护理机构的 AKS 夜间故障 runbook

夜班时护理记录应用打不开，员工首先不需要 Kubernetes 讲解。需要先知道记录页面、交接表、服药确认、巡视记录是否受影响。具体材料是 alert、页面、电话表、Service Health、AKS node 状态和下一次更新时间。

Key Points

• 先看谁受影响、哪个 workflow 停了。
• Claude Code 整理脱敏 alert、kubectl、Service Health 和联系人。
• 人来决定安全、纸面 fallback、家属联系、vendor 和 rollback。
• runbook 要留下下一次更新时间，而不只是技术诊断。
• 观察首次通知分钟数、影响确认时间、无个人信息 vendor note 和复发次数。

Night Incident Workflow

具体材料是 alert text、时间、受影响页面、夜班报告、Azure Service Health、Container insights、kubectl snapshot 和联系顺序。流程把护理影响和 AKS signal 放在一张表里。

Official sources checked: What is AKS?, AKS best practices, deployment and cluster reliability best practices, enable monitoring for AKS, Azure Service Health overview, AKS troubleshooting documentation, node NotReady troubleshooting, and high CPU in AKS. Related internal guide: Kubernetes deployment guide.

What Claude Code Handles And What Humans Decide

只给 Claude Code 脱敏 alert、node 数、pod 状态、namespace、最近 deploy 时间、Service Health、contact role 和 rollback 候选。不要给住户姓名、护理记录、医疗数据、地址、家属电话或私人联系方式。

Claude Code organizes alert text, impact scope, contact order, privacy-safe vendor note, kubectl check list, and missing update time. Humans decide care safety, paper fallback, family or care manager contact, privacy boundary, vendor escalation, rollback, and recovery announcement.

3 Use Cases

Use case 1: Create the first night memo

• Input: alert text, incident time, affected screen, staff report, Service Health status.
• Output: impact scope, next update time, fallback condition, manager update.
• Human review: resident safety, medication or rounds impact, family contact, on-call decision.

Use case 2: Create kubectl snapshot table

• Input: kubectl get nodes, kubectl get pods, describe output, previous logs, recent release time.
• Output: Node NotReady, CrashLoopBackOff, high CPU, network clue, recent change table.
• Human review: production command, restart, rollback, vendor escalation, Azure permission.

Use case 3: Draft privacy-safe vendor note

• Input: redacted incident memo, affected screen, count, time, checked commands, contact role.
• Output: vendor note, reproduction condition without personal data, next update time, request.
• Human review: names removed, contract scope, urgency, phone escalation.

Copy-Paste Prompt

Act as a night incident runbook reviewer for a care provider using AKS.
Organize alerts, impact, Service Health, kubectl snapshot, contact order, privacy-safe vendor note, rollback, and next update time.
Do not run destructive commands.
Remove resident names, care records, medical data, addresses, and phone numbers.
Return the five rows to fix today.

Working Check Code

// verify-aks-night-incident-runbook.mjs
// No dependencies. Run with: node verify-aks-night-incident-runbook.mjs
const incidentMemo = {
  industry: "care",
  service: "night-shift-care-record",
  detectedAt: "22:18",
  alert: "AKS node NotReady and pod restart spike",
  serviceHealthChecked: false,
  affectedUsers: "unknown",
  kubectlSnapshot: {
    nodesNotReady: 2,
    podsCrashLooping: 3,
    highCpuPods: 1,
    recentDeployWithinHours: 2
  },
  contactOrder: [
    { role: "night shift lead", name: "Sato", phone: "" },
    { role: "system owner", name: "", phone: "080-0000-0000" },
    { role: "vendor", name: "Cloud partner", phone: "03-0000-0000" }
  ],
  personalDataShared: true,
  rollbackNote: "",
  nextUpdateMinutes: 0
};

const problems = [];

if (!incidentMemo.serviceHealthChecked) {
  problems.push({ item: "Azure Service Health", fix: "check Service Health before blaming the application" });
}
if (incidentMemo.affectedUsers === "unknown") {
  problems.push({ item: "affected users", fix: "write whether staff, managers, families, or residents are affected" });
}
if (incidentMemo.kubectlSnapshot.nodesNotReady > 0) {
  problems.push({ item: "node NotReady", fix: "capture node count, last change time, and network or CPU clues" });
}
if (incidentMemo.kubectlSnapshot.podsCrashLooping > 0) {
  problems.push({ item: "pod restart", fix: "capture namespace, deployment, previous logs, and recent release" });
}
for (const contact of incidentMemo.contactOrder) {
  if (!contact.name || !contact.phone) {
    problems.push({ item: "contact: " + contact.role, fix: "write name and phone before the night shift" });
  }
}
if (incidentMemo.personalDataShared) {
  problems.push({ item: "personal data", fix: "remove resident names and care details from vendor escalation notes" });
}
if (!incidentMemo.rollbackNote) {
  problems.push({ item: "rollback", fix: "write rollback or safe read-only mode before escalation" });
}
if (incidentMemo.nextUpdateMinutes <= 0) {
  problems.push({ item: "next update", fix: "write the next staff update time, such as 15 or 30 minutes" });
}

if (problems.length > 0) {
  console.table(problems);
  process.exitCode = 1;
} else {
  console.log("AKS night incident memo passed.");
}

Pitfall: Common Failure Cases

Do not answer night staff with only technical terms. Pair AKS state with the affected screen, paper fallback, and next update time. Do not skip Service Health. Do not share personal data with vendors. Do not trust an old contact sheet. Fix the runbook with Service Health, impact scope, sanitized note, contact order, rollback, and update time.

FAQ

Q. Is AKS too much for care operations?

A. It may be too much for a simple record app. It fits when Kubernetes is already used, multiple services exist, or night recovery matters.

Q. Can Claude Code read incident logs?

A. Yes, after removing resident names, care notes, medical details, addresses, phone numbers, and credentials.

Q. What should the team inspect first?

A. Affected screen, Service Health, Node NotReady, pod restarts, and next update time.

Q. What metric matters?

A. Minutes to first update, impact confirmation time, privacy-safe vendor notes, repeat incidents, and missed records.

Training And Consultation Signal

If night incidents create unclear calls and privacy-risky notes, AKS operations and incident runbooks are a fit for ClaudeCodeLab training.

What I Verified

I checked official AKS and Service Health sources, CTA, executable JavaScript, internal link, external links, locale coverage, and queue removal. The first action is to add affected screen, Service Health, contact order, rollback, and next update time to the memo.