blob: 5333dcbbc67c7f0453c0503f30828c0b0670bbec [file] [log] [blame]
// Copyright 2020 The Chromium OS Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
package main
import (
pb "chromiumos/vm_tools/vm_crash"
lxd ""
const (
crashHostPort = "7779"
var lxdCmd = []string{"/usr/sbin/lxd", "--group", "lxd", "--syslog"}
type lxdHelper struct {
lxd lxd.ContainerServer
process *os.Process
mutex sync.Mutex
shutdownSignal chan os.Signal
// runWithRetry runs cmd, and whenever cmd exits runs it again unless
// shutdownSignal has been closed or the command has run too many times recently.
func runWithRetry(cmd func() error, shutdownSignal chan os.Signal, failureWindow time.Duration, numFailures int) {
var failures []time.Time
for {
waitReadySucceeded := make(chan bool)
go func() {
_, err := lxdWaitready()
if err == nil {
waitReadySucceeded <- true
err := cmd()
// If we reach here LXD has stopped running or never started.
// Process shutdownSignal first so we never ignore a
// shutdown signal in favour of waitReadySucceeded.
select {
case <-shutdownSignal:
log.Print("Shutting down LXD")
// We're shutting down, nothing to do here.
// Check if the waitready operation succeeded. If so,
// that means LXD started up and then failed, so
// indicate that to cicerone.
select {
case <-waitReadySucceeded:
log.Print("LXD died unexpectedly with error: ", err)
log.Print("LXD failed to start with error: ", err)
// At this point we know LXD failed and needs to be
// restarted, so do that.
for len(failures) > 0 && failures[0].Before(time.Now().Add(-failureWindow)) {
failures = failures[1:]
if len(failures) >= numFailures {
// We can't do anything with no LXD, so give up. Keep Tremplin
// running so we can report failures when the user tries to do
// anything, which is the best we can offer at the moment.
log.Print("LXD keeps dying, giving up. All future calls to Tremplin will fail until VM restart.")
// We're now giving up on starting LXD, so signal the LXD failure to cicerone.
failures = append(failures, time.Now())
// signalLxdFailure tells cicerone that LXD failed. This is meant to
// be used only to indicate either failures after startup (i.e. after
// lxd waitready succeeds) or if we fail to find any way to start up
// LXD.
func signalLxdFailure() {
conn, err := grpc.Dial(crashHostPort,
if err != nil {
log.Print("Could not connect to crash listener: ", err)
defer conn.Close()
crashListener := pb.NewCrashListenerClient(conn)
var failureReport pb.FailureReport
failureReport.FailedProcess = "lxd"
_, err = crashListener.SendFailureReport(context.Background(), &failureReport)
if err != nil {
log.Print("Failed to report lxd failure: ", err)
// runLxd launches LXD and blocks until LXD exits. Will set l.process.
func (l *lxdHelper) runLxd() error {
log.Printf("Running %q", strings.Join(lxdCmd, " "))
cmd := exec.Command(lxdCmd[0], lxdCmd[1:]...)
err := cmd.Start()
if err == nil {
log.Print("LXD started")
l.process = cmd.Process
return cmd.Wait()
return err
// lxdWaitready waits for LXD to be ready to handle requests.
func lxdWaitready() (lxd.ContainerServer, error) {
var lastErr error
const lxdReadyTimeout = 2 * time.Minute
const lxdInterval = 500 * time.Millisecond
start := time.Now()
attempt := 0
for time.Now().Before(start.Add(lxdReadyTimeout)) {
// Wait until there's a socket with something listening at the other end.
c, err := lxd.ConnectLXDUnix("", nil)
if err != nil {
if attempt%10 == 0 {
log.Print("ConnectLXDUnix not ready yet, retrying. Error was: ", err)
lastErr = err
// Wait until the server says it's ready.
_, _, err = c.RawQuery("GET", "/internal/ready", nil, "")
if err != nil {
if attempt%10 == 0 {
log.Print("internal/ready not ready yet, retrying. Error was: ", err)
lastErr = err
// LXD is ready.
log.Print("LXD launched")
return c, nil
return nil, lastErr
// LaunchLxd with launch LXD and return a connection to the server, or an error
// if launching failed. Will respawn LXD if it ever exits.
func (l *lxdHelper) LaunchLxd() (lxd.ContainerServer, error) {
defer l.mutex.Unlock()
if l.shutdownSignal == nil {
l.shutdownSignal = make(chan os.Signal, 1)
// Launch LXD, we allow 10 failures per 30 seconds (copied from maitre'd)
// before giving up.
go runWithRetry(l.runLxd, l.shutdownSignal, 30*time.Second, 10)
if l.lxd == nil {
// Wait for LXD to be ready.
c, err := lxdWaitready()
if err != nil {
// Timed out
return nil, fmt.Errorf("Timed out waiting for LXD to start. Last error was: %w", err)
log.Print("LXD running and ready")
l.lxd = c
return l.lxd, nil
// StopLxd signals LXD to shut down with the provided signal and stops
// respawning it automatically. Returns as soon as the signal is sent instead of
// blocking on LXD actually shutting down.
func (l *lxdHelper) StopLxd() error {
signal := syscall.SIGPWR
defer l.mutex.Unlock()
log.Print("Telling LXD to shut down")
if l.shutdownSignal != nil {
// This can happen if we shut down before we started LXD.
log.Print("No channel to signal LXD on (is it running?). Nothing to do.")
return nil
l.shutdownSignal <- signal
return l.process.Signal(signal)