Health checks

After deploy_command succeeds, reconD can wait for the workload to report healthy before recording the deployment. A failed health check counts as a deployment failure, which leads to a retry or rollback.

Without a health check, reconD records the deployment as soon as the deploy command exits with status 0.

Configure a health check for automatic rollback

Without a health check, reconD rolls back only when the deploy command itself fails. A release that deploys cleanly but then crashes or serves errors is kept. Configure a health check to catch those too.

health_check.type	Success means
none (default)	No check.
http	GET <url> returns status 200 within timeout.
tcp	A TCP connection to the host and port in url opens within timeout.
command	Recognized but not implemented. Configs using it are rejected.

HTTP

health_check:
  type: http
  url: http://localhost:8080/health
  retries: 12
  interval: 10s
  timeout: 5s

• Only status 200 counts as healthy. Other 2xx codes do not.
• Redirects are not followed; a 301 or 302 counts as unhealthy.

TCP

health_check:
  type: tcp
  url: tcp://localhost:5432

The URL must include a host and port.

Timing

reconD makes up to retries attempts, waiting interval between them, and stops at the first success. With the defaults (12 attempts, 10s apart, 5s timeout), a workload has between about 110 seconds (when each check fails fast, for example with connection refused) and 170 seconds (when every check times out) to become healthy before the attempt counts as failed.

The same check runs after a rollback. A rollback that fails its health check counts as a failed rollback attempt.