Add per-node health checks with status indicators and scheduled probes.

service/internal/auth/ping.go

@@ -0,0 +1,89 @@
+package auth
+
+import (
+	"fmt"
+	"net"
+	"os/exec"
+	"strings"
+	"time"
+)
+
+const (
+	PingMethodICMP = "icmp"
+	PingMethodTCP  = "tcp"
+
+	defaultPingCount   = "1"
+	defaultPingWaitSec = "2"
+	defaultTCPTimeout  = 3 * time.Second
+)
+
+// PingResult is the outcome of a reachability probe.
+type PingResult struct {
+	OK     bool
+	Method string
+	Error  error
+}
+
+type pingCommandFunc func(hostIP string) error
+type tcpDialFunc func(address string, timeout time.Duration) error
+
+var (
+	runICMPPing = defaultRunICMPPing
+	dialTCP     = defaultDialTCP
+)
+
+// PingHost probes host reachability: ICMP first, then TCP :22 when ICMP is
+// unavailable or the host does not respond to ICMP.
+func PingHost(hostIP string) PingResult {
+	hostIP = strings.TrimSpace(hostIP)
+	if hostIP == "" {
+		return PingResult{OK: false, Error: fmt.Errorf("host IP is required")}
+	}
+
+	icmpErr := runICMPPing(hostIP)
+	if icmpErr == nil {
+		return PingResult{OK: true, Method: PingMethodICMP}
+	}
+
+	address := net.JoinHostPort(hostIP, defaultSSHPort)
+	tcpErr := dialTCP(address, defaultTCPTimeout)
+	if tcpErr == nil {
+		return PingResult{OK: true, Method: PingMethodTCP}
+	}
+
+	return PingResult{
+		OK:     false,
+		Method: PingMethodTCP,
+		Error: fmt.Errorf(
+			"icmp: %v; tcp %s: %w",
+			icmpErr,
+			address,
+			tcpErr,
+		),
+	}
+}
+
+func defaultRunICMPPing(hostIP string) error {
+	pingPath, err := exec.LookPath("ping")
+	if err != nil {
+		return fmt.Errorf("ping unavailable: %w", err)
+	}
+
+	cmd := exec.Command(pingPath, "-c", defaultPingCount, "-W", defaultPingWaitSec, hostIP)
+	output, runErr := cmd.CombinedOutput()
+	if runErr == nil {
+		return nil
+	}
+
+	combined := strings.TrimSpace(string(output) + " " + runErr.Error())
+	return fmt.Errorf("%s", strings.TrimSpace(combined))
+}
+
+func defaultDialTCP(address string, timeout time.Duration) error {
+	conn, err := net.DialTimeout("tcp", address, timeout)
+	if err != nil {
+		return err
+	}
+	_ = conn.Close()
+	return nil
+}

service/internal/auth/ping_test.go

@@ -0,0 +1,96 @@
+package auth
+
+import (
+	"errors"
+	"testing"
+	"time"
+)
+
+func TestPingHostUsesICMPWhenAvailable(t *testing.T) {
+	originalPing := runICMPPing
+	originalDial := dialTCP
+	t.Cleanup(func() {
+		runICMPPing = originalPing
+		dialTCP = originalDial
+	})
+
+	runICMPPing = func(hostIP string) error {
+		if hostIP != "10.0.0.1" {
+			t.Fatalf("host = %q", hostIP)
+		}
+		return nil
+	}
+	dialTCP = func(address string, timeout time.Duration) error {
+		t.Fatal("tcp should not be used when icmp succeeds")
+		return nil
+	}
+
+	result := PingHost("10.0.0.1")
+	if !result.OK {
+		t.Fatalf("expected ok, got %#v", result)
+	}
+	if result.Method != PingMethodICMP {
+		t.Fatalf("method = %q", result.Method)
+	}
+}
+
+func TestPingHostFallsBackToTCPWhenICMPFails(t *testing.T) {
+	originalPing := runICMPPing
+	originalDial := dialTCP
+	t.Cleanup(func() {
+		runICMPPing = originalPing
+		dialTCP = originalDial
+	})
+
+	runICMPPing = func(hostIP string) error {
+		return errors.New("ping unavailable: executable file not found")
+	}
+	dialTCP = func(address string, timeout time.Duration) error {
+		if address != "10.0.0.2:22" {
+			t.Fatalf("address = %q", address)
+		}
+		return nil
+	}
+
+	result := PingHost("10.0.0.2")
+	if !result.OK {
+		t.Fatalf("expected ok, got %#v", result)
+	}
+	if result.Method != PingMethodTCP {
+		t.Fatalf("method = %q", result.Method)
+	}
+}
+
+func TestPingHostFailsWhenBothFail(t *testing.T) {
+	originalPing := runICMPPing
+	originalDial := dialTCP
+	t.Cleanup(func() {
+		runICMPPing = originalPing
+		dialTCP = originalDial
+	})
+
+	runICMPPing = func(hostIP string) error {
+		return errors.New("100% packet loss")
+	}
+	dialTCP = func(address string, timeout time.Duration) error {
+		return errors.New("connection refused")
+	}
+
+	result := PingHost("10.0.0.3")
+	if result.OK {
+		t.Fatal("expected failure")
+	}
+	if result.Error == nil {
+		t.Fatal("expected error")
+	}
+	if result.Method != PingMethodTCP {
+		t.Fatalf("method = %q", result.Method)
+	}
+}
+
+func TestPingHostRequiresHostIP(t *testing.T) {
+	result := PingHost("  ")
+	if result.OK || result.Error == nil {
+		t.Fatalf("expected failure, got %#v", result)
+	}
+}