技能 编程开发 Mistral AI生产部署检查清单

Mistral AI生产部署检查清单

v20260423
mistral-prod-checklist
本清单是部署 Mistral AI 集成到生产环境的完整指南。它涵盖了从凭证安全管理、代码质量门禁、设置健康端点、实现熔断器、执行渐进式发布到定义紧急回滚等关键操作步骤,确保AI系统在上线和运行过程中的稳定性和高可用性。
获取技能
249 次下载
概览

Mistral AI Production Checklist

Overview

Complete checklist for deploying Mistral AI integrations to production. Covers credential management, code quality gates, health endpoints, circuit breaker resilience, gradual rollout, and rollback procedures.

Prerequisites

  • Staging environment tested and verified
  • Production API keys from La Plateforme
  • Deployment pipeline (CI/CD) configured
  • Monitoring and alerting ready (see mistral-observability)

Instructions

Step 1: Pre-Deployment Verification

Credentials

  • [ ] Production API key stored in secret manager (never in env files or code)
  • [ ] Key tested with curl -H "Authorization: Bearer $KEY" https://api.mistral.ai/v1/models
  • [ ] Key has appropriate model access scope
  • [ ] Fallback key available for rotation

Code Quality

  • [ ] npm run typecheck passes
  • [ ] npm test passes (unit + integration)
  • [ ] No hardcoded keys: grep -r "MISTRAL_API_KEY\|sk-" src/ --include="*.ts"
  • [ ] Error handling covers 401, 429, 500+ status codes
  • [ ] Rate limiting/backoff implemented
  • [ ] Logging excludes message content and API keys

Model Configuration

  • [ ] Using versioned model IDs or -latest aliases intentionally
  • [ ] maxTokens set to prevent runaway costs
  • [ ] temperature set appropriately (0 for deterministic, 0.7 for creative)
  • [ ] Token budget alerts configured

Step 2: Health Check Endpoint

import { Mistral } from '@mistralai/mistralai';

interface HealthStatus {
  status: 'healthy' | 'degraded' | 'unhealthy';
  provider: 'mistral';
  latencyMs: number;
  model?: string;
  error?: string;
}

export async function checkHealth(): Promise<HealthStatus> {
  const start = performance.now();
  try {
    const client = new Mistral({ apiKey: process.env.MISTRAL_API_KEY! });
    const models = await client.models.list();
    const latencyMs = Math.round(performance.now() - start);

    return {
      status: latencyMs > 5000 ? 'degraded' : 'healthy',
      provider: 'mistral',
      latencyMs,
      model: models.data?.[0]?.id,
    };
  } catch (error: any) {
    return {
      status: 'unhealthy',
      provider: 'mistral',
      latencyMs: Math.round(performance.now() - start),
      error: error.message,
    };
  }
}

// Express route
app.get('/health', async (req, res) => {
  const health = await checkHealth();
  res.status(health.status === 'unhealthy' ? 503 : 200).json(health);
});

Step 3: Circuit Breaker

class MistralCircuitBreaker {
  private failures = 0;
  private lastFailure = 0;
  private state: 'closed' | 'open' | 'half-open' = 'closed';
  private readonly threshold = 5;
  private readonly resetMs = 60_000;

  async execute<T>(fn: () => Promise<T>, fallback?: () => T): Promise<T> {
    if (this.state === 'open') {
      if (Date.now() - this.lastFailure > this.resetMs) {
        this.state = 'half-open';
      } else if (fallback) {
        return fallback();
      } else {
        throw new Error('Circuit breaker open — Mistral unavailable');
      }
    }

    try {
      const result = await fn();
      if (this.state === 'half-open') {
        this.state = 'closed';
        this.failures = 0;
      }
      return result;
    } catch (error: any) {
      if (error.status >= 500 || error.status === 429) {
        this.failures++;
        this.lastFailure = Date.now();
        if (this.failures >= this.threshold) {
          this.state = 'open';
        }
      }
      throw error;
    }
  }
}

Step 4: Gradual Rollout

set -euo pipefail
# Deploy to canary (10% traffic)
kubectl set image deployment/mistral-app app=mistral-app:v2
kubectl rollout pause deployment/mistral-app

# Monitor for 10 minutes
echo "Monitoring canary..."
for i in $(seq 1 10); do
  curl -sf https://yourapp.com/health | jq '.services.mistral'
  sleep 60
done

# If healthy, resume rollout
kubectl rollout resume deployment/mistral-app
kubectl rollout status deployment/mistral-app

Step 5: Post-Deployment Verification

set -euo pipefail
# 1. Health check
curl -sf https://yourapp.com/health | jq '.'

# 2. Smoke test
curl -X POST https://yourapp.com/api/chat \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"role":"user","content":"ping"}]}' | jq '.choices[0].message.content'

# 3. Check error rate in monitoring
echo "Check Grafana/Datadog for mistral_errors_total"

Step 6: Emergency Rollback

set -euo pipefail
# Immediate rollback
kubectl rollout undo deployment/mistral-app
kubectl rollout status deployment/mistral-app

# Verify
curl -sf https://yourapp.com/health | jq '.'

Alert Configuration

AlertConditionSeverity
API Down5xx errors > 10/minP1
High Latencyp95 > 5000ms for 5minP2
Rate Limited429 errors > 5/minP2
Auth FailureAny 401 errorP1
Circuit OpenBreaker triggeredP2
Cost SpikeSpend > $10/hourP3

Documentation Requirements

  • [ ] Incident runbook created (see mistral-incident-runbook)
  • [ ] Key rotation procedure documented
  • [ ] Rollback procedure tested
  • [ ] On-call escalation path defined
  • [ ] API usage limits documented

Output

  • Production deployment with verified credentials
  • Health check endpoint with latency monitoring
  • Circuit breaker for graceful degradation
  • Gradual rollout procedure
  • Emergency rollback tested

Error Handling

IssueDetectionResolution
Deploy failurekubectl rollout statuskubectl rollout undo
Health check 503Alert triggeredCheck Mistral status, verify credentials
Circuit openMetrics alertInvestigate availability, wait for reset
High error rateMonitoring alertCheck logs, consider rollback

Resources

信息
分类 编程开发
名称 mistral-prod-checklist
版本 v20260423
大小 4.07KB
更新时间 2026-04-28
语言