move health check to separate client binary

12 files changed, 96 insertions, 43 deletions

diff --git a/Makefile b/Makefile
index ec3e756..4b50733 100644
--- a/Makefile
+++ b/Makefile
@@ -1,6 +1,6 @@
 GO ?= go
 all: build test
-build: dserver dcat dgrep dmap dtail
+build: dserver dcat dgrep dmap dtail dtailhealthcheck
 dserver:
 ifndef USE_ACL
 	${GO} build ${GO_FLAGS} -o dserver ./cmd/dserver/main.go
@@ -15,6 +15,8 @@ dmap:
 	${GO} build ${GO_FLAGS} -o dmap ./cmd/dmap/main.go
 dtail:
 	${GO} build ${GO_FLAGS} -o dtail ./cmd/dtail/main.go
+dtailhealthcheck:
+	${GO} build ${GO_FLAGS} -o dtailhealthcheck ./cmd/dtailhealthcheck/main.go
 install:
 ifndef USE_ACL
 	${GO} install ./cmd/dserver/main.go
@@ -25,6 +27,7 @@ endif
 	${GO} install ./cmd/dgrep/main.go
 	${GO} install ./cmd/dmap/main.go
 	${GO} install ./cmd/dtail/main.go
+	${GO} install ./cmd/dtailhealthcheck/main.go
 clean:
 	ls ./cmd/ | while read cmd; do \
 	  test -f $$cmd && rm $$cmd; \
diff --git a/cmd/dtail/main.go b/cmd/dtail/main.go
index d333825..301fc08 100644
--- a/cmd/dtail/main.go
+++ b/cmd/dtail/main.go
@@ -41,7 +41,7 @@ func main() {
 	flag.BoolVar(&args.RegexInvert, "invert", false, "Invert regex")
 	flag.BoolVar(&args.Spartan, "spartan", false, "Spartan output mode")
 	flag.BoolVar(&args.TrustAllHosts, "trustAllHosts", false, "Trust all unknown host keys")
-	flag.BoolVar(&checkHealth, "checkHealth", false, "Only check for server health")
+	flag.BoolVar(&checkHealth, "checkHealth", false, "Deprecated, flag will be removed soon")
 	flag.BoolVar(&displayColorTable, "colorTable", false, "Show color table")
 	flag.BoolVar(&displayWideColorTable, "wideColorTable", false, "Show a large color table")
 	flag.BoolVar(&displayVersion, "version", false, "Display version")
@@ -66,27 +66,23 @@ func main() {
 	if grep != "" {
 		args.RegexStr = grep
 	}
-	sourceProcess := source.Client
-	if checkHealth {
-		sourceProcess = source.HealthCheck
-	}
-	config.Setup(sourceProcess, &args, flag.Args())
-
+	config.Setup(source.Client, &args, flag.Args())
 	if displayVersion {
 		version.PrintAndExit()
 	}
 	if !args.Spartan {
+		if !checkHealth {
+			version.Print()
+		}
 		if displayWideColorTable {
 			color.TablePrintAndExit(true)
 		}
 		if displayColorTable {
 			color.TablePrintAndExit(false)
 		}
-		version.Print()
 	}
 
 	ctx, cancel := context.WithCancel(context.Background())
-
 	if shutdownAfter > 0 {
 		ctx, cancel = context.WithTimeout(ctx, time.Duration(shutdownAfter)*time.Second)
 		defer cancel()
@@ -94,12 +90,12 @@ func main() {
 
 	var wg sync.WaitGroup
 	wg.Add(1)
-	dlog.Start(ctx, &wg, sourceProcess, config.Common.LogLevel)
+	dlog.Start(ctx, &wg, source.Client, config.Common.LogLevel)
 
 	if checkHealth {
-		defer cancel()
-		healthClient, _ := clients.NewHealthClient(args)
-		os.Exit(healthClient.Start(ctx, signal.InterruptCh(ctx)))
+		fmt.Println("WARN: DTail health check has moved to separate binary dtailhealtcheck - please adjust the monitoring scripts!")
+		cancel()
+		os.Exit(1)
 	}
 
 	if pprof > -1 {
diff --git a/cmd/dtailhealthcheck/main.go b/cmd/dtailhealthcheck/main.go
new file mode 100644
index 0000000..e0cb795
--- /dev/null
+++ b/cmd/dtailhealthcheck/main.go
@@ -0,0 +1,31 @@
+package main
+
+import (
+	"context"
+	"flag"
+	"os"
+	"sync"
+
+	"github.com/mimecast/dtail/internal/clients"
+	"github.com/mimecast/dtail/internal/config"
+	"github.com/mimecast/dtail/internal/io/dlog"
+	"github.com/mimecast/dtail/internal/io/signal"
+	"github.com/mimecast/dtail/internal/source"
+)
+
+// The evil begins here.
+func main() {
+	var args config.Args
+	flag.StringVar(&args.ServersStr, "server", "", "Remote server to connect")
+	flag.Parse()
+	config.Setup(source.HealthCheck, &args, flag.Args())
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	var wg sync.WaitGroup
+	wg.Add(1)
+
+	dlog.Start(ctx, &wg, source.HealthCheck, config.Common.LogLevel)
+	healthClient, _ := clients.NewHealthClient(args)
+	os.Exit(healthClient.Start(ctx, signal.NoCh(ctx)))
+}
diff --git a/internal/clients/baseclient.go b/internal/clients/baseclient.go
index b474208..d5d7c2c 100644
--- a/internal/clients/baseclient.go
+++ b/internal/clients/baseclient.go
@@ -88,7 +88,6 @@ func (c *baseClient) Start(ctx context.Context, statsCh <-chan string) (status i
 		go func(i int, conn connectors.Connector) {
 			defer wg.Done()
 			connStatus := c.startConnection(ctx, i, conn)
-			// Update global status.
 			mutex.Lock()
 			defer mutex.Unlock()
 			if connStatus > status {
diff --git a/internal/clients/handlers/healthhandler.go b/internal/clients/handlers/healthhandler.go
index 4b16ce4..10ba1f7 100644
--- a/internal/clients/handlers/healthhandler.go
+++ b/internal/clients/handlers/healthhandler.go
@@ -1,7 +1,6 @@
 package handlers
 
 import (
-	"fmt"
 	"strings"
 
 	"github.com/mimecast/dtail/internal"
@@ -22,7 +21,7 @@ func NewHealthHandler(server string) *HealthHandler {
 			server:       server,
 			shellStarted: false,
 			commands:     make(chan string),
-			status:       -1,
+			status:       2, // Assume CRITICAL status by default.
 			done:         internal.NewDone(),
 		},
 	}
@@ -34,14 +33,12 @@ func (h *HealthHandler) Write(p []byte) (n int, err error) {
 		switch b {
 		case '\n', protocol.MessageDelimiter:
 			message := h.baseHandler.receiveBuf.String()
-			dlog.Client.Debug(message)
 			h.handleMessage(message)
 			h.baseHandler.receiveBuf.Reset()
 		default:
 			h.baseHandler.receiveBuf.WriteByte(b)
 		}
 	}
-
 	return len(p), nil
 }
 
@@ -52,21 +49,7 @@ func (h *HealthHandler) handleMessage(message string) {
 	}
 	s := strings.Split(message, protocol.FieldDelimiter)
 	message = s[len(s)-1]
-	status := strings.Split(message, ":")
-	fmt.Println(status)
-	/*
-		switch status {
-		case "OK":
-			h.HealthStatusCh <- 0
-		case "WARNING":
-			h.HealthStatusCh <- 1
-		case "CRITICAL":
-			h.HealthStatusCh <- 2
-		case "UNKNOWN":
-			h.HealthStatusCh <- 3
-		default:
-			fmt.Println("CRITICAL: Unexpected server response: '%s'")
-			h.HealthStatusCh <- 2
-		}
-	*/
+	if message == "OK" {
+		h.baseHandler.status = 0
+	}
 }
diff --git a/internal/clients/healthclient.go b/internal/clients/healthclient.go
index e2c8ccb..ac1dc20 100644
--- a/internal/clients/healthclient.go
+++ b/internal/clients/healthclient.go
@@ -1,6 +1,8 @@
 package clients
 
 import (
+	"context"
+	"fmt"
 	"runtime"
 
 	"github.com/mimecast/dtail/internal/clients/handlers"
@@ -42,3 +44,30 @@ func (c HealthClient) makeCommands() (commands []string) {
 	commands = append(commands, "health")
 	return
 }
+
+func (c *HealthClient) Start(ctx context.Context, statsCh <-chan string) int {
+	status := c.baseClient.Start(ctx, statsCh)
+
+	switch status {
+	case 0:
+		if c.Serverless {
+			fmt.Printf("WARNING: All seems fine but the check only run in serverless mode, please specify a remote server via --server hostname:port\n")
+			return 1
+		}
+		fmt.Printf("OK: All fine at %s :-)\n", c.ServersStr)
+	case 2:
+		if c.Serverless {
+			fmt.Printf("CRITICAL: DTail server not operating properly (using serverless connction)!\n")
+			return 2
+		}
+		fmt.Printf("CRITICAL: DTail server not operating properly at %s!\n", c.ServersStr)
+	default:
+		if c.Serverless {
+			fmt.Printf("UNKNOWN: Received unknown status code %d (using serverless connection)\n", status)
+			return status
+		}
+		fmt.Printf("UNKNOWN: Received unknown status code %d from %s!\n", status, c.ServersStr)
+	}
+
+	return status
+}
diff --git a/internal/config/initializer.go b/internal/config/initializer.go
index a58f82a..ec758c8 100644
--- a/internal/config/initializer.go
+++ b/internal/config/initializer.go
@@ -96,11 +96,19 @@ func (c *initializer) transformConfig(sourceProcess source.Source, args *Args, a
 		common.SSHPort = args.SSHPort
 	}
 
-	if args.Discovery == "" && args.ServersStr == "" {
+	if args.Discovery == "" && (args.ServersStr == "" ||
+		strings.ToLower(args.ServersStr) == "serverless") {
 		// We are not connecting to any servers.
 		args.Serverless = true
 	}
 
+	if sourceProcess == source.HealthCheck {
+		args.TrustAllHosts = true
+		if !args.Serverless && strings.ToLower(args.ServersStr) == "" {
+			args.ServersStr = fmt.Sprintf("localhost:%d", DefaultSSHPort)
+		}
+	}
+
 	// Interpret additional args as file list.
 	if args.What == "" {
 		var files []string
diff --git a/internal/io/dlog/dlog.go b/internal/io/dlog/dlog.go
index a17d6e9..2ae3b04 100644
--- a/internal/io/dlog/dlog.go
+++ b/internal/io/dlog/dlog.go
@@ -59,7 +59,7 @@ func Start(ctx context.Context, wg *sync.WaitGroup, sourceProcess source.Source,
 		Common = Server
 	case source.HealthCheck:
 		// Health check isn't logging anything.
-		impl := loggers.STDOUT
+		impl := loggers.NONE
 		Client = New(source.HealthCheck, source.Client, level, impl, strategy)
 		Server = New(source.HealthCheck, source.Server, level, impl, strategy)
 		Common = Client
diff --git a/internal/io/signal/signal.go b/internal/io/signal/signal.go
index 500c530..14056c4 100644
--- a/internal/io/signal/signal.go
+++ b/internal/io/signal/signal.go
@@ -44,3 +44,8 @@ func InterruptCh(ctx context.Context) <-chan string {
 
 	return statsCh
 }
+
+// NoCh doesn't listen on a signal.
+func NoCh(ctx context.Context) <-chan string {
+	return make(chan string)
+}
diff --git a/internal/server/handlers/healthhandler.go b/internal/server/handlers/healthhandler.go
index 3f3b932..347ff66 100644
--- a/internal/server/handlers/healthhandler.go
+++ b/internal/server/handlers/healthhandler.go
@@ -48,8 +48,8 @@ func (h *HealthHandler) handleHealthCommand(ctx context.Context, argc int, args
 
 	switch commandName {
 	case "health":
-		h.send(h.serverMessages, "OK: DTail SSH Server seems fine")
-	case "ack", ".ack":
+		h.send(h.serverMessages, "OK")
+	case ".ack":
 		h.handleAckCommand(argc, args)
 	default:
 		h.send(h.serverMessages, dlog.Server.Error(h.user, "Received unknown health command", commandName, argc, args))
diff --git a/internal/server/handlers/serverhandler.go b/internal/server/handlers/serverhandler.go
index aaffe14..aed8956 100644
--- a/internal/server/handlers/serverhandler.go
+++ b/internal/server/handlers/serverhandler.go
@@ -106,7 +106,7 @@ func (h *ServerHandler) handleUserCommand(ctx context.Context, argc int, args []
 			commandFinished()
 		}()
 
-	case "ack", ".ack":
+	case ".ack":
 		h.handleAckCommand(argc, args)
 		commandFinished()
 
diff --git a/samples/check_dserver.sh.sample b/samples/check_dserver.sh.sample
index 96c96de..9462037 100755
--- a/samples/check_dserver.sh.sample
+++ b/samples/check_dserver.sh.sample
@@ -1,4 +1,3 @@
 #!/bin/bash
 
-declare -r CONFIG_FILE=/etc/dserver/dtail.json
-exec /usr/local/bin/dtail --cfg $CONFIG_FILE --checkHealth
+exec /usr/local/bin/dtailhealthcheck --server localhost:2222