what users feel: tx success, latency, wallet connect, pool health.
  • •Separate signal from noise: actionable alerts only.
  • •Include on-chain + off-chain metrics.

Workflow

  1. •Metrics selection
    • •RPC: latency, error rates, slot lag.
    • •Tx pipeline: submit latency, confirmation time, failure codes.
    • •Program: error counts by code, compute units used.
    • •Market: price, liquidity depth, volume, holder concentration.
  2. •Instrumentation
    • •Add logs with error codes; emit metrics from services/bots.
    • •Subscribe to webhooks for program logs/events.
  3. •Dashboards
    • •Build views for user journeys (connect, sign, swap/mint) and infra (RPC health).
  4. •Alerts
    • •Set thresholds and runbooks (e.g., tx fail rate >3% over 5m -> switch RPC).
    • •Pager paths with severity levels.
  5. •Testing
    • •Fire drill alerts; validate runbooks; ensure contacts current.

Templates / Playbooks

  • •Alert table: metric | threshold | duration | action | owner.
  • •Standard runbook entries for RPC failover, blockhash errors, LP imbalance.

Common Failure Modes + Debugging

  • •Alert fatigue: too many low-priority alerts; prune.
  • •Missing program error visibility; add msg! with codes and parse logs.
  • •Slot lag misread due to provider differences; monitor per provider.
  • •No runbook -> slow response; write and link.

Quality Bar / Validation

  • •Dashboards live with top metrics; alerts tested.
  • •Each alert has runbook and owner.
  • •On-call rotation known; contact methods tested.

Output Format

Provide monitoring plan: metrics list, dashboards needed, alert thresholds with runbooks, and ownership map.

Examples

  • •Simple: Dashboard for tx success + RPC latency; alert to Slack on error spike; runbook to switch RPC.
  • •Complex: Full stack including program log parsing, pool depth alerts, holder concentration tracking; PagerDuty rotation with quarterly drills.