plugin/forward: add startup health checks with start_failed (#8663)

* plugin/forward: add startup health checks with start_failed

Signed-off-by: houyuwushang <180804215+houyuwushang@users.noreply.github.com>

* test: separate CoreDNS and third-party imports

Signed-off-by: houyuwushang <180804215+houyuwushang@users.noreply.github.com>

---------

Signed-off-by: houyuwushang <180804215+houyuwushang@users.noreply.github.com>

plugin/pkg/proxy/doq_test.go

@@ -521,6 +521,7 @@ func TestDoQHealthCheck(t *testing.T) {
 	})
 
 	p := NewProxy("TestDoQHealth", server.addr(), transport.QUIC)
+	p.initiallyDown.Store(true)
 	p.SetTLSConfig(clientTLS)
 	defer p.Stop()
 	hc := p.GetHealthchecker()
@@ -529,6 +530,9 @@ func TestDoQHealthCheck(t *testing.T) {
 	if err := hc.Check(p); err != nil {
 		t.Fatalf("health check failed: %v", err)
 	}
+	if p.Down(2) {
+		t.Error("successful QUIC health check did not clear startup failure")
+	}
 	msg := <-query
 	if len(msg.Question) != 1 || msg.Question[0].Name != "health.example." || msg.Question[0].Qtype != dns.TypeNS {
 		t.Fatalf("unexpected health query: %v", msg.Question)

plugin/pkg/proxy/health.go

@@ -6,7 +6,6 @@ import (
 	"errors"
 	"net"
 	"net/http"
-	"sync/atomic"
 	"time"
 
 	"github.com/coredns/coredns/plugin/pkg/doh"
@@ -130,7 +129,7 @@ func (h *doqHc) Check(p *Proxy) error {
 		return err
 	}
 
-	atomic.StoreUint32(&p.fails, 0)
+	p.resetFails()
 	return nil
 }
 
@@ -219,7 +218,7 @@ func (h *dnsHc) Check(p *Proxy) error {
 		return err
 	}
 
-	atomic.StoreUint32(&p.fails, 0)
+	p.resetFails()
 	return nil
 }
 
@@ -293,7 +292,7 @@ func (h *dohHc) Check(p *Proxy) error {
 		return err
 	}
 
-	atomic.StoreUint32(&p.fails, 0)
+	p.resetFails()
 	return nil
 }
 

plugin/pkg/proxy/health_test.go

@@ -33,11 +33,15 @@ func TestHealth(t *testing.T) {
 	hc.SetWriteTimeout(10 * time.Millisecond)
 
 	p := NewProxy("TestHealth", s.Addr, transport.DNS)
+	p.initiallyDown.Store(true)
 	p.readTimeout = 10 * time.Millisecond
 	err := hc.Check(p)
 	if err != nil {
 		t.Errorf("check failed: %v", err)
 	}
+	if p.Down(2) {
+		t.Error("successful DNS health check did not clear startup failure")
+	}
 
 	time.Sleep(20 * time.Millisecond)
 	i1 := atomic.LoadUint32(&i)
@@ -110,11 +114,15 @@ func TestHealthHTTPS(t *testing.T) {
 	hc.SetWriteTimeout(1 * time.Second)
 
 	p := NewProxy("TestHealthHTTPS", s.URL, transport.HTTPS)
+	p.initiallyDown.Store(true)
 	p.readTimeout = 10 * time.Millisecond
 	err := hc.Check(p)
 	if err != nil {
 		t.Fatalf("check failed: %v", err)
 	}
+	if p.Down(2) {
+		t.Error("successful HTTPS health check did not clear startup failure")
+	}
 
 	time.Sleep(20 * time.Millisecond)
 	i1 := atomic.LoadUint32(&i)

plugin/pkg/proxy/proxy.go

@@ -15,9 +15,10 @@ import (
 
 // Proxy defines an upstream host.
 type Proxy struct {
-	fails     uint32
-	addr      string
-	proxyName string
+	fails         uint32
+	initiallyDown atomic.Bool
+	addr          string
+	proxyName     string
 
 	transport *Transport
 	doq       *doqTransport
@@ -136,14 +137,28 @@ func (p *Proxy) Healthcheck() {
 	})
 }
 
-// Down returns true if this proxy is down, i.e. has *more* fails than maxfails.
+// StartFailed marks the proxy down until its first successful health check and
+// starts probing immediately. Start must be called first to set the probe interval.
+func (p *Proxy) StartFailed() {
+	p.initiallyDown.Store(true)
+	p.Healthcheck()
+}
+
+// Down reports whether the proxy is awaiting its first successful startup check
+// or has more failures than maxfails. A zero maxfails disables both checks.
 func (p *Proxy) Down(maxfails uint32) bool {
 	if maxfails == 0 {
 		return false
 	}
 
 	fails := atomic.LoadUint32(&p.fails)
-	return fails > maxfails
+	return p.initiallyDown.Load() || fails > maxfails
+}
+
+// resetFails makes the proxy available after a successful health check.
+func (p *Proxy) resetFails() {
+	atomic.StoreUint32(&p.fails, 0)
+	p.initiallyDown.Store(false)
 }
 
 // Stop stops health checking and closes the DoQ transport, when configured.