Production observability

Health probes, structured logs, and queue failure signals for deployed Tyravel apps.

Health endpoint

Scaffolded apps register HealthServiceProvider and expose liveness and readiness routes when config/health.ts is enabled:

// config/health.ts
export default {
  enabled: true,
  path: '/health',
  livenessPath: '/health/live',
  readinessPath: '/health/ready',
  checks: {
    database: true,
    redis: true, // enable when using Redis
  },
} satisfies HealthConfig;

Path	Use
/health/live	Liveness — returns { status: 'ok' } without probing dependencies
/health/ready	Readiness — runs database/Redis checks; 503 when a check fails
/health	Alias for readiness

Point load balancer readiness probes at /health/ready. A 503 on readiness means a dependency failed — inspect the JSON body for per-check status.

Disable Redis checks in single-process dev when Redis is not running:

checks: { database: true, redis: envBool('REDIS_HEALTH_CHECK', false) },

Structured logging

config/log.ts defaults to a stack channel (stdout + file). In production, prefer stdout so your platform captures logs:

export default {
  default: env('LOG_CHANNEL', 'stdout'),
  channels: {
    stdout: { channel: 'stdout' },
    file: { channel: 'file', path: 'storage/logs/tyravel.log' },
    stack: { channel: 'stack', channels: ['stdout'] },
  },
} satisfies LogConfig;

Emit request-scoped context from controllers or middleware:

import { Log } from '@tyravel/core';

Log.info('order.placed', { orderId: order.id, userId: user.id });

Set APP_DEBUG=false in production so exception responses do not leak stack traces to clients.

Queue and job failures

When using QUEUE_CONNECTION=database, monitor:

Signal	Where
Pending jobs	jobs table row count
Failed jobs	failed_jobs table
Worker alive	Separate tyravel queue:work process or container

Retry failed jobs after fixing the root cause:

tyravel queue:retry all

Run at least one worker process in production — HTTP requests only enqueue work; they do not process it.

Graceful shutdown

serve() in @tyravel/core handles SIGTERM: in-flight requests drain before the process exits. Orchestrators (Fly, Railway, Kubernetes) send SIGTERM on scale-in — allow a 30s grace period in your platform config.

Related

• Deployment overview — env vars and process model
• Queues & jobs — dispatch, workers, failed job tables
• Configuration reference — LOG_* and health keys