devices/arm_cdcether.go

package devices

import (
	"context"
	"fmt"
	"log"
	"os"
	"os/exec"
	"strconv"
	"syscall"
	"time"

	"go.fuchsia.dev/infra/devices/power"
	"go.fuchsia.dev/tools/netboot"
)

// ArmCDCether represents a generic arm cdcether  product. It currently encompasses
// 1) Sherlock
// 2) Astro
// 3) Nelson
type ArmCDCether struct {
	fuchsiaDevice
}

// NewArmCDCether initializes Astros, Sherlocks, and Nelsons as Arm CDCether  objects.
func NewArmCDCether(config DeviceConfig, bootserverCmdStub []string) (*ArmCDCether, error) {
	var powerManager powerManager = nil
	if config.Power != nil {
		port, err := strconv.Atoi(config.Power.PDUPort)
		if err != nil {
			return nil, err
		}
		powerManager = power.NewPDUPowerManager(
			config.Power.PDUIp,
			config.Power.Username,
			config.Power.Password,
			port,
		)
	}

	sd := &ArmCDCether{
		fuchsiaDevice{
			power: powerManager,
		},
	}
	if err := initFuchsiaDevice(&sd.fuchsiaDevice, config, bootserverCmdStub); err != nil {
		return nil, err
	}

	// All transitions in a ArmCDCether attempt to get it into zedboot.
	transitionMap := map[DeviceState]*Transition{
		Initial: &Transition{
			PerformAction: sd.sshRebootRecovery,
			Validate:      sd.checkTransitionSuccess,
			SuccessState:  Healthy,
			FailureState:  "ssh_failed",
		},
		"ssh_failed": &Transition{
			PerformAction: sd.serialRebootRecovery,
			Validate:      sd.checkTransitionSuccess,
			SuccessState:  Healthy,
			FailureState:  "serial_failed",
		},
		"serial_failed": &Transition{
			PerformAction: sd.PowercycleIntoRecovery,
			Validate:      sd.checkTransitionSuccess,
			SuccessState:  Healthy,
			FailureState:  Unrecoverable,
		},
	}
	sd.transitionMap = transitionMap
	return sd, nil
}

// PowercycleIntoRecovery powercycles the device and then sends a serial command
// to boot into the recovery partition.
func (s *ArmCDCether) PowercycleIntoRecovery(ctx context.Context) error {
	if err := s.Powercycle(ctx); err != nil {
		return err
	}
	time.Sleep(powercycleWait)
	return s.SoftReboot(ctx, "R", "serial")
}

// checkTransitionSuccess is the validation function used by all transitions
// performed by a ArmCDCether. It verfies that the devices is in zedboot and
// broadcasting.
func (s *ArmCDCether) checkTransitionSuccess(ctx context.Context) error {
	// Give the device time to transition.
	time.Sleep(1 * time.Minute)
	client := netboot.NewClient(netbootTimeout)
	if err := checkBroadcasting(client); err != nil {
		return err
	}
	return deviceInZedboot(client, s.nodename)
}

// checkBroadcasting ensures that the ArmCDCether is sending out advertising
// packets.
func checkBroadcasting(n *netboot.Client) error {
	if _, err := n.Beacon(); err != nil {
		return err
	}
	return nil
}

// ToTaskState puts the ArmCDCether into a state in which it can run tasks.
func (s *ArmCDCether) ToTaskState(ctx context.Context) error {
	cmd := exec.Command(s.BootserverCmd[0], s.BootserverCmd[1:]...)

	// Spin up handler to exit subprocess cleanly on sigterm.
	processDone := make(chan bool, 1)
	go func() {
		select {
		case <-processDone:
		case <-ctx.Done():
			if err := cmd.Process.Signal(syscall.SIGTERM); err != nil {
				log.Printf("exited cmd with error %v", err)
			}
		}
	}()

	// Ensure the context still exists before running the subprocess.
	if ctx.Err() != nil {
		return nil
	}

	// Run the bootserver.
	cmd.Stdout = os.Stdout
	cmd.Stderr = os.Stderr
	cmd.Run()
	processDone <- true
	if !cmd.ProcessState.Success() {
		return fmt.Errorf("bootserver for %s exited with exit code %d", s.nodename, cmd.ProcessState.ExitCode())
	}
	return nil
}