feat: add deploy config for stabilization delay and probe timeouts

New config options: deploy.stabilization_delay (2s),
deploy.tcp_probe_timeout (30s), deploy.http_probe_timeout (60s).

commit: 547a2f42452f4e38d202de7ac5ca95a873e4c628
author: brice · 7mo ago · 2026-03-02 16:44 UTC
parent: 0f1a5e4
signature: unsigned

2 files · +11 -3

M internal/app/run.go +6 -0


@@ -1193,6 +1193,9 @@ func createContainerService(ctx context.Context, v *viper.Viper, cfg Config, svc
1193	1193		ReadinessDelay: v.GetDuration("deploy.readiness_delay"),
1194	1194		ReadinessMode: v.GetString("deploy.readiness_mode"),
1195	1195		HealthTimeout: v.GetDuration("deploy.health_timeout"),
	1196	+	StabilizationDelay: v.GetDuration("deploy.stabilization_delay"),
	1197	+	TCPProbeTimeout: v.GetDuration("deploy.tcp_probe_timeout"),
	1198	+	HTTPProbeTimeout: v.GetDuration("deploy.http_probe_timeout"),
1196	1199		DrainDelay: v.GetDuration("deploy.drain_delay"),
1197	1200		DrainMode: v.GetString("deploy.drain_mode"),
1198	1201		DrainTimeout: v.GetDuration("deploy.drain_timeout"),

@@ -2374,6 +2377,9 @@ func loadConfig(v *viper.Viper, configPath string) error {
2374	2377		v.SetDefault("deploy.readiness_delay", "5s")
2375	2378		v.SetDefault("deploy.readiness_mode", "auto")
2376	2379		v.SetDefault("deploy.health_timeout", "90s")
	2380	+	v.SetDefault("deploy.stabilization_delay", "2s")
	2381	+	v.SetDefault("deploy.tcp_probe_timeout", "30s")
	2382	+	v.SetDefault("deploy.http_probe_timeout", "60s")
2377	2383		v.SetDefault("deploy.drain_mode", "auto")
2378	2384		v.SetDefault("deploy.drain_timeout", "30s")
2379	2385


@@ -1193,6 +1193,9 @@ func createContainerService(ctx context.Context, v *viper.Viper, cfg Config, svc
1193	ReadinessDelay: v.GetDuration("deploy.readiness_delay"),	1193		ReadinessDelay: v.GetDuration("deploy.readiness_delay"),
1194	ReadinessMode: v.GetString("deploy.readiness_mode"),	1194		ReadinessMode: v.GetString("deploy.readiness_mode"),
1195	HealthTimeout: v.GetDuration("deploy.health_timeout"),	1195		HealthTimeout: v.GetDuration("deploy.health_timeout"),
		1196	+	StabilizationDelay: v.GetDuration("deploy.stabilization_delay"),
		1197	+	TCPProbeTimeout: v.GetDuration("deploy.tcp_probe_timeout"),
		1198	+	HTTPProbeTimeout: v.GetDuration("deploy.http_probe_timeout"),
1196	DrainDelay: v.GetDuration("deploy.drain_delay"),	1199		DrainDelay: v.GetDuration("deploy.drain_delay"),
1197	DrainMode: v.GetString("deploy.drain_mode"),	1200		DrainMode: v.GetString("deploy.drain_mode"),
1198	DrainTimeout: v.GetDuration("deploy.drain_timeout"),	1201		DrainTimeout: v.GetDuration("deploy.drain_timeout"),

@@ -2374,6 +2377,9 @@ func loadConfig(v *viper.Viper, configPath string) error {
2374	v.SetDefault("deploy.readiness_delay", "5s")	2377		v.SetDefault("deploy.readiness_delay", "5s")
2375	v.SetDefault("deploy.readiness_mode", "auto")	2378		v.SetDefault("deploy.readiness_mode", "auto")
2376	v.SetDefault("deploy.health_timeout", "90s")	2379		v.SetDefault("deploy.health_timeout", "90s")
		2380	+	v.SetDefault("deploy.stabilization_delay", "2s")
		2381	+	v.SetDefault("deploy.tcp_probe_timeout", "30s")
		2382	+	v.SetDefault("deploy.http_probe_timeout", "60s")
2377	v.SetDefault("deploy.drain_mode", "auto")	2383		v.SetDefault("deploy.drain_mode", "auto")
2378	v.SetDefault("deploy.drain_timeout", "30s")	2384		v.SetDefault("deploy.drain_timeout", "30s")
2379		2385

M internal/usecase/container/service.go +5 -3


@@ -48,7 +48,9 @@ type Config struct {
48	48		DrainDelayConfigured bool // True when deploy.drain_delay was explicitly configured
49	49		DrainMode string // Drain strategy: auto, inflight, delay
50	50		DrainTimeout time.Duration // Max wait for in-flight requests to drain
51		-	StabilizationDelay time.Duration // Pause after traffic switch before verifying new container health
	51	+	StabilizationDelay time.Duration // Post-switch monitoring window (default 2s)
	52	+	TCPProbeTimeout time.Duration // TCP probe timeout (default 30s)
	53	+	HTTPProbeTimeout time.Duration // HTTP probe timeout (default 60s)
52	54		}
53	55
54	56		var tracer = otel.Tracer("gordon.container")

@@ -2265,7 +2267,7 @@ func (s *Service) readinessCascade(ctx context.Context, containerID string, cont
2265	2267		ip, port, probeErr := s.resolveContainerEndpoint(ctx, containerID, containerConfig)
2266	2268		if probeErr == nil && ip != "" && port > 0 {
2267	2269		url := fmt.Sprintf("http://%s:%d%s", ip, port, healthPath)
2268		-	timeout := cfg.HealthTimeout
	2270	+	timeout := cfg.HTTPProbeTimeout
2269	2271		if timeout == 0 {
2270	2272		timeout = 60 * time.Second
2271	2273		}

@@ -2282,7 +2284,7 @@ func (s *Service) readinessCascade(ctx context.Context, containerID string, cont
2282	2284		ip, port, probeErr := s.resolveContainerEndpoint(ctx, containerID, containerConfig)
2283	2285		if probeErr == nil && ip != "" && port > 0 {
2284	2286		addr := fmt.Sprintf("%s:%d", ip, port)
2285		-	timeout := cfg.HealthTimeout
	2287	+	timeout := cfg.TCPProbeTimeout
2286	2288		if timeout == 0 {
2287	2289		timeout = 30 * time.Second
2288	2290		}

 	DrainDelayConfigured     bool          // True when deploy.drain_delay was explicitly configured
 	DrainMode                string        // Drain strategy: auto, inflight, delay
 	DrainTimeout             time.Duration // Max wait for in-flight requests to drain
-	StabilizationDelay       time.Duration // Pause after traffic switch before verifying new container health
+	StabilizationDelay       time.Duration // Post-switch monitoring window (default 2s)
+	TCPProbeTimeout          time.Duration // TCP probe timeout (default 30s)
+	HTTPProbeTimeout         time.Duration // HTTP probe timeout (default 60s)
 }
 var tracer = otel.Tracer("gordon.container")
 			ip, port, probeErr := s.resolveContainerEndpoint(ctx, containerID, containerConfig)
 			if probeErr == nil && ip != "" && port > 0 {
 				url := fmt.Sprintf("http://%s:%d%s", ip, port, healthPath)
-				timeout := cfg.HealthTimeout
+				timeout := cfg.HTTPProbeTimeout
 				if timeout == 0 {
 					timeout = 60 * time.Second
 				}
 		ip, port, probeErr := s.resolveContainerEndpoint(ctx, containerID, containerConfig)
 		if probeErr == nil && ip != "" && port > 0 {
 			addr := fmt.Sprintf("%s:%d", ip, port)
-			timeout := cfg.HealthTimeout
+			timeout := cfg.TCPProbeTimeout
 			if timeout == 0 {
 				timeout = 30 * time.Second
 			}