// NewExecutor returns an Executor
func NewExecutor(logger *log.Logger) Executor {
exec := &UniversalExecutor{
logger: logger,
processExited: make(chan interface{}),
totalCpuStats: stats.NewCpuStats(),
userCpuStats: stats.NewCpuStats(),
systemCpuStats: stats.NewCpuStats(),
pids: make(map[int]*nomadPid),
}
return exec
}
// NewExecutor returns an Executor
func NewExecutor(logger *log.Logger) Executor {
if err := shelpers.Init(); err != nil {
logger.Printf("[FATAL] executor: unable to initialize stats: %v", err)
return nil
}
exec := &UniversalExecutor{
logger: logger,
processExited: make(chan interface{}),
totalCpuStats: stats.NewCpuStats(),
userCpuStats: stats.NewCpuStats(),
systemCpuStats: stats.NewCpuStats(),
pids: make(map[int]*nomadPid),
}
return exec
}
// NewExecutor returns an Executor
func NewExecutor(logger *log.Logger) Executor {
exec := &UniversalExecutor{
logger: logger,
processExited: make(chan interface{}),
cpuStats: stats.NewCpuStats(),
}
return exec
}
// scanPids scans all the pids on the machine running the current executor and
// returns the child processes of the executor.
func (e *UniversalExecutor) scanPids(parentPid int, allPids []ps.Process) (map[int]*nomadPid, error) {
processFamily := make(map[int]struct{})
processFamily[parentPid] = struct{}{}
// A mapping of pids to their parent pids. It is used to build the process
// tree of the executing task
pidsRemaining := make(map[int]int, len(allPids))
for _, pid := range allPids {
pidsRemaining[pid.Pid()] = pid.PPid()
}
for {
// flag to indicate if we have found a match
foundNewPid := false
for pid, ppid := range pidsRemaining {
_, childPid := processFamily[ppid]
// checking if the pid is a child of any of the parents
if childPid {
processFamily[pid] = struct{}{}
delete(pidsRemaining, pid)
foundNewPid = true
}
}
// not scanning anymore if we couldn't find a single match
if !foundNewPid {
break
}
}
res := make(map[int]*nomadPid)
for pid := range processFamily {
np := nomadPid{
pid: pid,
cpuStatsTotal: stats.NewCpuStats(),
cpuStatsUser: stats.NewCpuStats(),
cpuStatsSys: stats.NewCpuStats(),
}
res[pid] = &np
}
return res, nil
}
// scanPids scans all the pids on the machine running the current executor and
// returns the child processes of the executor.
func (e *UniversalExecutor) scanPids(parentPid int, allPids []ps.Process) (map[int]*nomadPid, error) {
processFamily := make(map[int]struct{})
processFamily[parentPid] = struct{}{}
// A buffer for holding pids which haven't matched with any parent pid
var pidsRemaining []ps.Process
for {
// flag to indicate if we have found a match
foundNewPid := false
for _, pid := range allPids {
_, childPid := processFamily[pid.PPid()]
// checking if the pid is a child of any of the parents
if childPid {
processFamily[pid.Pid()] = struct{}{}
foundNewPid = true
} else {
// if it is not, then we add the pid to the buffer
pidsRemaining = append(pidsRemaining, pid)
}
// scan only the pids which are left in the buffer
allPids = pidsRemaining
}
// not scanning anymore if we couldn't find a single match
if !foundNewPid {
break
}
}
res := make(map[int]*nomadPid)
for pid := range processFamily {
np := nomadPid{
pid: pid,
cpuStatsTotal: stats.NewCpuStats(),
cpuStatsUser: stats.NewCpuStats(),
cpuStatsSys: stats.NewCpuStats(),
}
res[pid] = &np
}
return res, nil
}
// getAllPids returns the pids of all the processes spun up by the executor. We
// use the libcontainer apis to get the pids when the user is using cgroup
// isolation and we scan the entire process table if the user is not using any
// isolation
func (e *UniversalExecutor) getAllPids() (map[int]*nomadPid, error) {
if e.command.ResourceLimits {
manager := getCgroupManager(e.resConCtx.groups, e.resConCtx.cgPaths)
pids, err := manager.GetAllPids()
if err != nil {
return nil, err
}
np := make(map[int]*nomadPid, len(pids))
for _, pid := range pids {
np[pid] = &nomadPid{
pid: pid,
cpuStatsTotal: stats.NewCpuStats(),
cpuStatsSys: stats.NewCpuStats(),
cpuStatsUser: stats.NewCpuStats(),
}
}
return np, nil
}
allProcesses, err := ps.Processes()
if err != nil {
return nil, err
}
return e.scanPids(os.Getpid(), allProcesses)
}
// Open creates the driver to monitor an existing LXC container
func (d *LxcDriver) Open(ctx *ExecContext, handleID string) (DriverHandle, error) {
pid := &lxcPID{}
if err := json.Unmarshal([]byte(handleID), pid); err != nil {
return nil, fmt.Errorf("Failed to parse handle '%s': %v", handleID, err)
}
var container *lxc.Container
containers := lxc.Containers(pid.LxcPath)
for _, c := range containers {
if c.Name() == pid.ContainerName {
container = &c
break
}
}
if container == nil {
return nil, fmt.Errorf("container %v not found", pid.ContainerName)
}
handle := lxcDriverHandle{
container: container,
initPid: container.InitPid(),
lxcPath: pid.LxcPath,
logger: d.logger,
killTimeout: pid.KillTimeout,
maxKillTimeout: d.DriverContext.config.MaxKillTimeout,
totalCpuStats: stats.NewCpuStats(),
userCpuStats: stats.NewCpuStats(),
systemCpuStats: stats.NewCpuStats(),
waitCh: make(chan *dstructs.WaitResult, 1),
doneCh: make(chan bool, 1),
}
go handle.run()
return &handle, nil
}
// getAllPids returns the pids of all the processes spun up by the executor. We
// use the libcontainer apis to get the pids when the user is using cgroup
// isolation and we scan the entire process table if the user is not using any
// isolation
func (e *UniversalExecutor) getAllPids() ([]*nomadPid, error) {
if e.command.ResourceLimits {
manager := getCgroupManager(e.groups, e.cgPaths)
pids, err := manager.GetAllPids()
if err != nil {
return nil, err
}
np := make([]*nomadPid, len(pids))
for idx, pid := range pids {
np[idx] = &nomadPid{pid, stats.NewCpuStats()}
}
return np, nil
}
allProcesses, err := ps.Processes()
if err != nil {
return nil, err
}
return e.scanPids(os.Getpid(), allProcesses)
}
// Start starts the LXC Driver
func (d *LxcDriver) Start(ctx *ExecContext, task *structs.Task) (DriverHandle, error) {
var driverConfig LxcDriverConfig
if err := mapstructure.WeakDecode(task.Config, &driverConfig); err != nil {
return nil, err
}
lxcPath := lxc.DefaultConfigPath()
if path := d.config.Read("driver.lxc.path"); path != "" {
lxcPath = path
}
containerName := fmt.Sprintf("%s-%s", task.Name, ctx.AllocID)
c, err := lxc.NewContainer(containerName, lxcPath)
if err != nil {
return nil, fmt.Errorf("unable to initialize container: %v", err)
}
var verbosity lxc.Verbosity
switch driverConfig.Verbosity {
case "verbose":
verbosity = lxc.Verbose
case "", "quiet":
verbosity = lxc.Quiet
default:
return nil, fmt.Errorf("lxc driver config 'verbosity' can only be either quiet or verbose")
}
c.SetVerbosity(verbosity)
var logLevel lxc.LogLevel
switch driverConfig.LogLevel {
case "trace":
logLevel = lxc.TRACE
case "debug":
logLevel = lxc.DEBUG
case "info":
logLevel = lxc.INFO
case "warn":
logLevel = lxc.WARN
case "", "error":
logLevel = lxc.ERROR
default:
return nil, fmt.Errorf("lxc driver config 'log_level' can only be trace, debug, info, warn or error")
}
c.SetLogLevel(logLevel)
logFile := filepath.Join(ctx.AllocDir.LogDir(), fmt.Sprintf("%v-lxc.log", task.Name))
c.SetLogFile(logFile)
options := lxc.TemplateOptions{
Template: driverConfig.Template,
Distro: driverConfig.Distro,
Release: driverConfig.Release,
Arch: driverConfig.Arch,
FlushCache: driverConfig.FlushCache,
DisableGPGValidation: driverConfig.DisableGPGValidation,
ExtraArgs: driverConfig.TemplateArgs,
}
if err := c.Create(options); err != nil {
return nil, fmt.Errorf("unable to create container: %v", err)
}
// Set the network type to none
if err := c.SetConfigItem("lxc.network.type", "none"); err != nil {
return nil, fmt.Errorf("error setting network type configuration: %v", err)
}
// Bind mount the shared alloc dir and task local dir in the container
taskDir, ok := ctx.AllocDir.TaskDirs[task.Name]
if !ok {
return nil, fmt.Errorf("failed to find task local directory: %v", task.Name)
}
secretdir, err := ctx.AllocDir.GetSecretDir(task.Name)
if err != nil {
return nil, fmt.Errorf("faild getting secret path for task: %v", err)
}
taskLocalDir := filepath.Join(taskDir, allocdir.TaskLocal)
mounts := []string{
fmt.Sprintf("%s local none rw,bind,create=dir", taskLocalDir),
fmt.Sprintf("%s alloc none rw,bind,create=dir", ctx.AllocDir.SharedDir),
fmt.Sprintf("%s secret none rw,bind,create=dir", secretdir),
}
for _, mnt := range mounts {
if err := c.SetConfigItem("lxc.mount.entry", mnt); err != nil {
return nil, fmt.Errorf("error setting bind mount %q error: %v", mnt, err)
}
}
// Start the container
if err := c.Start(); err != nil {
return nil, fmt.Errorf("unable to start container: %v", err)
}
// Set the resource limits
if err := c.SetMemoryLimit(lxc.ByteSize(task.Resources.MemoryMB) * lxc.MB); err != nil {
return nil, fmt.Errorf("unable to set memory limits: %v", err)
}
if err := c.SetCgroupItem("cpu.shares", strconv.Itoa(task.Resources.CPU)); err != nil {
return nil, fmt.Errorf("unable to set cpu shares: %v", err)
}
handle := lxcDriverHandle{
container: c,
initPid: c.InitPid(),
lxcPath: lxcPath,
logger: d.logger,
killTimeout: GetKillTimeout(task.KillTimeout, d.DriverContext.config.MaxKillTimeout),
maxKillTimeout: d.DriverContext.config.MaxKillTimeout,
totalCpuStats: stats.NewCpuStats(),
userCpuStats: stats.NewCpuStats(),
systemCpuStats: stats.NewCpuStats(),
waitCh: make(chan *dstructs.WaitResult, 1),
doneCh: make(chan bool, 1),
}
go handle.run()
return &handle, nil
}