Merge pull request #204 from hashicorp/f-check-freq

Add deferred sync for check output

command/agent/config.go

@@ -173,6 +173,14 @@ type Config struct {
 	// true, we ignore the leave, and rejoin the cluster on start.
 	RejoinAfterLeave bool `mapstructure:"rejoin_after_leave"`
 
+	// CheckUpdateInterval controls the interval on which the output of a health check
+	// is updated if there is no change to the state. For example, a check in a steady
+	// state may run every 5 second generating a unique output (timestamp, etc), forcing
+	// constant writes. This allows Consul to defer the write for some period of time,
+	// reducing the write pressure when the state is steady.
+	CheckUpdateInterval    time.Duration `mapstructure:"-"`
+	CheckUpdateIntervalRaw string        `mapstructure:"check_update_interval" json:"-"`
+
 	// AEInterval controls the anti-entropy interval. This is how often
 	// the agent attempts to reconcile it's local state with the server'
 	// representation of our state. Defaults to every 60s.
@@ -221,6 +229,7 @@ func DefaultConfig() *Config {
 			MaxStale: 5 * time.Second,
 		},
 		Protocol:            consul.ProtocolVersionMax,
+		CheckUpdateInterval: 5 * time.Minute,
 		AEInterval:          time.Minute,
 	}
 }
@@ -309,6 +318,14 @@ func DecodeConfig(r io.Reader) (*Config, error) {
 		}
 	}
 
+	if raw := result.CheckUpdateIntervalRaw; raw != "" {
+		dur, err := time.ParseDuration(raw)
+		if err != nil {
+			return nil, fmt.Errorf("CheckUpdateInterval invalid: %v", err)
+		}
+		result.CheckUpdateInterval = dur
+	}
+
 	return &result, nil
 }
 
@@ -536,6 +553,9 @@ func MergeConfig(a, b *Config) *Config {
 	if b.DNSConfig.MaxStale != 0 {
 		result.DNSConfig.MaxStale = b.DNSConfig.MaxStale
 	}
+	if b.CheckUpdateIntervalRaw != "" {
+		result.CheckUpdateInterval = b.CheckUpdateInterval
+	}
 
 	// Copy the start join addresses
 	result.StartJoin = make([]string, 0, len(a.StartJoin)+len(b.StartJoin))

command/agent/config_test.go

@@ -324,6 +324,17 @@ func TestDecodeConfig(t *testing.T) {
 	if config.DNSConfig.ServiceTTL["web"] != 30*time.Second {
 		t.Fatalf("bad: %#v", config)
 	}
+
+	// CheckUpdateInterval
+	input = `{"check_update_interval": "10m"}`
+	config, err = DecodeConfig(bytes.NewReader([]byte(input)))
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	if config.CheckUpdateInterval != 10*time.Minute {
+		t.Fatalf("bad: %#v", config)
+	}
 }
 
 func TestDecodeConfig_Service(t *testing.T) {
@@ -422,6 +433,7 @@ func TestMergeConfig(t *testing.T) {
 		LeaveOnTerm:            false,
 		SkipLeaveOnInt:         false,
 		EnableDebug:            false,
+		CheckUpdateIntervalRaw: "8m",
 	}
 
 	b := &Config{
@@ -466,6 +478,8 @@ func TestMergeConfig(t *testing.T) {
 		UiDir:                  "/opt/consul-ui",
 		EnableSyslog:           true,
 		RejoinAfterLeave:       true,
+		CheckUpdateInterval:    8 * time.Minute,
+		CheckUpdateIntervalRaw: "8m",
 	}
 
 	c := MergeConfig(a, b)

command/agent/local.go

@@ -20,6 +20,7 @@ const (
 type syncStatus struct {
 	remoteDelete bool        // Should this be deleted from the server
 	inSync       bool        // Is this in sync with the server
+	deferSync    *time.Timer // Defer sync until this time
 }
 
 // localState is used to represent the node's services,
@@ -191,6 +192,25 @@ func (l *localState) UpdateCheck(checkID, status, output string) {
 		return
 	}
 
+	// Defer a sync if the output has changed. This is an optimization around
+	// frequent updates of output. Instead, we update the output internally,
+	// and periodically do a write-back to the servers. If there is a status
+	// change we do the write immediately.
+	if l.config.CheckUpdateInterval > 0 && check.Status == status {
+		check.Output = output
+		status := l.checkStatus[checkID]
+		if status.deferSync == nil && status.inSync {
+			deferSync := time.AfterFunc(l.config.CheckUpdateInterval, func() {
+				l.Lock()
+				l.checkStatus[checkID] = syncStatus{inSync: false}
+				l.changeMade()
+				l.Unlock()
+			})
+			l.checkStatus[checkID] = syncStatus{inSync: true, deferSync: deferSync}
+		}
+		return
+	}
+
 	// Update status and mark out of sync
 	check.Status = status
 	check.Output = output

command/agent/local_test.go

@@ -252,3 +252,77 @@ func TestAgentAntiEntropy_Checks(t *testing.T) {
 		}
 	}
 }
+
+func TestAgentAntiEntropy_Check_DeferSync(t *testing.T) {
+	conf := nextConfig()
+	conf.CheckUpdateInterval = 100 * time.Millisecond
+	dir, agent := makeAgent(t, conf)
+	defer os.RemoveAll(dir)
+	defer agent.Shutdown()
+
+	testutil.WaitForLeader(t, agent.RPC, "dc1")
+
+	// Create a check
+	check := &structs.HealthCheck{
+		Node:    agent.config.NodeName,
+		CheckID: "web",
+		Name:    "web",
+		Status:  structs.HealthPassing,
+		Output:  "",
+	}
+	agent.state.AddCheck(check)
+
+	// Trigger anti-entropy run and wait
+	agent.StartSync()
+	time.Sleep(200 * time.Millisecond)
+
+	// Verify that we are in sync
+	req := structs.NodeSpecificRequest{
+		Datacenter: "dc1",
+		Node:       agent.config.NodeName,
+	}
+	var checks structs.IndexedHealthChecks
+	if err := agent.RPC("Health.NodeChecks", &req, &checks); err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
+	// Verify checks in place
+	if len(checks.HealthChecks) != 2 {
+		t.Fatalf("checks: %v", check)
+	}
+
+	// Update the check output! Should be defered
+	agent.state.UpdateCheck("web", structs.HealthPassing, "output")
+
+	// Should not update for 100 milliseconds
+	time.Sleep(50 * time.Millisecond)
+	if err := agent.RPC("Health.NodeChecks", &req, &checks); err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
+	// Verify not updated
+	for _, chk := range checks.HealthChecks {
+		switch chk.CheckID {
+		case "web":
+			if chk.Output != "" {
+				t.Fatalf("early update: %v", chk)
+			}
+		}
+	}
+
+	// Wait for a defered update
+	time.Sleep(100 * time.Millisecond)
+	if err := agent.RPC("Health.NodeChecks", &req, &checks); err != nil {
+		t.Fatalf("err: %v", err)
+	}
+
+	// Verify not updated
+	for _, chk := range checks.HealthChecks {
+		switch chk.CheckID {
+		case "web":
+			if chk.Output != "output" {
+				t.Fatalf("no update: %v", chk)
+			}
+		}
+	}
+}

website/source/docs/agent/options.html.markdown

@@ -179,6 +179,14 @@ definitions support being updated during a reload.
   The certificate is provided to clients or servers to verify the agents authenticity.
   Must be provided along with the `key_file`.
 
+* `check_update_interval` - This interval controls how often check output from
+  checks in a steady state is syncronized with the server. By default, this is
+  set to 5 minutes ("5m"). Many checks which are in a steady state produce
+  slightly different output per run (timestamps, etc) which cause constant writes.
+  This configuration allows defering the sync of check output for a given interval to
+  reduce write pressure. If a check ever changes state, the new state and associated
+  output is syncronized immediately. To disable this behavior, set the value to "0s".
+
 * `domain` - By default, Consul responds to DNS queries in the "consul." domain.
   This flag can be used to change that domain. All queries in this domain are assumed
   to be handled by Consul, and will not be recursively resolved.