fix(graceful-shutdown): implement configurable sutdown delay

prevents in-flight requests to land on a pod that has already stopped.

.github/workflows/ci-helm-deploy-nginx.yml

@@ -56,6 +56,8 @@ jobs:
           sleep 15 # wait for the scheduler to create pods
           kubectl -n kube-system wait pods -l app.kubernetes.io/name=kubenurse --for=condition=Ready
           kubectl -n kube-system get pods -l app.kubernetes.io/name=kubenurse
+          kubectl rollout restart daemonset kubenurse
+          kubectl rollout status  daemonset kubenurse --timeout=1m
           sleep 60 # Wait to generate some checks etc.
       - name: Check deployment
         uses: ./.github/actions/check-deployment

.github/workflows/ci-helm-deploy-traefik.yml

@@ -60,6 +60,8 @@ jobs:
           sleep 15 # wait for the scheduler to create pods
           kubectl -n kube-system wait pods -l app=kubenurse --for=condition=Ready
           kubectl -n kube-system get pods -l app=kubenurse
+          kubectl rollout restart daemonset kubenurse
+          kubectl rollout status  daemonset kubenurse --timeout=1m
           sleep 60 # Wait to generate some checks etc.
       - name: Check deployment
         uses: ./.github/actions/check-deployment

.github/workflows/ci-kustomize-deploy.yml

@@ -52,6 +52,8 @@ jobs:
           sleep 15 # wait for the scheduler to create pods
           kubectl wait pods -l app.kubernetes.io/name=kubenurse --for=condition=Ready
           kubectl get pods -l  app.kubernetes.io/name=kubenurse
+          kubectl rollout restart daemonset kubenurse
+          kubectl rollout status  daemonset kubenurse --timeout=1m
           sleep 60 # Wait to generate some checks etc.
       - name: Check deployment
         uses: ./.github/actions/check-deployment

.github/workflows/lint.yml

@@ -10,7 +10,7 @@ jobs:
     - uses: actions/checkout@v4
     - uses: golangci/golangci-lint-action@v3
       with:
-        version: v1.52
+        version: v1.55
         args: --timeout 5m
   lint-helm:
     runs-on: ubuntu-latest

.golangci.yml

@@ -42,13 +42,10 @@ linters-settings:
       - gocognit
       - funlen
       - gocyclo
-
 linters:
   disable-all: true
   enable:
     - bodyclose
-    - deadcode
-    - depguard
     - dogsled
     - dupl
     - errcheck
@@ -71,19 +68,18 @@ linters:
     - misspell
     - nakedret
     - prealloc
+    - protogetter
     - rowserrcheck
     - exportloopref
     - staticcheck
-    - structcheck
     - stylecheck
+    - sqlclosecheck
     - typecheck
     - unconvert
     - unparam
     - unused
-    - varcheck
     - whitespace
     - wsl
-issues:
   exclude:
     # Very commonly not checked.
     - 'Error return value of .(l.Sync|.*Close|.*.Write|.*Flush|os\.Remove(All)?|os\.(Un)?Setenv). is not checked'

Dockerfile

@@ -1,5 +1,5 @@
 FROM alpine:latest
-MAINTAINER OpenSource PF <[email protected]>
+LABEL OpenSource="PF <[email protected]>"
 
 RUN apk --no-cache add ca-certificates curl
 COPY kubenurse /bin/kubenurse

internal/kubediscovery/kubediscovery.go

@@ -5,6 +5,7 @@ import (
 	"context"
 	"fmt"
 
+	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
 	"k8s.io/client-go/kubernetes"
@@ -35,7 +36,8 @@ type Neighbour struct {
 	HostIP          string
 	NodeName        string
 	NodeSchedulable NodeSchedulability
-	Phase           string // Pod Phase
+	Phase           v1.PodPhase
+	Terminating     bool
 }
 
 // New creates a new kubediscovery client. The context is used to stop the k8s watchers/informers.
@@ -92,8 +94,9 @@ func (c *Client) GetNeighbours(ctx context.Context, namespace, labelSelector str
 			PodName:         pod.Name,
 			PodIP:           pod.Status.PodIP,
 			HostIP:          pod.Status.HostIP,
-			Phase:           string(pod.Status.Phase),
+			Phase:           pod.Status.Phase,
 			NodeName:        pod.Spec.NodeName,
+			Terminating:     pod.DeletionTimestamp != nil,
 			NodeSchedulable: sched,
 		}
 		neighbours[idx] = n

internal/kubenurse/server.go

@@ -46,6 +46,7 @@ type Server struct {
 // * KUBERNETES_SERVICE_PORT
 // * KUBENURSE_NAMESPACE
 // * KUBENURSE_NEIGHBOUR_FILTER
+// * KUBENURSE_SHUTDOWN_DURATION
 // * KUBENURSE_CHECK_API_SERVER_DIRECT
 // * KUBENURSE_CHECK_API_SERVER_DNS
 // * KUBENURSE_CHECK_ME_INGRESS
@@ -107,12 +108,24 @@ func New(ctx context.Context, k8s kubernetes.Interface) (*Server, error) {
 		return nil, err
 	}
 
+	shutdownDuration := 5 * time.Second
+
+	if v, ok := os.LookupEnv("KUBENURSE_SHUTDOWN_DURATION"); ok {
+		var err error
+		shutdownDuration, err = time.ParseDuration(v)
+
+		if err != nil {
+			return nil, err
+		}
+	}
+
 	chk.KubenurseIngressURL = os.Getenv("KUBENURSE_INGRESS_URL")
 	chk.KubenurseServiceURL = os.Getenv("KUBENURSE_SERVICE_URL")
 	chk.KubernetesServiceHost = os.Getenv("KUBERNETES_SERVICE_HOST")
 	chk.KubernetesServicePort = os.Getenv("KUBERNETES_SERVICE_PORT")
 	chk.KubenurseNamespace = os.Getenv("KUBENURSE_NAMESPACE")
 	chk.NeighbourFilter = os.Getenv("KUBENURSE_NEIGHBOUR_FILTER")
+	chk.ShutdownDuration = shutdownDuration
 
 	//nolint:goconst // No need to make "false" a constant in my opinion, readability is better like this.
 	chk.SkipCheckAPIServerDirect = os.Getenv("KUBENURSE_CHECK_API_SERVER_DIRECT") == "false"
@@ -198,6 +211,12 @@ func (s *Server) Shutdown(ctx context.Context) error {
 	s.ready = false
 	s.mu.Unlock()
 
+	// wait before actually shutting down the http/s server, as the updated
+	// endpoints for the kubenurse service might not have propagated everywhere
+	// (other kubenurse/ingress controller) yet, which will lead to
+	// me_ingress or path errors in other pods
+	time.Sleep(s.checker.ShutdownDuration)
+
 	// stop the scheduled checker
 	s.checker.StopScheduled()
 

internal/servicecheck/servicecheck.go

@@ -11,6 +11,7 @@ import (
 
 	"github.com/postfinance/kubenurse/internal/kubediscovery"
 	"github.com/prometheus/client_golang/prometheus"
+	v1 "k8s.io/api/core/v1"
 )
 
 const (
@@ -170,7 +171,7 @@ func (c *Checker) MeIngress() (string, error) {
 		return skippedStr, nil
 	}
 
-	return c.doRequest(c.KubenurseIngressURL + "/alwayshappy")
+	return c.doRequest(c.KubenurseIngressURL + "/alwayshappy") //nolint:goconst // readability
 }
 
 // MeService checks if the kubenurse is reachable at the /alwayshappy endpoint through the kubernetes service
@@ -186,8 +187,10 @@ func (c *Checker) MeService() (string, error) {
 // which are not schedulable are excluded from this check to avoid possible false errors.
 func (c *Checker) checkNeighbours(nh []kubediscovery.Neighbour) {
 	for _, neighbour := range nh {
-		neighbour := neighbour // pin
-		if c.allowUnschedulable || neighbour.NodeSchedulable == kubediscovery.NodeSchedulable {
+		neighbour := neighbour                 // pin
+		if neighbour.Phase == v1.PodRunning && // only query running pods (excludes pending ones)
+			!neighbour.Terminating && // exclude terminating pods
+			(c.allowUnschedulable || neighbour.NodeSchedulable == kubediscovery.NodeSchedulable) {
 			check := func() (string, error) {
 				if c.UseTLS {
 					return c.doRequest("https://" + neighbour.PodIP + ":8443/alwayshappy")

internal/servicecheck/transport.go

@@ -66,7 +66,7 @@ func generateRoundTripper(extraCA string, insecure bool) (http.RoundTripper, err
 
 	// Append extra CA, if set
 	if extraCA != "" {
-		caCert, err := os.ReadFile(extraCA) //nolint:gosec // Intentionally included by the user.
+		caCert, err := os.ReadFile(extraCA) // Intentionally included by the user.
 		if err != nil {
 			return nil, fmt.Errorf("could not load certificate %s: %w", extraCA, err)
 		}

internal/servicecheck/types.go

@@ -16,6 +16,9 @@ type Checker struct {
 	SkipCheckMeIngress  bool
 	SkipCheckMeService  bool
 
+	// shutdownDuration defines the time during which kubenurse will wait before stopping
+	ShutdownDuration time.Duration
+
 	// Kubernetes API
 	KubernetesServiceHost    string
 	KubernetesServicePort    string