Add dry-run/monitoring-only mode for TxThrottler

changelog/18.0/18.0.0/summary.md

@@ -89,6 +89,14 @@ Vitess upgrade process from an earlier version if you need to use such a workflo
 Any MoveTables or Migrate workflow that moves a sequence table should only be run after all vitess components have been
 upgraded, and no upgrade should be done while such a workflow is in progress.
 
+#### <a id="vttablet-tx-throttler-dry-run"/>New Dry-run/monitoring-only mode for the transaction throttler
+
+A new CLI flag `--tx-throttler-dry-run` to set the Transaction Throttler to monitoring-only/dry-run mode has been added.
+If the transaction throttler is enabled with `--enable-tx-throttler` and the new dry-run flag is also specified, the
+tablet will not actually throttle any transactions; however, it will increase the counters for transactions throttled
+(`vttablet_transaction_throttler_throttled`). This allows to deploy the transaction throttler in production and gain
+observability on how much throttling would take place, without actually throttling any requests.
+
 ### <a id="docker"/>Docker
 
 #### <a id="debian-bookworm"/>Bookworm added and made default

go/flags/endtoend/vttablet.txt

@@ -347,6 +347,7 @@ Usage of vttablet:
       --twopc_enable                                                     if the flag is on, 2pc is enabled. Other 2pc flags must be supplied.
       --tx-throttler-config string                                       Synonym to -tx_throttler_config (default "target_replication_lag_sec:2 max_replication_lag_sec:10 initial_rate:100 max_increase:1 emergency_decrease:0.5 min_duration_between_increases_sec:40 max_duration_between_increases_sec:62 min_duration_between_decreases_sec:20 spread_backlog_across_sec:20 age_bad_rate_after_sec:180 bad_rate_increase:0.1 max_rate_approach_threshold:0.9")
       --tx-throttler-default-priority int                                Default priority assigned to queries that lack priority information (default 100)
+      --tx-throttler-dry-run                                             If present, the TxThrottler only records metrics about requests received and throttled, but does not actually throttle any requests.
       --tx-throttler-healthcheck-cells strings                           Synonym to -tx_throttler_healthcheck_cells
       --tx-throttler-tablet-types strings                                A comma-separated list of tablet types. Only tablets of this type are monitored for replication lag by the transaction throttler. Supported types are replica and/or rdonly. (default replica)
       --tx_throttler_config string                                       The configuration of the transaction throttler as a text-formatted throttlerdata.Configuration protocol buffer message. (default "target_replication_lag_sec:2 max_replication_lag_sec:10 initial_rate:100 max_increase:1 emergency_decrease:0.5 min_duration_between_increases_sec:40 max_duration_between_increases_sec:62 min_duration_between_decreases_sec:20 spread_backlog_across_sec:20 age_bad_rate_after_sec:180 bad_rate_increase:0.1 max_rate_approach_threshold:0.9")

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

@@ -185,6 +185,7 @@ func registerTabletEnvFlags(fs *pflag.FlagSet) {
 	flagutil.DualFormatStringListVar(fs, &currentConfig.TxThrottlerHealthCheckCells, "tx_throttler_healthcheck_cells", defaultConfig.TxThrottlerHealthCheckCells, "A comma-separated list of cells. Only tabletservers running in these cells will be monitored for replication lag by the transaction throttler.")
 	fs.IntVar(&currentConfig.TxThrottlerDefaultPriority, "tx-throttler-default-priority", defaultConfig.TxThrottlerDefaultPriority, "Default priority assigned to queries that lack priority information")
 	fs.Var(currentConfig.TxThrottlerTabletTypes, "tx-throttler-tablet-types", "A comma-separated list of tablet types. Only tablets of this type are monitored for replication lag by the transaction throttler. Supported types are replica and/or rdonly.")
+	fs.BoolVar(&currentConfig.TxThrottlerDryRun, "tx-throttler-dry-run", defaultConfig.TxThrottlerDryRun, "If present, the TxThrottler only records metrics about requests received and throttled, but does not actually throttle any requests.")
 
 	fs.BoolVar(&enableHotRowProtection, "enable_hot_row_protection", false, "If true, incoming transactions for the same row (range) will be queued and cannot consume all txpool slots.")
 	fs.BoolVar(&enableHotRowProtectionDryRun, "enable_hot_row_protection_dry_run", false, "If true, hot row protection is not enforced but logs if transactions would have been queued.")
@@ -364,6 +365,7 @@ type TabletConfig struct {
 	TxThrottlerHealthCheckCells []string                      `json:"-"`
 	TxThrottlerDefaultPriority  int                           `json:"-"`
 	TxThrottlerTabletTypes      *topoproto.TabletTypeListFlag `json:"-"`
+	TxThrottlerDryRun           bool                          `json:"-"`
 
 	EnableTableGC bool `json:"-"` // can be turned off programmatically by tests
 
@@ -832,6 +834,7 @@ var defaultConfig = TabletConfig{
 	TxThrottlerHealthCheckCells: []string{},
 	TxThrottlerDefaultPriority:  sqlparser.MaxPriorityValue, // This leads to all queries being candidates to throttle
 	TxThrottlerTabletTypes:      &topoproto.TabletTypeListFlag{topodatapb.TabletType_REPLICA},
+	TxThrottlerDryRun:           false,
 
 	TransactionLimitConfig: defaultTransactionLimitConfig(),
 

go/vt/vttablet/tabletserver/txthrottler/tx_throttler.go

@@ -136,7 +136,7 @@ type txThrottler struct {
 
 	// state holds an open transaction throttler state. It is nil
 	// if the TransactionThrottler is closed.
-	state *txThrottlerState
+	state txThrottlerState
 
 	target     *querypb.Target
 	topoServer *topo.Server
@@ -158,6 +158,10 @@ type txThrottlerConfig struct {
 	// returns false.
 	enabled bool
 
+	// if dryRun is true, the txThrottler will run only on monitoring mode, meaning that it will increase counters for
+	// total and actually throttled requests, but it will not actually return that a transaction should be throttled.
+	dryRun bool
+
 	throttlerConfig *throttlerdatapb.Configuration
 	// healthCheckCells stores the cell names in which running vttablets will be monitored for
 	// replication lag.
@@ -167,8 +171,14 @@ type txThrottlerConfig struct {
 	tabletTypes map[topodatapb.TabletType]bool
 }
 
-// txThrottlerState holds the state of an open TxThrottler object.
-type txThrottlerState struct {
+type txThrottlerState interface {
+	deallocateResources()
+	StatsUpdate(tabletStats *discovery.TabletHealth)
+	throttle() bool
+}
+
+// txThrottlerStateImpl holds the state of an open TxThrottler object.
+type txThrottlerStateImpl struct {
 	config      *txThrottlerConfig
 	txThrottler *txThrottler
 
@@ -202,6 +212,7 @@ func NewTxThrottler(env tabletenv.Env, topoServer *topo.Server) TxThrottler {
 
 		throttlerConfig = &txThrottlerConfig{
 			enabled:          true,
+			dryRun:           env.Config().TxThrottlerDryRun,
 			tabletTypes:      tabletTypes,
 			throttlerConfig:  env.Config().TxThrottlerConfig.Get(),
 			healthCheckCells: healthCheckCells,
@@ -280,10 +291,10 @@ func (t *txThrottler) Throttle(priority int) (result bool) {
 		t.requestsThrottled.Add(1)
 	}
 
-	return result
+	return result && !t.config.dryRun
 }
 
-func newTxThrottlerState(txThrottler *txThrottler, config *txThrottlerConfig, target *querypb.Target) (*txThrottlerState, error) {
+func newTxThrottlerState(txThrottler *txThrottler, config *txThrottlerConfig, target *querypb.Target) (txThrottlerState, error) {
 	maxReplicationLagModuleConfig := throttler.MaxReplicationLagModuleConfig{Configuration: config.throttlerConfig}
 
 	t, err := throttlerFactory(
@@ -300,7 +311,7 @@ func newTxThrottlerState(txThrottler *txThrottler, config *txThrottlerConfig, ta
 		t.Close()
 		return nil, err
 	}
-	state := &txThrottlerState{
+	state := &txThrottlerStateImpl{
 		config:      config,
 		throttler:   t,
 		txThrottler: txThrottler,
@@ -324,7 +335,7 @@ func newTxThrottlerState(txThrottler *txThrottler, config *txThrottlerConfig, ta
 	return state, nil
 }
 
-func createTxThrottlerHealthCheck(topoServer *topo.Server, config *txThrottlerConfig, result *txThrottlerState, cell string) {
+func createTxThrottlerHealthCheck(topoServer *topo.Server, config *txThrottlerConfig, result *txThrottlerStateImpl, cell string) {
 	ctx, cancel := context.WithCancel(context.Background())
 	result.stopHealthCheck = cancel
 	result.healthCheck = healthCheckFactory(topoServer, cell, config.healthCheckCells)
@@ -341,7 +352,7 @@ func createTxThrottlerHealthCheck(topoServer *topo.Server, config *txThrottlerCo
 	}(ctx)
 }
 
-func (ts *txThrottlerState) throttle() bool {
+func (ts *txThrottlerStateImpl) throttle() bool {
 	if ts.throttler == nil {
 		log.Error("throttle called after deallocateResources was called")
 		return false
@@ -352,7 +363,7 @@ func (ts *txThrottlerState) throttle() bool {
 	return ts.throttler.Throttle(0 /* threadId */) > 0
 }
 
-func (ts *txThrottlerState) deallocateResources() {
+func (ts *txThrottlerStateImpl) deallocateResources() {
 	// We don't really need to nil out the fields here
 	// as deallocateResources is not expected to be called
 	// more than once, but it doesn't hurt to do so.
@@ -365,14 +376,14 @@ func (ts *txThrottlerState) deallocateResources() {
 	ts.healthCheck.Close()
 	ts.healthCheck = nil
 
-	// After ts.healthCheck is closed txThrottlerState.StatsUpdate() is guaranteed not
+	// After ts.healthCheck is closed txThrottlerStateImpl.StatsUpdate() is guaranteed not
 	// to be executing, so we can safely close the throttler.
 	ts.throttler.Close()
 	ts.throttler = nil
 }
 
 // StatsUpdate updates the health of a tablet with the given healthcheck.
-func (ts *txThrottlerState) StatsUpdate(tabletStats *discovery.TabletHealth) {
+func (ts *txThrottlerStateImpl) StatsUpdate(tabletStats *discovery.TabletHealth) {
 	if ts.config.tabletTypes == nil {
 		return
 	}

go/vt/vttablet/tabletserver/txthrottler/tx_throttler_test.go

@@ -187,3 +187,54 @@ func TestNewTxThrottler(t *testing.T) {
 		assert.Equal(t, []string{"cell1", "cell2"}, throttlerImpl.config.healthCheckCells)
 	}
 }
+
+func TestDryRunThrottler(t *testing.T) {
+	config := tabletenv.NewDefaultConfig()
+	env := tabletenv.NewEnv(config, t.Name())
+
+	testCases := []struct {
+		Name                           string
+		txThrottlerStateShouldThrottle bool
+		throttlerDryRun                bool
+		expectedResult                 bool
+	}{
+		{Name: "Real run throttles when txThrottlerStateImpl says it should", txThrottlerStateShouldThrottle: true, throttlerDryRun: false, expectedResult: true},
+		{Name: "Real run does not throttle when txThrottlerStateImpl says it should not", txThrottlerStateShouldThrottle: false, throttlerDryRun: false, expectedResult: false},
+		{Name: "Dry run does not throttle when txThrottlerStateImpl says it should", txThrottlerStateShouldThrottle: true, throttlerDryRun: true, expectedResult: false},
+		{Name: "Dry run does not throttle when txThrottlerStateImpl says it should not", txThrottlerStateShouldThrottle: false, throttlerDryRun: true, expectedResult: false},
+	}
+
+	for _, aTestCase := range testCases {
+		theTestCase := aTestCase
+
+		t.Run(theTestCase.Name, func(t *testing.T) {
+			aTxThrottler := &txThrottler{
+				config: &txThrottlerConfig{
+					enabled: true,
+					dryRun:  theTestCase.throttlerDryRun,
+				},
+				state:             &mockTxThrottlerState{shouldThrottle: theTestCase.txThrottlerStateShouldThrottle},
+				throttlerRunning:  env.Exporter().NewGauge("TransactionThrottlerRunning", "transaction throttler running state"),
+				requestsTotal:     env.Exporter().NewCounter("TransactionThrottlerRequests", "transaction throttler requests"),
+				requestsThrottled: env.Exporter().NewCounter("TransactionThrottlerThrottled", "transaction throttler requests throttled"),
+			}
+
+			assert.Equal(t, theTestCase.expectedResult, aTxThrottler.Throttle(100))
+		})
+	}
+}
+
+type mockTxThrottlerState struct {
+	shouldThrottle bool
+}
+
+func (t *mockTxThrottlerState) deallocateResources() {
+
+}
+func (t *mockTxThrottlerState) StatsUpdate(tabletStats *discovery.TabletHealth) {
+
+}
+
+func (t *mockTxThrottlerState) throttle() bool {
+	return t.shouldThrottle
+}