UAP Ops Agent Golang Wrapper.

cmd/ops_agent_uap_wrapper/main.go

@@ -0,0 +1,126 @@
+package main
+
+import (
+
+	"log"
+	"net"
+	"os"
+	"os/exec"
+	"os/signal"
+	"syscall"
+	"time"
+)
+const MaximumWaitForProcessStart = 5 * time.Second
+
+func main() {
+	// Define the paths and commands from the systemd unit file.
+	prefix := "/opt/google-cloud-ops-agent"  // @PREFIX@
+	sysconfdir := "/etc" // @SYSCONFDIR@
+	logsDirectory := "google-cloud-ops-agent"
+	fluentBitStateDiectory := "google-cloud-ops-agent/fluent-bit"
+	//fluentBitRuntimeDirectory := "google-cloud-ops-agent-fluent-bit"
+
+
+	// Wait for network to be online before starting the process
+	waitForNetwork()
+	// Start google-cloud-ops-agent.service
+
+	// Starting ExecStartPre commands 
+	execStartPreCmd := exec.Command(
+		prefix+"/libexec/google_cloud_ops_agent_engine",
+		"-in", sysconfdir+"/google-cloud-ops-agent/config.yaml",
+	)
+	if err := runCommand(execStartPreCmd); err != nil {
+		log.Fatalf("config validation faile: ", err)
+	}
+
+	execStartPreFluentBitCmd := exec.Command(
+		prefix+"/libexec/google_cloud_ops_agent_engine",
+		"-service", "fluentbit",
+		"-in", sysconfdir+"/google-cloud-ops-agent/config.yaml",
+		"-logs", logsDirectory,  "-state", fluentBitStateDiectory)
+
+	if err := runCommand(execStartPreFluentBitCmd); err != nil {
+		log.Fatalf("exec pre command for FluentBit faile: ", err)
+	}
+
+	execStartPreOtelCmd := exec.Command(
+		prefix+"/libexec/google_cloud_ops_agent_engine",
+		"-service", "otel",
+		"-in", sysconfdir+"/google-cloud-ops-agent/config.yaml",
+		"-logs", logsDirectory)
+
+	if err := runCommand(execStartPreOtelCmd); err != nil {
+		log.Fatalf("exec pre command for Otel faile: ", err)
+	}
+
+
+	// Starting Diagnostics Service 
+	execDiagnosticsCmd := exec.Command(
+		prefix+"/libexec/google_cloud_ops_agent_diagnostics",
+		"-config", sysconfdir+"/google-cloud-ops-agent/config.yaml",
+	)
+
+	restartCommand(execDiagnosticsCmd)
+
+}
+
+
+
+func handleSignals(cmd *exec.Cmd) {
+	// Relay signals that should be passed down to the subprocess we are wrapping.
+	sigs := make(chan os.Signal, 1)
+	signal.Notify(sigs, syscall.SIGTERM, syscall.SIGINT, syscall.SIGCONT)
+	for {
+		sig := <-sigs
+		start := time.Now()
+		// It is possible that we receive a signal before the code for `cmd.Run()` set cmd.Process.
+		// In this case we wait up to MaximumWaitForProcessStart before giving up relaying the signal.
+		for {
+			if cmd.Process != nil {
+				cmd.Process.Signal(sig)
+				break
+			} else if time.Until(start.Add(MaximumWaitForProcessStart)) <= 0 {
+				// We waited MaximumWaitForProcessStart and the subprocess did not start. Give up on relaying the signal.
+				log.Printf("Failed to relay signal %v to subprocess as it has not started yet", sig)
+				break
+			}
+			time.Sleep(50 * time.Millisecond)
+		}
+	}
+}
+
+func runCommand(cmd *exec.Cmd) error {
+	cmd.SysProcAttr = &syscall.SysProcAttr{
+		Pdeathsig: syscall.SIGKILL,
+		Setpgid: true,
+	}
+	go handleSignals(cmd)
+	return cmd.Run()
+}
+
+func restartCommand(cmd *exec.Cmd) {
+	cmdToRestart := exec.Command(cmd.Path, cmd.Args...)
+	if err := runCommand(cmdToRestart); err != nil {
+		log.Fatalf("command failed: ", err)
+	} else {
+		log.Print("command succeeded")
+	}
+	log.Print("Restarting command")
+
+	restartCommand(cmdToRestart)
+}
+
+func waitForNetwork() {
+	for {
+		// Try to connect to a known reliable host (e.g., Google DNS)
+		conn, err := net.DialTimeout("tcp", "8.8.8.8:53", 5*time.Second)
+		if err == nil {
+			conn.Close()
+			log.Print("Network online. Proceeding...")
+			break
+		}
+		log.Print("Waiting for network...")
+		time.Sleep(5 * time.Second)
+	}
+}