chore: change to fail health check right after getting sigterm signal

Signed-off-by: Alessandro Yuichi Okimoto <yuichijpn@gmail.com>

Author: cre8ivejp

manifests/bucketeer/charts/api/templates/deployment.yaml

@@ -185,10 +185,11 @@ spec:
                   - -c
                   - |
                     admin_port={{ .Values.envoy.adminPort }}
-                    # Wait for load balancer propagation (must match app container propagation delay)
-                    sleep 15
                     wget -q -T 1 -O- --method=POST --body-data='' \
                       "http://localhost:${admin_port}/drain_listeners?graceful" || true
+
+                    # Wait for load balancer propagation (must match app container propagation delay)
+                    sleep 15
                     exit 0
           command: ["envoy"]
           args:

manifests/bucketeer/charts/batch/templates/deployment.yaml

@@ -223,10 +223,11 @@ spec:
                   - -c
                   - |
                     admin_port={{ .Values.envoy.adminPort }}
-                    # Wait for load balancer propagation (must match app container propagation delay)
-                    sleep 15
                     wget -q -T 1 -O- --method=POST --body-data='' \
                       "http://localhost:${admin_port}/drain_listeners?graceful" || true
+
+                    # Wait for load balancer propagation (must match app container propagation delay)
+                    sleep 15
                     exit 0
           command: ["envoy"]
           args:

manifests/bucketeer/charts/subscriber/templates/deployment.yaml

@@ -205,10 +205,11 @@ spec:
                   - -c
                   - |
                     admin_port={{ .Values.envoy.adminPort }}
-                    # Wait for load balancer propagation (must match app container propagation delay)
-                    sleep 15
                     wget -q -T 1 -O- --method=POST --body-data='' \
                       "http://localhost:${admin_port}/drain_listeners?graceful" || true
+
+                    # Wait for load balancer propagation (must match app container propagation delay)
+                    sleep 15
                     exit 0
           command: ["envoy"]
           args:

manifests/bucketeer/charts/web/templates/deployment.yaml

@@ -273,10 +273,11 @@ spec:
                   - -c
                   - |
                     admin_port={{ .Values.envoy.adminPort }}
-                    # Wait for load balancer propagation (must match app container propagation delay)
-                    sleep 15
                     wget -q -T 1 -O- --method=POST --body-data='' \
                       "http://localhost:${admin_port}/drain_listeners?graceful" || true
+
+                    # Wait for load balancer propagation (must match app container propagation delay)
+                    sleep 15
                     exit 0
           command: ["envoy"]
           args:

pkg/api/cmd/server.go

@@ -567,16 +567,16 @@ func (s *server) Run(ctx context.Context, metrics metrics.Metrics, logger *zap.L
 	defer func() {
 		shutdownStartTime := time.Now()
 
-		// Wait for K8s endpoint propagation
-		// This prevents "context deadline exceeded" errors during high traffic.
-		time.Sleep(propagationDelay)
-
 		// Mark as unhealthy so readiness probes fail
 		// This ensures Kubernetes readiness probe fails on next check,
 		// preventing new traffic from being routed to this pod.
 		healthChecker.Stop()
 		restHealthChecker.Stop()
 
+		// Wait for K8s endpoint propagation
+		// This prevents "context deadline exceeded" errors during high traffic.
+		time.Sleep(propagationDelay)
+
 		// Shutdown order matters due to dependencies:
 		// 1. apiGateway/httpServer make gRPC calls to the backend server
 		// 2. We MUST drain them BEFORE stopping the backend sever

pkg/batch/cmd/server/server.go

@@ -632,15 +632,15 @@ func (s *server) Run(ctx context.Context, metrics metrics.Metrics, logger *zap.L
 	defer func() {
 		shutdownStartTime := time.Now()
 
-		// Wait for K8s endpoint propagation
-		// This prevents "context deadline exceeded" errors during high traffic.
-		time.Sleep(propagationDelay)
-
 		// Mark as unhealthy so readiness probes fail
 		// This ensures Kubernetes readiness probe fails on next check,
 		// preventing new traffic from being routed to this pod.
 		healthChecker.Stop()
 
+		// Wait for K8s endpoint propagation
+		// This prevents "context deadline exceeded" errors during high traffic.
+		time.Sleep(propagationDelay)
+
 		// Gracefully stop gRPC Gateway (calls the gRPC server internally)
 		batchGateway.Stop(serverShutDownTimeout)
 

pkg/web/cmd/server/server.go

@@ -861,16 +861,16 @@ func (s *server) Run(ctx context.Context, metrics metrics.Metrics, logger *zap.L
 	defer func() {
 		shutdownStartTime := time.Now()
 
-		// Wait for K8s endpoint propagation
-		// This prevents "context deadline exceeded" errors during high traffic.
-		time.Sleep(propagationDelay)
-
 		// Mark as unhealthy so readiness probes fail
 		// This ensures Kubernetes readiness probe fails on next check,
 		// preventing new traffic from being routed to this pod.
 		healthcheckServer.Stop(5 * time.Second)
 		restHealthChecker.Stop()
 
+		// Wait for K8s endpoint propagation
+		// This prevents "context deadline exceeded" errors during high traffic.
+		time.Sleep(propagationDelay)
+
 		// Stop REST servers in parallel (these call gRPC servers internally)
 		// Stop these first to drain REST traffic before stopping gRPC
 		var restWg sync.WaitGroup