telemetry: Honor the server's request to toggle certain metrics

telemetry/collection.go

@@ -104,7 +104,7 @@ func Reset() {
 // go keyword after the call to SendHello so it
 // doesn't block crucial code.
 func Set(key string, val interface{}) {
-	if !enabled {
+	if !enabled || isDisabled(key) {
 		return
 	}
 	bufferMu.Lock()
@@ -123,10 +123,8 @@ func Set(key string, val interface{}) {
 // If key is new, a new list will be created.
 // If key maps to a type that is not a list,
 // a panic is logged, and this is a no-op.
-//
-// TODO: is this function needed/useful?
 func Append(key string, value interface{}) {
-	if !enabled {
+	if !enabled || isDisabled(key) {
 		return
 	}
 	bufferMu.Lock()
@@ -161,7 +159,7 @@ func Append(key string, value interface{}) {
 // that is not a counting set, a panic is logged,
 // and this is a no-op.
 func AppendUnique(key string, value interface{}) {
-	if !enabled {
+	if !enabled || isDisabled(key) {
 		return
 	}
 	bufferMu.Lock()
@@ -204,7 +202,7 @@ func Increment(key string) {
 // atomicAdd adds amount (negative to subtract)
 // to key.
 func atomicAdd(key string, amount int) {
-	if !enabled {
+	if !enabled || isDisabled(key) {
 		return
 	}
 	bufferMu.Lock()
@@ -225,3 +223,14 @@ func atomicAdd(key string, amount int) {
 	buffer[key] = intVal + amount
 	bufferMu.Unlock()
 }
+
+// isDisabled returns whether key is
+// a disabled metric key. ALL collection
+// functions should call this and not
+// save the value if this returns true.
+func isDisabled(key string) bool {
+	disabledMetricsMu.RLock()
+	_, ok := disabledMetrics[key]
+	disabledMetricsMu.RUnlock()
+	return ok
+}

telemetry/telemetry.go

@@ -155,6 +155,18 @@ func emit(final bool) error {
 			continue
 		}
 
+		// update the list of enabled/disabled keys, if any
+		for _, key := range reply.EnableKeys {
+			disabledMetricsMu.Lock()
+			delete(disabledMetrics, key)
+			disabledMetricsMu.Unlock()
+		}
+		for _, key := range reply.DisableKeys {
+			disabledMetricsMu.Lock()
+			disabledMetrics[key] = struct{}{}
+			disabledMetricsMu.Unlock()
+		}
+
 		// make sure we didn't send the update too soon; if so,
 		// just wait and try again -- this is a special case of
 		// error that we handle differently, as you can see
@@ -259,6 +271,18 @@ type Response struct {
 	// Error will be populated with an error message, if any.
 	// This field should be empty if the status code is < 400.
 	Error string `json:"error,omitempty"`
+
+	// DisableKeys will contain a list of keys/metrics that
+	// should NOT be sent until further notice. The client
+	// must NOT store these items in its buffer or send them
+	// to the telemetry server while they are disabled. If
+	// this list and EnableKeys have the same value (which is
+	// not supposed to happen), this field should dominate.
+	DisableKeys []string `json:"disable_keys,omitempty"`
+
+	// EnableKeys will contain a list of keys/metrics that
+	// MAY be sent until further notice.
+	EnableKeys []string `json:"enable_keys,omitempty"`
 }
 
 // Payload is the data that gets sent to the telemetry server.
@@ -335,6 +359,12 @@ var (
 	updateTimer   *time.Timer
 	updateTimerMu sync.Mutex
 
+	// disabledMetrics is a list of metric keys
+	// that should NOT be saved to the buffer
+	// or sent to the telemetry server.
+	disabledMetrics   = make(map[string]struct{})
+	disabledMetricsMu sync.RWMutex
+
 	// instanceUUID is the ID of the current instance.
 	// This MUST be set to emit telemetry.
 	// This MUST NOT be openly exposed to clients, for privacy.