devices/vim.go

package devices

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

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

// Vim represents a VIM2 (and eventually a VIM3)
type Vim struct {
	fuchsiaDevice
}

// NewVim initializes a Vim for use by other infra utilities.
func NewVim(config DeviceConfig, bootserverCmdStub []string) (*Vim, error) {
	v := &Vim{
		fuchsiaDevice{
			power: power.NewWOLPowerManager(config.Network.Mac),
		},
	}
	if err := initFuchsiaDevice(&v.fuchsiaDevice, config, bootserverCmdStub); err != nil {
		return nil, err
	}

	// All transitions in Vim attempt to get the device into zedboot.
	transitionMap := map[DeviceState]*Transition{
		Initial: &Transition{
			PerformAction: v.sshRebootRecovery,
			Validate:      v.checkTransitionSuccess,
			SuccessState:  Healthy,
			FailureState:  "ssh_failed",
		},
		"ssh_failed": &Transition{
			PerformAction: v.serialRebootRecovery,
			Validate:      v.checkTransitionSuccess,
			SuccessState:  Healthy,
			FailureState:  "serial_failed",
		},
		"serial_failed": &Transition{
			PerformAction: v.Powercycle,
			Validate:      v.checkTransitionSuccess,
			SuccessState:  Healthy,
			FailureState:  Unrecoverable,
		},
	}
	v.transitionMap = transitionMap
	return v, nil
}

// checkTransitionSuccess is the validation function used by all transitions
// performed by Vim. It checks that the device is in zedboot.
func (v *Vim) checkTransitionSuccess(ctx context.Context) error {
	// Give the device time to transition.
	time.Sleep(1 * time.Minute)
	client := netboot.NewClient(netbootTimeout)
	return deviceInZedboot(client, v.nodename)
}

// ToTaskState puts the Vim into a state in which it can run a test task.
func (v *Vim) ToTaskState(ctx context.Context) error {
	cmd := exec.Command(v.BootserverCmd[0], v.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", v.nodename, cmd.ProcessState.ExitCode())
	}
	return nil
}