Adjust kubeip GCP address listing to include random sleep for better address allocation distribution. Add max wait time constant to control sleep duration.

Author: alexei-led

examples/gcp/gke.tf

@@ -280,8 +280,9 @@ resource "kubernetes_daemonset" "kubeip_daemonset" {
         termination_grace_period_seconds = 30
         priority_class_name              = "system-node-critical"
         container {
-          name  = "kubeip-agent"
-          image = "doitintl/kubeip-agent:${var.kubeip_version}"
+          name              = "kubeip-agent"
+          image             = "doitintl/kubeip-agent:${var.kubeip_version}"
+          image_pull_policy = "Always"
           env {
             name = "NODE_NAME"
             value_from {

internal/address/gcp.go

@@ -3,6 +3,7 @@ package address
 import (
 	"context"
 	"fmt"
+	"math/rand"
 	"strings"
 	"time"
 
@@ -26,6 +27,7 @@ const (
 	accessConfigKind            = "compute#accessConfig"
 	defaultPrefixLength         = 96
 	maxRetries                  = 10 // number of retries for assigning ephemeral public IP address
+	maxWaitListTime             = 10 // max time to wait before listing addresses
 )
 
 var (
@@ -221,6 +223,11 @@ func (a *gcpAssigner) Assign(ctx context.Context, instanceID, zone string, filte
 		return errors.Wrapf(err, "check if static public IP is already assigned to instance %s", instanceID)
 	}
 
+	// add random sleep to reduce the chance of multiple kubeip instances getting the same address list
+	waitTime := time.Duration(rand.Intn(maxWaitListTime)) * time.Second //nolint:gosec
+	a.logger.WithField("waitTime", waitTime).Debug("waiting before listing addresses")
+	time.Sleep(waitTime)
+
 	// get available reserved public IP addresses
 	addresses, err := a.listAddresses(filter, orderBy, reservedStatus)
 	if err != nil {