Support for starting a single beacon or no beacon on node startup (#1097

)

* Support for starting a single beacon or no beacon on node startup
To reload a beacon, one can use:
```
drand load --control <node-port> --id <beacon-name>
```
Fixes #1046
* Emit logs when starting node with no active beacons

cmd/drand-cli/cli.go

@@ -360,7 +360,7 @@ var appCommands = []*cli.Command{
 		Flags: toArray(folderFlag, tlsCertFlag, tlsKeyFlag,
 			insecureFlag, controlFlag, privListenFlag, pubListenFlag, metricsFlag,
 			certsDirFlag, pushFlag, verboseFlag, oldGroupFlag,
-			skipValidationFlag, jsonFlag,
+			skipValidationFlag, jsonFlag, beaconIDFlag,
 			storageTypeFlag, pgDSNFlag),
 		Action: func(c *cli.Context) error {
 			banner()
@@ -393,7 +393,7 @@ var appCommands = []*cli.Command{
 	{
 		Name:   "load",
 		Usage:  "Launch a sharing protocol from filesystem",
-		Flags:  toArray(controlFlag, beaconIDFlag),
+		Flags:  toArray(controlFlag, beaconIDFlag, insecureFlag),
 		Action: loadCmd,
 	},
 	{

cmd/drand-cli/cli_test.go

@@ -483,7 +483,6 @@ func testStartedDrandFunctional(t *testing.T, ctrlPort, rootPath, address string
 	require.Error(t, err)
 }
 
-//nolint:unused // We want to provide convenience functions
 func testPing(t *testing.T, ctrlPort string) {
 	t.Helper()
 
@@ -511,13 +510,27 @@ func testStatus(t *testing.T, ctrlPort, beaconID string) {
 		status := []string{"drand", "util", "status", "--control", ctrlPort, "--id", beaconID}
 		err = CLI().Run(status)
 		if err == nil {
-			break
+			return
 		}
 		time.Sleep(500 * time.Millisecond)
 	}
 	require.NoError(t, err)
 }
 
+func testFailStatus(t *testing.T, ctrlPort, beaconID string) {
+	t.Helper()
+
+	var err error
+
+	t.Logf(" + running STATUS command with %s on beacon [%s]", ctrlPort, beaconID)
+	for i := 0; i < 3; i++ {
+		status := []string{"drand", "util", "status", "--control", ctrlPort, "--id", beaconID}
+		err = CLI().Run(status)
+		require.Error(t, err)
+		time.Sleep(500 * time.Millisecond)
+	}
+}
+
 //nolint:unused // We want to provide convenience functions
 func testListSchemes(t *testing.T, ctrlPort string) {
 	t.Helper()
@@ -698,7 +711,6 @@ func TestDrandListSchemes(t *testing.T) {
 }
 
 func TestDrandReloadBeacon(t *testing.T) {
-	t.Skipf("test fails when error checking commands")
 	sch := scheme.GetSchemeFromEnv()
 	beaconID := test.GetBeaconIDFromEnv()
 
@@ -707,49 +719,95 @@ func TestDrandReloadBeacon(t *testing.T) {
 
 	for i, inst := range instances {
 		if i == 0 {
-			inst.shareLeader(t, n, n, 2, beaconID, sch)
+			inst.shareLeader(t, n, n, 1, beaconID, sch)
 		} else {
 			inst.share(t, instances[0].addr, beaconID)
 		}
-		time.Sleep(500 * time.Millisecond)
 	}
 
+	t.Log("waiting for initial set up to settle on all nodes")
+	time.Sleep(3 * time.Second)
+
 	defer func() {
 		for _, inst := range instances {
-			err := inst.stopAll()
-			require.NoError(t, err)
+			// We want to ignore this error, at least until the stop command won't return an error
+			// when correctly running the stop command.
+			_ = inst.stopAll()
 		}
 	}()
 
-	time.Sleep(1 * time.Second)
+	t.Log("waiting for initial setup to finish")
+	time.Sleep(5 * time.Second)
 
 	// try to reload a beacon which is already loaded
 	err := instances[3].load(beaconID)
 	require.Error(t, err)
 
-	// wait some time to generate some randomness
-	time.Sleep(1 * time.Minute)
-
 	// Stop beacon process... not the entire node
 	err = instances[3].stop(beaconID)
 	require.NoError(t, err)
 
 	// check the node is still alive
 	testPing(t, instances[3].ctrlPort)
 
+	t.Log("waiting for beacons to be generated while a beacon process is stopped on a node")
+	time.Sleep(10 * time.Second)
+
 	// reload a beacon
 	err = instances[3].load(beaconID)
 	require.NoError(t, err)
 
 	// test beacon process status
 	testStatus(t, instances[3].ctrlPort, beaconID)
 
-	time.Sleep(5 * time.Second)
+	time.Sleep(3 * time.Second)
 
 	// test beacon process status
 	testStatus(t, instances[3].ctrlPort, beaconID)
 }
 
+func TestDrandLoadNotPresentBeacon(t *testing.T) {
+	sch := scheme.GetSchemeFromEnv()
+	beaconID := test.GetBeaconIDFromEnv()
+
+	n := 4
+	instances := launchDrandInstances(t, n)
+
+	for i, inst := range instances {
+		if i == 0 {
+			inst.shareLeader(t, n, n, 1, beaconID, sch)
+		} else {
+			inst.share(t, instances[0].addr, beaconID)
+		}
+	}
+
+	t.Log("waiting for initial set up to settle on all nodes")
+	time.Sleep(3 * time.Second)
+
+	defer func() {
+		for _, inst := range instances {
+			_ = inst.stopAll()
+		}
+	}()
+
+	t.Log("waiting for initial setup to finish")
+	time.Sleep(5 * time.Second)
+
+	// Stop beacon process... not the entire node
+	err := instances[3].stop(beaconID)
+	require.NoError(t, err)
+
+	t.Log("waiting for beacons to be generated while a beacon process is stopped on a node")
+	time.Sleep(10 * time.Second)
+
+	// reload a different beacon
+	err = instances[3].load("not-a-valid-beacon-name-here")
+	require.Error(t, err)
+
+	// test original beacon process status and hope it's still off
+	testFailStatus(t, instances[3].ctrlPort, beaconID)
+}
+
 func TestDrandStatus(t *testing.T) {
 	t.Skipf("test fails when error checking commands")
 	n := 4
@@ -847,13 +905,11 @@ func (d *drandInstance) stopAll() error {
 	return CLI().Run([]string{"drand", "stop", "--control", d.ctrlPort})
 }
 
-//nolint:unused // We want to provide convenience functions
 func (d *drandInstance) stop(beaconID string) error {
 	return CLI().Run([]string{"drand", "stop", "--control", d.ctrlPort, "--id", beaconID})
 }
 
-//nolint:unused // We want to provide convenience functions
-func (d *drandInstance) shareLeader(t *testing.T, nodes, threshold, period int, beaconID string, sch scheme.Scheme) {
+func (d *drandInstance) shareLeader(t *testing.T, nodes, threshold, periodSeconds int, beaconID string, sch scheme.Scheme) {
 	t.Helper()
 
 	shareArgs := []string{
@@ -862,7 +918,7 @@ func (d *drandInstance) shareLeader(t *testing.T, nodes, threshold, period int,
 		"--leader",
 		"--nodes", strconv.Itoa(nodes),
 		"--threshold", strconv.Itoa(threshold),
-		"--period", fmt.Sprintf("%ds", period),
+		"--period", fmt.Sprintf("%ds", periodSeconds),
 		"--control", d.ctrlPort,
 		"--scheme", sch.ID,
 		"--id", beaconID,
@@ -874,7 +930,6 @@ func (d *drandInstance) shareLeader(t *testing.T, nodes, threshold, period int,
 	}()
 }
 
-//nolint:unused // We want to provide convenience functions
 func (d *drandInstance) share(t *testing.T, leaderURL, beaconID string) {
 	t.Helper()
 
@@ -892,7 +947,6 @@ func (d *drandInstance) share(t *testing.T, leaderURL, beaconID string) {
 	}()
 }
 
-//nolint:unused // We want to provide convenience functions
 func (d *drandInstance) load(beaconID string) error {
 	reloadArgs := []string{
 		"drand",

cmd/drand-cli/daemon.go

@@ -17,8 +17,13 @@ func startCmd(c *cli.Context) error {
 		return fmt.Errorf("can't instantiate drand daemon %w", err)
 	}
 
+	singleBeacon := false
+	if c.IsSet(beaconIDFlag.Name) {
+		singleBeacon = true
+	}
+
 	// Check stores and start BeaconProcess
-	err = drandDaemon.LoadBeaconsFromDisk(c.String(metricsFlag.Name))
+	err = drandDaemon.LoadBeaconsFromDisk(c.String(metricsFlag.Name), singleBeacon, c.String(beaconIDFlag.Name))
 	if err != nil {
 		return fmt.Errorf("couldn't load existing beacons: %w", err)
 	}

core/drand_daemon.go

@@ -241,8 +241,17 @@ func (dd *DrandDaemon) RemoveBeaconHandler(beaconID string, bp *BeaconProcess) {
 }
 
 // LoadBeaconsFromDisk checks for existing stores and creates the corresponding BeaconProcess
-// accordingly to each stored BeaconID
-func (dd *DrandDaemon) LoadBeaconsFromDisk(metricsFlag string) error {
+// accordingly to each stored BeaconID.
+// When singleBeacon is set, and the singleBeaconName matches one of the stored beacons, then
+// only that beacon will be loaded.
+// If the singleBeaconName is an empty string, no beacon will be loaded.
+func (dd *DrandDaemon) LoadBeaconsFromDisk(metricsFlag string, singleBeacon bool, singleBeaconName string) error {
+	// Are we trying to start the daemon without any beacon running?
+	if singleBeacon && singleBeaconName == "" {
+		dd.log.Warnw("starting daemon with no active beacon")
+		return nil
+	}
+
 	// Load possible existing stores
 	stores, err := key.NewFileStores(dd.opts.ConfigFolderMB())
 	if err != nil {
@@ -251,7 +260,12 @@ func (dd *DrandDaemon) LoadBeaconsFromDisk(metricsFlag string) error {
 
 	metricsHandlers := make([]metrics.Handler, 0, len(stores))
 
+	startedAtLeastOne := false
 	for beaconID, fs := range stores {
+		if singleBeacon && singleBeaconName != beaconID {
+			continue
+		}
+
 		bp, err := dd.LoadBeaconFromStore(beaconID, fs)
 		if err != nil {
 			return err
@@ -261,6 +275,12 @@ func (dd *DrandDaemon) LoadBeaconsFromDisk(metricsFlag string) error {
 			bp.log.Infow("", "metrics", "adding handler")
 			metricsHandlers = append(metricsHandlers, bp.MetricsHandlerForPeer)
 		}
+
+		startedAtLeastOne = true
+	}
+
+	if !startedAtLeastOne {
+		dd.log.Warnw("starting daemon with no active beacon")
 	}
 
 	// Start metrics server
@@ -279,7 +299,7 @@ func (dd *DrandDaemon) LoadBeaconFromDisk(beaconID string) (*BeaconProcess, erro
 func (dd *DrandDaemon) LoadBeaconFromStore(beaconID string, store key.Store) (*BeaconProcess, error) {
 	bp, err := dd.InstantiateBeaconProcess(beaconID, store)
 	if err != nil {
-		dd.log.Error("beacon id", beaconID, "can't instantiate randomness beacon. err:", err)
+		dd.log.Errorw("beacon id", beaconID, "can't instantiate randomness beacon. err:", err)
 		return nil, err
 	}