[tablet] Split health_check_interval into two flags

go/cmd/vttablet/vttablet.go

@@ -106,7 +106,7 @@ func main() {
 		VREngine:            vreplication.NewEngine(config, ts, tabletAlias.Cell, mysqld, qsc.LagThrottler()),
 		MetadataManager:     &mysqlctl.MetadataManager{},
 	}
-	if err := tm.Start(tablet, config.Healthcheck.IntervalSeconds.Get()); err != nil {
+	if err := tm.Start(tablet, config.Healthcheck.ReplicationIntervalSeconds.Get()); err != nil {
 		log.Exitf("failed to parse -tablet-path or initialize DB credentials: %v", err)
 	}
 	servenv.OnClose(func() {

go/vt/vttablet/tabletserver/tabletenv/config.go

@@ -40,6 +40,8 @@ const (
 	NotOnMaster = "notOnMaster"
 	Polling     = "polling"
 	Heartbeat   = "heartbeat"
+
+	ReplicationHealthCheckIntervalSecondsFlagName = "replication_health_check_interval"
 )
 
 var (
@@ -69,17 +71,18 @@ var (
 	deprecatedFoundRowsPoolSize             int
 
 	// The following vars are used for custom initialization of Tabletconfig.
-	enableHotRowProtection       bool
-	enableHotRowProtectionDryRun bool
-	enableConsolidator           bool
-	enableConsolidatorReplicas   bool
-	enableHeartbeat              bool
-	heartbeatInterval            time.Duration
-	healthCheckInterval          time.Duration
-	degradedThreshold            time.Duration
-	unhealthyThreshold           time.Duration
-	transitionGracePeriod        time.Duration
-	enableReplicationReporter    bool
+	enableHotRowProtection         bool
+	enableHotRowProtectionDryRun   bool
+	enableConsolidator             bool
+	enableConsolidatorReplicas     bool
+	enableHeartbeat                bool
+	heartbeatInterval              time.Duration
+	healthCheckInterval            time.Duration
+	replicationHealthCheckInterval time.Duration
+	degradedThreshold              time.Duration
+	unhealthyThreshold             time.Duration
+	transitionGracePeriod          time.Duration
+	enableReplicationReporter      bool
 )
 
 func init() {
@@ -156,6 +159,7 @@ func init() {
 	flagutil.DualFormatBoolVar(&currentConfig.CacheResultFields, "enable_query_plan_field_caching", defaultConfig.CacheResultFields, "This option fetches & caches fields (columns) when storing query plans")
 
 	flag.DurationVar(&healthCheckInterval, "health_check_interval", 20*time.Second, "Interval between health checks")
+	flag.DurationVar(&replicationHealthCheckInterval, ReplicationHealthCheckIntervalSecondsFlagName, 20*time.Second, "Interval at which ReplicationManager checks replication health")
 	flag.DurationVar(&degradedThreshold, "degraded_threshold", 30*time.Second, "replication lag after which a replica is considered degraded")
 	flag.DurationVar(&unhealthyThreshold, "unhealthy_threshold", 2*time.Hour, "replication lag after which a replica is considered unhealthy")
 	flag.DurationVar(&transitionGracePeriod, "serving_state_grace_period", 0, "how long to pause after broadcasting health to vtgate, before enforcing a new serving state")
@@ -207,6 +211,13 @@ func Init() {
 	}
 
 	currentConfig.Healthcheck.IntervalSeconds.Set(healthCheckInterval)
+
+	var actualReplHealthCheckInterval = healthCheckInterval
+	if flagWasPassed(ReplicationHealthCheckIntervalSecondsFlagName) {
+		actualReplHealthCheckInterval = replicationHealthCheckInterval
+	}
+	currentConfig.Healthcheck.ReplicationIntervalSeconds.Set(actualReplHealthCheckInterval)
+
 	currentConfig.Healthcheck.DegradedThresholdSeconds.Set(degradedThreshold)
 	currentConfig.Healthcheck.UnhealthyThresholdSeconds.Set(unhealthyThreshold)
 	currentConfig.GracePeriods.TransitionSeconds.Set(transitionGracePeriod)
@@ -309,9 +320,10 @@ type HotRowProtectionConfig struct {
 
 // HealthcheckConfig contains the config for healthcheck.
 type HealthcheckConfig struct {
-	IntervalSeconds           Seconds `json:"intervalSeconds,omitempty"`
-	DegradedThresholdSeconds  Seconds `json:"degradedThresholdSeconds,omitempty"`
-	UnhealthyThresholdSeconds Seconds `json:"unhealthyThresholdSeconds,omitempty"`
+	IntervalSeconds            Seconds `json:"intervalSeconds,omitempty"`
+	ReplicationIntervalSeconds Seconds `json:"replicationIntervalSeconds,omitempty"`
+	DegradedThresholdSeconds   Seconds `json:"degradedThresholdSeconds,omitempty"`
+	UnhealthyThresholdSeconds  Seconds `json:"unhealthyThresholdSeconds,omitempty"`
 }
 
 // GracePeriodsConfig contains various grace periods.
@@ -433,9 +445,10 @@ var defaultConfig = TabletConfig{
 		MaxRows:             10000,
 	},
 	Healthcheck: HealthcheckConfig{
-		IntervalSeconds:           20,
-		DegradedThresholdSeconds:  30,
-		UnhealthyThresholdSeconds: 7200,
+		IntervalSeconds:            20,
+		ReplicationIntervalSeconds: 20,
+		DegradedThresholdSeconds:   30,
+		UnhealthyThresholdSeconds:  7200,
 	},
 	ReplicationTracker: ReplicationTrackerConfig{
 		Mode:                     Disable,
@@ -508,3 +521,13 @@ func defaultTransactionLimitConfig() TransactionLimitConfig {
 		TransactionLimitBySubcomponent: false,
 	}
 }
+
+func flagWasPassed(name string) bool {
+	found := false
+	flag.Visit(func(f *flag.Flag) {
+		if f.Name == name {
+			found = true
+		}
+	})
+	return found
+}

go/vt/vttablet/tabletserver/tabletenv/config_test.go

@@ -117,6 +117,7 @@ gracePeriods: {}
 healthcheck:
   degradedThresholdSeconds: 30
   intervalSeconds: 20
+  replicationIntervalSeconds: 20
   unhealthyThresholdSeconds: 7200
 hotRowProtection:
   maxConcurrency: 5
@@ -226,6 +227,7 @@ func TestFlags(t *testing.T) {
 	want.HotRowProtection.Mode = Disable
 	want.Consolidator = Enable
 	want.Healthcheck.IntervalSeconds = 20
+	want.Healthcheck.ReplicationIntervalSeconds = 20
 	want.Healthcheck.DegradedThresholdSeconds = 30
 	want.Healthcheck.UnhealthyThresholdSeconds = 7200
 	want.ReplicationTracker.HeartbeatIntervalSeconds = 1
@@ -309,9 +311,12 @@ func TestFlags(t *testing.T) {
 	assert.Equal(t, want, currentConfig)
 
 	healthCheckInterval = 1 * time.Second
+	replicationHealthCheckInterval = 2 * time.Second
 	currentConfig.Healthcheck.IntervalSeconds = 0
+	currentConfig.Healthcheck.ReplicationIntervalSeconds = 0
 	Init()
 	want.Healthcheck.IntervalSeconds = 1
+	want.Healthcheck.ReplicationIntervalSeconds = 1
 	assert.Equal(t, want, currentConfig)
 
 	degradedThreshold = 2 * time.Second