Enable AttemptToReviveEndpoints option in agent and riot clients

client/client.go

@@ -336,8 +336,7 @@ func (c *HTTPClient) healthCheck(timeout time.Duration) {
 		// the goroutines execute the health-check HTTP request and sets status
 		go func(endpointURL string) {
 
-			// Run a GET request against QED with a timeout
-			// TODO it should be a HEAD instead of a GET
+			// Run a HEAD request against QED with a timeout
 			ctx, cancel := context.WithTimeout(context.Background(), timeout)
 			defer cancel()
 

client/config.go

@@ -133,22 +133,27 @@ type Config struct {
 	// HealthCheckTimeout is the timeout in seconds the healthcheck waits for a response
 	// from a QED server.
 	HealthCheckTimeout time.Duration
+
+	// AttemptToReviveEndpoints sets if dead endpoints will be marked alive again after a
+	// round-robin round. This way, they will be picked up in the next try.
+	AttemptToReviveEndpoints bool
 }
 
 // DefaultConfig creates a Config structures with default values.
 func DefaultConfig() *Config {
 	return &Config{
-		Endpoints:               []string{"127.0.0.1:8800"},
-		APIKey:                  "my-key",
-		Insecure:                DefaultInsecure,
-		Timeout:                 DefaultTimeout,
-		DialTimeout:             DefaultDialTimeout,
-		HandshakeTimeout:        DefaultHandshakeTimeout,
-		ReadPreference:          Primary,
-		MaxRetries:              DefaultMaxRetries,
-		EnableTopologyDiscovery: DefaultTopologyDiscoveryEnabled,
-		EnableHealthChecks:      DefaultHealthCheckEnabled,
-		DiscoveryTimeout:        DefaultTopologyDiscoveryTimeout,
-		HealthCheckTimeout:      DefaultHealthCheckTimeout,
+		Endpoints:                []string{"127.0.0.1:8800"},
+		APIKey:                   "my-key",
+		Insecure:                 DefaultInsecure,
+		Timeout:                  DefaultTimeout,
+		DialTimeout:              DefaultDialTimeout,
+		HandshakeTimeout:         DefaultHandshakeTimeout,
+		ReadPreference:           Primary,
+		MaxRetries:               DefaultMaxRetries,
+		EnableTopologyDiscovery:  DefaultTopologyDiscoveryEnabled,
+		EnableHealthChecks:       DefaultHealthCheckEnabled,
+		DiscoveryTimeout:         DefaultTopologyDiscoveryTimeout,
+		HealthCheckTimeout:       DefaultHealthCheckTimeout,
+		AttemptToReviveEndpoints: false,
 	}
 }

client/options.go

@@ -38,6 +38,7 @@ func configToOptions(conf *Config) ([]HTTPClientOptionF, error) {
 			SetDiscoveryTimeout(conf.DiscoveryTimeout),
 			SetHealthchecks(conf.EnableHealthChecks),
 			SetHealthCheckTimeout(conf.HealthCheckTimeout),
+			SetAttemptToReviveEndpoints(conf.AttemptToReviveEndpoints),
 		}
 		if len(conf.Endpoints) > 0 {
 			options = append(options, SetURLs(conf.Endpoints[0], conf.Endpoints[1:]...))

gossip/auditor/auditor.go

@@ -107,6 +107,7 @@ func NewAuditor(conf Config) (*Auditor, error) {
 		client.SetURLs(conf.QEDUrls[0], conf.QEDUrls[1:]...),
 		client.SetAPIKey(conf.APIKey),
 		client.SetReadPreference(client.Any),
+		client.SetAttemptToReviveEndpoints(true),
 	)
 	if err != nil {
 		return nil, errors.Wrap(err, "Cannot start http client: ")

gossip/monitor/monitor.go

@@ -105,6 +105,7 @@ func NewMonitor(conf *Config) (*Monitor, error) {
 		client.SetURLs(conf.QEDUrls[0], conf.QEDUrls[1:]...),
 		client.SetAPIKey(conf.APIKey),
 		client.SetReadPreference(client.Any),
+		client.SetAttemptToReviveEndpoints(true),
 	)
 	if err != nil {
 		return nil, errors.Wrap(err, "Cannot start http client: ")

tests/riot.go

@@ -379,6 +379,7 @@ func newAttack(conf Config) {
 		client.SetURLs(conf.Endpoint[0], conf.Endpoint[1:]...),
 		client.SetAPIKey(conf.APIKey),
 		client.SetReadPreference(client.Any),
+		client.SetAttemptToReviveEndpoints(true),
 	)
 
 	if err != nil {