Health Check

Description

This article mainly introduces the health check function of Apache APISIX. The health check function can proxy requests to healthy nodes when the upstream node fails or migrates, avoiding the problem of service unavailability to the greatest extent. The health check function of APISIX is implemented using lua-resty-healthcheck, which is divided into active check and passive check.

Active check

Active health check mainly means that APISIX actively detects the survivability of upstream nodes through preset probe types. APISIX supports three probe types: HTTP, HTTPS, and TCP.

When N consecutive probes sent to healthy node A fail, the node will be marked as unhealthy, and the unhealthy node will be ignored by APISIX’s load balancer and cannot receive requests; if For an unhealthy node, if M consecutive probes are successful, the node will be re-marked as healthy and can be proxied.

Passive check

Passive health check refers to judging whether the corresponding upstream node is healthy by judging the response status of the request forwarded from APISIX to the upstream node. Compared with the active health check, the passive health check method does not need to initiate additional probes, but it cannot sense the node status in advance, and there may be a certain amount of failed requests.

If N consecutive requests to a healthy node A fail, the node will be marked as unhealthy.

note

Since unhealthy nodes cannot receive requests, nodes cannot be re-marked as healthy using the passive health check strategy alone, so combining the active health check strategy is usually necessary.

tip

• We only start the health check when the upstream is hit by a request. There won’t be any health check if an upstream is configured but isn’t in used.
• If there is no healthy node can be chosen, we will continue to access the upstream.
• We won’t start the health check when the upstream only has one node, as we will access it whether this unique node is healthy or not.

Configuration instructions

Configuration example

You can enable health checks in routes via the Admin API:

note

You can fetch the admin_key from config.yaml and save to an environment variable with the following command:

admin_key=$(yq '.deployment.admin.admin_key[0].key' conf/config.yaml | sed 's/"//g')

curl http://127.0.0.1:9180/apisix/admin/routes/1 -H "X-API-KEY: $admin_key" -X PUT -d '
{
    "uri": "/index.html",
    "plugins": {
        "limit-count": {
            "count": 2,
            "time_window": 60,
            "rejected_code": 503,
            "key": "remote_addr"
        }
    },
    "upstream": {
         "nodes": {
            "127.0.0.1:1980": 1,
            "127.0.0.1:1970": 1
        },
        "type": "roundrobin",
        "retries": 2,
        "checks": {
            "active": {
                "timeout": 5,
                "http_path": "/status",
                "host": "foo.com",
                "healthy": {
                    "interval": 2,
                    "successes": 1
                },
                "unhealthy": {
                    "interval": 1,
                    "http_failures": 2
                },
                "req_headers": ["User-Agent: curl/7.29.0"]
            },
            "passive": {
                "healthy": {
                    "http_statuses": [200, 201],
                    "successes": 3
                },
                "unhealthy": {
                    "http_statuses": [500],
                    "http_failures": 3,
                    "tcp_failures": 3
                }
            }
        }
    }
}'

If APISIX detects an unhealthy node, the following logs will be output in the error log:

enabled healthcheck passive while logging request
failed to receive status line from 'nil (127.0.0.1:1980)': closed
unhealthy TCP increment (1/2) for '(127.0.0.1:1980)'
failed to receive status line from 'nil (127.0.0.1:1980)': closed
unhealthy TCP increment (2/2) for '(127.0.0.1:1980'

tip

To observe the above log information, you need to adjust the error log level to info.

The health check status can be fetched via GET /v1/healthcheck in Control API.

curl http://127.0.0.1:9090/v1/healthcheck/upstreams/healthycheck -s | jq .

Health Check Status

APISIX provides comprehensive health check information, with particular emphasis on the status and counter parameters for effective health monitoring. In the APISIX context, nodes exhibit four states: healthy, unhealthy, mostly_unhealthy, and mostly_healthy. The mostly_healthy status indicates that the current node is considered healthy, but during health checks, the node’s health status is not consistently successful. The mostly_unhealthy status indicates that the current node is considered unhealthy, but during health checks, the node’s health detection is not consistently unsuccessful. The transition of a node’s state depends on the success or failure of the current health check, along with the recording of four key metrics in the counter: tcp_failure, http_failure, success, and timeout_failure.

To retrieve health check information, you can use the following curl command:

 curl -i http://127.0.0.1:9090/v1/healthcheck

Response Example:

[
  {
    "nodes": {},
    "name": "/apisix/routes/1",
    "type": "http"
  },
  {
    "nodes": [
      {
        "port": 1970,
        "hostname": "127.0.0.1",
        "status": "healthy",
        "ip": "127.0.0.1",
        "counter": {
          "tcp_failure": 0,
          "http_failure": 0,
          "success": 0,
          "timeout_failure": 0
        }
      },
      {
        "port": 1980,
        "hostname": "127.0.0.1",
        "status": "healthy",
        "ip": "127.0.0.1",
        "counter": {
          "tcp_failure": 0,
          "http_failure": 0,
          "success": 0,
          "timeout_failure": 0
        }
      }
    ],
    "name": "/apisix/routes/example-hc-route",
    "type": "http"
  }
]

State Transition Diagram

Note that all nodes start with the healthy status without any initial probes, and the counter only resets and updates with a state change. Hence, when nodes are healthy and all subsequent checks are successful, the success counter is not updated and remains zero.

Counter Information

In the event of a health check failure, the success count in the counter will be reset to zero. Upon a successful health check, the tcp_failure, http_failure, and timeout_failure data will be reset to zero.