health-checking.md

Checking health of Turnilo instance

Turnilo instance's health is defined in terms of being able to communicate with all configured Druid brokers and those brokers knowing about all segments in Zookeeper.

It can be checked by sending a GET request to a healthEndpoint path defined in Turnilo's server configuration.

Healthy Turnilo instance responds with HTTP status 200 while an unhealthy one responds with the status of 503. The body of a response contains health status of all configured brokers with optional error message on unhealthy brokers.

While processing the health checking request Turnilo server will send its own requests to all configured druid clusters' brokers for /druid/broker/v1/loadstatus endpoint. It will check that all brokers responds within individually defined cluster timeout (healthCheckTimeout property in cluster properties) and that the response body contains inventoryInitialized flag set to true. If any of the requests to brokers fail to meet the criteria defined above the Turnilo instance is marked as unhealthy.

Response examples

Healthy response example:

HTTP/1.1 200 OK
Content-Type: application/json; charset=utf-8

{
    "clusters": [
        {
            "host": "localhost:8082",
            "message": "",
            "status": "healthy"
        }
    ],
    "status": "healthy"
}

Unhealthy response example:

HTTP/1.1 503 Service Unavailable
Content-Type: application/json; charset=utf-8

{
    "clusters": [
        {
            "host": "localhost:8082",
            "message": "inventory not initialized",
            "status": "unhealthy"
        },
        {
            "host": "192.168.99.100:8082",
            "message": "connection error: 'Error: ESOCKETTIMEDOUT'",
            "status": "unhealthy"
        }
    ],
    "status": "unhealthy"
}