traefik/healthcheck/healthcheck.go

191 lines
5.4 KiB
Go
Raw Normal View History

2016-11-26 18:48:49 +00:00
package healthcheck
import (
"context"
"fmt"
"net"
2016-11-26 18:48:49 +00:00
"net/http"
"net/url"
"strconv"
2016-11-26 18:48:49 +00:00
"sync"
"time"
2017-01-31 21:55:02 +00:00
"github.com/containous/traefik/log"
"github.com/containous/traefik/safe"
"github.com/go-kit/kit/metrics"
2017-01-31 21:55:02 +00:00
"github.com/vulcand/oxy/roundrobin"
2016-11-26 18:48:49 +00:00
)
var singleton *HealthCheck
var once sync.Once
// GetHealthCheck returns the health check which is guaranteed to be a singleton.
func GetHealthCheck(metrics metricsRegistry) *HealthCheck {
2016-11-26 18:48:49 +00:00
once.Do(func() {
singleton = newHealthCheck(metrics)
2016-11-26 18:48:49 +00:00
})
return singleton
}
// Options are the public health check options.
type Options struct {
2017-11-21 10:06:03 +00:00
Path string
Port int
Transport http.RoundTripper
Interval time.Duration
LB LoadBalancer
}
func (opt Options) String() string {
2017-09-18 13:50:03 +00:00
return fmt.Sprintf("[Path: %s Port: %d Interval: %s]", opt.Path, opt.Port, opt.Interval)
}
2016-11-26 18:48:49 +00:00
// BackendHealthCheck HealthCheck configuration for a backend
type BackendHealthCheck struct {
Options
name string
disabledURLs []*url.URL
requestTimeout time.Duration
2016-11-26 18:48:49 +00:00
}
//HealthCheck struct
type HealthCheck struct {
Backends map[string]*BackendHealthCheck
metrics metricsRegistry
cancel context.CancelFunc
2016-11-26 18:48:49 +00:00
}
// LoadBalancer includes functionality for load-balancing management.
type LoadBalancer interface {
2016-11-26 18:48:49 +00:00
RemoveServer(u *url.URL) error
UpsertServer(u *url.URL, options ...roundrobin.ServerOption) error
Servers() []*url.URL
}
func newHealthCheck(metrics metricsRegistry) *HealthCheck {
return &HealthCheck{
Backends: make(map[string]*BackendHealthCheck),
metrics: metrics,
}
2016-11-26 18:48:49 +00:00
}
// metricsRegistry is a local interface in the healthcheck package, exposing only the required metrics
// necessary for the healthcheck package. This makes it easier for the tests.
type metricsRegistry interface {
BackendServerUpGauge() metrics.Gauge
}
2016-11-26 18:48:49 +00:00
// NewBackendHealthCheck Instantiate a new BackendHealthCheck
func NewBackendHealthCheck(options Options, backendName string) *BackendHealthCheck {
return &BackendHealthCheck{
Options: options,
name: backendName,
requestTimeout: 5 * time.Second,
}
2016-11-26 18:48:49 +00:00
}
2016-11-29 18:30:51 +00:00
//SetBackendsConfiguration set backends configuration
2017-02-06 08:31:20 +00:00
func (hc *HealthCheck) SetBackendsConfiguration(parentCtx context.Context, backends map[string]*BackendHealthCheck) {
2016-11-26 18:48:49 +00:00
hc.Backends = backends
if hc.cancel != nil {
hc.cancel()
}
2017-01-31 21:55:02 +00:00
ctx, cancel := context.WithCancel(parentCtx)
hc.cancel = cancel
2016-11-26 18:48:49 +00:00
2018-01-15 16:27:37 +00:00
for _, backend := range backends {
currentBackend := backend
2017-01-31 21:55:02 +00:00
safe.Go(func() {
hc.execute(ctx, currentBackend)
2017-01-31 21:55:02 +00:00
})
}
2016-11-26 18:48:49 +00:00
}
func (hc *HealthCheck) execute(ctx context.Context, backend *BackendHealthCheck) {
log.Debugf("Initial health check for backend: %q", backend.name)
hc.checkBackend(backend)
ticker := time.NewTicker(backend.Interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
log.Debug("Stopping current health check goroutines of backend: %s", backend.name)
return
case <-ticker.C:
log.Debugf("Refreshing health check for backend: %s", backend.name)
hc.checkBackend(backend)
}
}
}
func (hc *HealthCheck) checkBackend(backend *BackendHealthCheck) {
enabledURLs := backend.LB.Servers()
var newDisabledURLs []*url.URL
for _, url := range backend.disabledURLs {
serverUpMetricValue := float64(0)
if err := checkHealth(url, backend); err == nil {
log.Warnf("Health check up: Returning to server list. Backend: %q URL: %q", backend.name, url.String())
backend.LB.UpsertServer(url, roundrobin.Weight(1))
serverUpMetricValue = 1
} else {
log.Warnf("Health check still failing. Backend: %q URL: %q Reason: %s", backend.name, url.String(), err)
newDisabledURLs = append(newDisabledURLs, url)
}
labelValues := []string{"backend", backend.name, "url", url.String()}
hc.metrics.BackendServerUpGauge().With(labelValues...).Set(serverUpMetricValue)
}
backend.disabledURLs = newDisabledURLs
for _, url := range enabledURLs {
serverUpMetricValue := float64(1)
if err := checkHealth(url, backend); err != nil {
log.Warnf("Health check failed: Remove from server list. Backend: %q URL: %q Reason: %s", backend.name, url.String(), err)
backend.LB.RemoveServer(url)
backend.disabledURLs = append(backend.disabledURLs, url)
serverUpMetricValue = 0
}
labelValues := []string{"backend", backend.name, "url", url.String()}
hc.metrics.BackendServerUpGauge().With(labelValues...).Set(serverUpMetricValue)
}
}
func (backend *BackendHealthCheck) newRequest(serverURL *url.URL) (*http.Request, error) {
2017-09-18 13:50:03 +00:00
if backend.Port == 0 {
2017-11-20 08:40:03 +00:00
return http.NewRequest(http.MethodGet, serverURL.String()+backend.Path, nil)
}
// copy the url and add the port to the host
u := &url.URL{}
*u = *serverURL
2017-09-18 13:50:03 +00:00
u.Host = net.JoinHostPort(u.Hostname(), strconv.Itoa(backend.Port))
u.Path = u.Path + backend.Path
2017-11-20 08:40:03 +00:00
return http.NewRequest(http.MethodGet, u.String(), nil)
}
// checkHealth returns a nil error in case it was successful and otherwise
// a non-nil error with a meaningful description why the health check failed.
func checkHealth(serverURL *url.URL, backend *BackendHealthCheck) error {
2016-11-30 21:48:09 +00:00
client := http.Client{
2017-11-21 10:06:03 +00:00
Timeout: backend.requestTimeout,
Transport: backend.Options.Transport,
2016-11-30 21:48:09 +00:00
}
req, err := backend.newRequest(serverURL)
if err != nil {
return fmt.Errorf("failed to create HTTP request: %s", err)
}
resp, err := client.Do(req)
if err == nil {
defer resp.Body.Close()
2016-11-26 18:48:49 +00:00
}
switch {
case err != nil:
return fmt.Errorf("HTTP request failed: %s", err)
case resp.StatusCode != http.StatusOK:
return fmt.Errorf("received non-200 status code: %v", resp.StatusCode)
}
return nil
2016-11-26 18:48:49 +00:00
}