feat(kds): allow to delay full resync when ticker

docs/generated/kuma-cp.md

@@ -728,6 +728,8 @@ experimental:
     flushInterval: 5s # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FLUSH_INTERVAL
     # How often we schedule full KDS resync when experimental event based watchdog is used.
     fullResyncInterval: 60s # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FULL_RESYNC_INTERVAL
+    # If true, then initial full resync is going to be delayed by 0 to FullResyncInterval.
+    delayFullResync: false # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_DELAY_FULL_RESYNC
 
 proxy:
   gateway:

docs/generated/raw/kuma-cp.yaml

@@ -725,6 +725,8 @@ experimental:
     flushInterval: 5s # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FLUSH_INTERVAL
     # How often we schedule full KDS resync when experimental event based watchdog is used.
     fullResyncInterval: 60s # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FULL_RESYNC_INTERVAL
+    # If true, then initial full resync is going to be delayed by 0 to FullResyncInterval.
+    delayFullResync: false # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_DELAY_FULL_RESYNC
 
 proxy:
   gateway:

pkg/config/app/kuma-cp/config.go

@@ -239,6 +239,7 @@ var DefaultConfig = func() Config {
 				Enabled:            false,
 				FlushInterval:      config_types.Duration{Duration: 5 * time.Second},
 				FullResyncInterval: config_types.Duration{Duration: 1 * time.Minute},
+				DelayFullResync:    false,
 			},
 		},
 		Proxy:    xds.DefaultProxyConfig(),
@@ -408,6 +409,8 @@ type ExperimentalKDSEventBasedWatchdog struct {
 	FlushInterval config_types.Duration `json:"flushInterval" envconfig:"KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FLUSH_INTERVAL"`
 	// How often we schedule full KDS resync when experimental event based watchdog is used.
 	FullResyncInterval config_types.Duration `json:"fullResyncInterval" envconfig:"KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FULL_RESYNC_INTERVAL"`
+	// If true, then initial full resync is going to be delayed by 0 to FullResyncInterval.
+	DelayFullResync bool `json:"delayFullResync" envconfig:"KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_DELAY_FULL_RESYNC"`
 }
 
 func (e ExperimentalConfig) Validate() error {

pkg/config/app/kuma-cp/kuma-cp.defaults.yaml

@@ -725,6 +725,8 @@ experimental:
     flushInterval: 5s # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FLUSH_INTERVAL
     # How often we schedule full KDS resync when experimental event based watchdog is used.
     fullResyncInterval: 60s # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FULL_RESYNC_INTERVAL
+    # If true, then initial full resync is going to be delayed by 0 to FullResyncInterval.
+    delayFullResync: false # ENV: KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_DELAY_FULL_RESYNC
 
 proxy:
   gateway:

pkg/config/loader_test.go

@@ -346,6 +346,7 @@ var _ = Describe("Config loader", func() {
 			Expect(cfg.Experimental.KDSEventBasedWatchdog.Enabled).To(BeTrue())
 			Expect(cfg.Experimental.KDSEventBasedWatchdog.FlushInterval.Duration).To(Equal(10 * time.Second))
 			Expect(cfg.Experimental.KDSEventBasedWatchdog.FullResyncInterval.Duration).To(Equal(15 * time.Second))
+			Expect(cfg.Experimental.KDSEventBasedWatchdog.DelayFullResync).To(BeTrue())
 
 			Expect(cfg.Proxy.Gateway.GlobalDownstreamMaxConnections).To(BeNumerically("==", 1))
 			Expect(cfg.EventBus.BufferSize).To(Equal(uint(30)))
@@ -682,6 +683,7 @@ experimental:
     enabled: true
     flushInterval: 10s
     fullResyncInterval: 15s
+    delayFullResync: true
 proxy:
   gateway:
     globalDownstreamMaxConnections: 1
@@ -933,6 +935,7 @@ eventBus:
 				"KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_ENABLED":                                       "true",
 				"KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FLUSH_INTERVAL":                                "10s",
 				"KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_FULL_RESYNC_INTERVAL":                          "15s",
+				"KUMA_EXPERIMENTAL_KDS_EVENT_BASED_WATCHDOG_DELAY_FULL_RESYNC":                             "true",
 				"KUMA_PROXY_GATEWAY_GLOBAL_DOWNSTREAM_MAX_CONNECTIONS":                                     "1",
 				"KUMA_TRACING_OPENTELEMETRY_ENDPOINT":                                                      "otel-collector:4317",
 				"KUMA_EVENT_BUS_BUFFER_SIZE":                                                               "30",

pkg/kds/v2/server/components.go

@@ -2,6 +2,7 @@ package server
 
 import (
 	"context"
+	"math/rand"
 	"time"
 
 	envoy_core "github.com/envoyproxy/go-control-plane/envoy/config/core/v3"
@@ -104,7 +105,22 @@ func newSyncTracker(
 					return time.NewTicker(experimentalWatchdogCfg.FlushInterval.Duration)
 				},
 				NewFullResyncTicker: func() *time.Ticker {
-					return time.NewTicker(experimentalWatchdogCfg.FullResyncInterval.Duration)
+					if experimentalWatchdogCfg.DelayFullResync {
+						// To ensure an even distribution of connections over time, we introduce a random delay within
+						// the full resync interval. This prevents clustering connections within a short timeframe
+						// and spreads them evenly across the entire interval. After the initial trigger, we reset
+						// the ticker, returning it to its full resync interval.
+						// #nosec G404 - math rand is enough
+						delay := time.Duration(experimentalWatchdogCfg.FullResyncInterval.Duration.Seconds()*rand.Float64()) * time.Second
+						ticker := time.NewTicker(experimentalWatchdogCfg.FullResyncInterval.Duration + delay)
+						go func() {
+							<-time.After(delay)
+							ticker.Reset(experimentalWatchdogCfg.FullResyncInterval.Duration)
+						}()
+						return ticker
+					} else {
+						return time.NewTicker(experimentalWatchdogCfg.FullResyncInterval.Duration)
+					}
 				},
 			}, nil
 		}