fix: healthcheck config

2025-08-06 22:57:14 +02:00 · 2025-05-22 15:25:41 -03:00 · 2025-05-22 15:25:41 -03:00 · 88ec017e85
commit 88ec017e85
parent 1f8d174e63
9 changed files with 38 additions and 18 deletions
--- a/docs/content/reference/dynamic-configuration/docker-labels.yml
+++ b/docs/content/reference/dynamic-configuration/docker-labels.yml
@ -264,10 +264,10 @@
 - "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.expected=foobar"
 - "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.interval=42s"
 - "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.payload=foobar"
- "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.port=foobar"
 - "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.serverstransport=foobar"
 - "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.tls=true"
 - "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.timeout=42s"
+- "traefik.tcp.services.tcpservice01.loadbalancer.healthcheck.unhealthyinterval=42s"
 - "traefik.tcp.services.tcpservice01.loadbalancer.proxyprotocol=true"
 - "traefik.tcp.services.tcpservice01.loadbalancer.proxyprotocol.version=42"
 - "traefik.tcp.services.tcpservice01.loadbalancer.serverstransport=foobar"
--- a/docs/content/reference/dynamic-configuration/file.toml
+++ b/docs/content/reference/dynamic-configuration/file.toml
@ -465,10 +465,10 @@
          address = "foobar"
          tls = true
        [tcp.services.TCPService01.loadBalancer.healthCheck]
-          address = "foobar"
          tls = true
          serversTransport = "foobar"
          interval = "42s"
+          unhealthyInterval = "42s"
          timeout = "42s"
          payload = "foobar"
          expected = "foobar"
--- a/docs/content/reference/dynamic-configuration/file.yaml
+++ b/docs/content/reference/dynamic-configuration/file.yaml
@ -528,10 +528,10 @@ tcp:
        serversTransport: foobar
        terminationDelay: 42
        healthCheck:
-          address: foobar
          tls: true
          serversTransport: foobar
          interval: 42s
+          unhealthyInterval: 42s
          timeout: 42s
          payload: foobar
          expected: foobar
--- a/docs/content/reference/dynamic-configuration/kv-ref.md
+++ b/docs/content/reference/dynamic-configuration/kv-ref.md
@ -393,13 +393,13 @@ THIS FILE MUST NOT BE EDITED BY HAND
 | `traefik/tcp/serversTransports/TCPServersTransport1/tls/spiffe/ids/0` | `foobar` |
 | `traefik/tcp/serversTransports/TCPServersTransport1/tls/spiffe/ids/1` | `foobar` |
 | `traefik/tcp/serversTransports/TCPServersTransport1/tls/spiffe/trustDomain` | `foobar` |
-| `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/address` | `foobar` |
 | `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/expected` | `foobar` |
 | `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/interval` | `42s` |
 | `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/payload` | `foobar` |
 | `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/serversTransport` | `foobar` |
 | `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/timeout` | `42s` |
 | `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/tls` | `true` |
+| `traefik/tcp/services/TCPService01/loadBalancer/healthCheck/unhealthyInterval` | `42s` |
 | `traefik/tcp/services/TCPService01/loadBalancer/proxyProtocol/version` | `42` |
 | `traefik/tcp/services/TCPService01/loadBalancer/servers/0/address` | `foobar` |
 | `traefik/tcp/services/TCPService01/loadBalancer/servers/0/tls` | `true` |
--- a/pkg/config/dynamic/tcp_config.go
+++ b/pkg/config/dynamic/tcp_config.go
@ -178,10 +178,11 @@ func (t *TCPServersTransport) SetDefaults() {

 // TCPServerHealthCheck holds the HealthCheck configuration.
 type TCPServerHealthCheck struct {
-	TLS              bool            `json:"tls,omitempty" toml:"tls,omitempty" yaml:"tls,omitempty" export:"true"`
-	ServersTransport string          `json:"serversTransport,omitempty" toml:"serversTransport,omitempty" yaml:"serversTransport,omitempty" export:"true"`
-	Interval         ptypes.Duration `json:"interval,omitempty" toml:"interval,omitempty" yaml:"interval,omitempty" export:"true"`
-	Timeout          ptypes.Duration `json:"timeout,omitempty" toml:"timeout,omitempty" yaml:"timeout,omitempty" export:"true"`
-	Payload          string          `json:"payload,omitempty" toml:"payload,omitempty" yaml:"payload,omitempty" export:"true"`
-	Expected         string          `json:"expected,omitempty" toml:"expected,omitempty" yaml:"expected,omitempty" export:"true"`
+	TLS               bool             `json:"tls,omitempty" toml:"tls,omitempty" yaml:"tls,omitempty" export:"true"`
+	ServersTransport  string           `json:"serversTransport,omitempty" toml:"serversTransport,omitempty" yaml:"serversTransport,omitempty" export:"true"`
+	Interval          ptypes.Duration  `json:"interval,omitempty" toml:"interval,omitempty" yaml:"interval,omitempty" export:"true"`
+	UnhealthyInterval *ptypes.Duration `json:"unhealthyInterval,omitempty" toml:"unhealthyInterval,omitempty" yaml:"unhealthyInterval,omitempty" export:"true"`
+	Timeout           ptypes.Duration  `json:"timeout,omitempty" toml:"timeout,omitempty" yaml:"timeout,omitempty" export:"true"`
+	Payload           string           `json:"payload,omitempty" toml:"payload,omitempty" yaml:"payload,omitempty" export:"true"`
+	Expected          string           `json:"expected,omitempty" toml:"expected,omitempty" yaml:"expected,omitempty" export:"true"`
 }
--- a/pkg/config/runtime/runtime_tcp.go
+++ b/pkg/config/runtime/runtime_tcp.go
@ -120,7 +120,7 @@ func (s *TCPServiceInfo) UpdateServerStatus(server string, status bool) {
 	}

 	if currentStatus, loaded := s.serverStatus.LoadOrStore(server, status); loaded && currentStatus != status {
-		s.serverStatus.Swap(server, status)
+		s.serverStatus.Store(server, status)
 	}
 }

--- a/pkg/healthcheck/tcp.go
+++ b/pkg/healthcheck/tcp.go
@ -27,16 +27,17 @@ type ServiceTCPHealthChecker struct {
 	serviceName string
 }

-func NewServiceTCPHealthChecker(dialerManager *tcp.DialerManager, metrics metricsHealthCheck, config *dynamic.TCPServerHealthCheck, service StatusSetter, info *runtime.TCPServiceInfo, targets map[string]*net.TCPAddr, serviceName string) *ServiceTCPHealthChecker {
+func NewServiceTCPHealthChecker(ctx context.Context, dialerManager *tcp.DialerManager, metrics metricsHealthCheck, config *dynamic.TCPServerHealthCheck, service StatusSetter, info *runtime.TCPServiceInfo, targets map[string]*net.TCPAddr, serviceName string) *ServiceTCPHealthChecker {
+	logger := log.Ctx(ctx)
 	interval := time.Duration(config.Interval)
 	if interval <= 0 {
-		log.Error().Msg("Health check interval smaller than zero")
+		logger.Error().Msg("Health check interval smaller than zero")
 		interval = time.Duration(dynamic.DefaultHealthCheckInterval)
 	}

 	timeout := time.Duration(config.Timeout)
 	if timeout <= 0 {
-		log.Error().Msg("Health check timeout smaller than zero")
+		logger.Error().Msg("Health check timeout smaller than zero")
 		timeout = time.Duration(dynamic.DefaultHealthCheckTimeout)
 	}

--- a/pkg/healthcheck/tcp_test.go
+++ b/pkg/healthcheck/tcp_test.go
@ -112,6 +112,22 @@ func Test_ServiceTCPHealthChecker_Check(t *testing.T) {
 			expGaugeValue:         0,
 			targetStatus:          truntime.StatusDown,
 		},
+		{
+			desc: "unhealthy server becoming healthy",
+			server: newTCPServer(t,
+				false,
+				tcpMockSequence{accept: false},
+				tcpMockSequence{accept: true},
+			),
+			config: &dynamic.TCPServerHealthCheck{
+				Interval: ptypes.Duration(time.Millisecond * 100),
+				Timeout:  ptypes.Duration(time.Millisecond * 99),
+			},
+			expNumRemovedServers:  1,
+			expNumUpsertedServers: 1,
+			expGaugeValue:         1,
+			targetStatus:          truntime.StatusUp,
+		},
 		{
 			desc: "healthy server with request and response",
 			server: newTCPServer(t,
@ -191,7 +207,7 @@ func Test_ServiceTCPHealthChecker_Check(t *testing.T) {
 					ServerName:         "example.com",
 				},
 			}})
-			service := NewServiceTCPHealthChecker(dialerManager, &MetricsMock{gauge}, test.config, lb, serviceInfo, targets, "serviceName")
+			service := NewServiceTCPHealthChecker(ctx, dialerManager, &MetricsMock{gauge}, test.config, lb, serviceInfo, targets, "serviceName")

 			for range test.server.StatusSequence {
 				test.server.Next()
--- a/pkg/server/service/tcp/service.go
+++ b/pkg/server/service/tcp/service.go
@ -28,9 +28,10 @@ type Manager struct {
 // NewManager creates a new manager.
 func NewManager(conf *runtime.Configuration, dialerManager *tcp.DialerManager) *Manager {
 	return &Manager{
-		dialerManager: dialerManager,
-		configs:       conf.TCPServices,
-		rand:          rand.New(rand.NewSource(time.Now().UnixNano())),
+		dialerManager:  dialerManager,
+		healthCheckers: make(map[string]*healthcheck.ServiceTCPHealthChecker),
+		configs:        conf.TCPServices,
+		rand:           rand.New(rand.NewSource(time.Now().UnixNano())),
 	}
 }

@ -108,6 +109,7 @@ func (m *Manager) BuildTCP(rootCtx context.Context, serviceName string) (tcp.Han

 		if conf.LoadBalancer.HealthCheck != nil {
 			m.healthCheckers[serviceName] = healthcheck.NewServiceTCPHealthChecker(
+				ctx,
 				m.dialerManager,
 				nil,
 				conf.LoadBalancer.HealthCheck,