moby/libcontainerd/client_windows.go

461 lines
14 KiB
Go
Raw Normal View History

package libcontainerd
import (
"errors"
"fmt"
"io"
"path/filepath"
"strings"
"syscall"
"golang.org/x/net/context"
"github.com/Microsoft/hcsshim"
"github.com/Sirupsen/logrus"
)
type client struct {
clientCommon
// Platform specific properties below here (none presently on Windows)
}
// Win32 error codes that are used for various workarounds
// These really should be ALL_CAPS to match golangs syscall library and standard
// Win32 error conventions, but golint insists on CamelCase.
const (
CoEClassstring = syscall.Errno(0x800401F3) // Invalid class string
ErrorNoNetwork = syscall.Errno(1222) // The network is not present or not started
ErrorBadPathname = syscall.Errno(161) // The specified path is invalid
ErrorInvalidObject = syscall.Errno(0x800710D8) // The object identifier does not represent a valid object
)
// defaultOwner is a tag passed to HCS to allow it to differentiate between
// container creator management stacks. We hard code "docker" in the case
// of docker.
const defaultOwner = "docker"
// Create is the entrypoint to create a container from a spec, and if successfully
// created, start it too.
func (clnt *client) Create(containerID string, checkpoint string, checkpointDir string, spec Spec, options ...CreateOption) error {
logrus.Debugln("libcontainerd: client.Create() with spec", spec)
configuration := &hcsshim.ContainerConfig{
SystemType: "Container",
Name: containerID,
Owner: defaultOwner,
VolumePath: spec.Root.Path,
IgnoreFlushesDuringBoot: false,
LayerFolderPath: spec.Windows.LayerFolder,
HostName: spec.Hostname,
}
if spec.Windows.Networking != nil {
configuration.EndpointList = spec.Windows.Networking.EndpointList
}
if spec.Windows.Resources != nil {
if spec.Windows.Resources.CPU != nil {
if spec.Windows.Resources.CPU.Shares != nil {
configuration.ProcessorWeight = *spec.Windows.Resources.CPU.Shares
}
if spec.Windows.Resources.CPU.Percent != nil {
configuration.ProcessorMaximum = *spec.Windows.Resources.CPU.Percent * 100 // ProcessorMaximum is a value between 1 and 10000
}
}
if spec.Windows.Resources.Memory != nil {
if spec.Windows.Resources.Memory.Limit != nil {
configuration.MemoryMaximumInMB = *spec.Windows.Resources.Memory.Limit / 1024 / 1024
}
}
if spec.Windows.Resources.Storage != nil {
if spec.Windows.Resources.Storage.Bps != nil {
configuration.StorageBandwidthMaximum = *spec.Windows.Resources.Storage.Bps
}
if spec.Windows.Resources.Storage.Iops != nil {
configuration.StorageIOPSMaximum = *spec.Windows.Resources.Storage.Iops
}
}
}
if spec.Windows.HvRuntime != nil {
configuration.VolumePath = "" // Always empty for Hyper-V containers
configuration.HvPartition = true
configuration.HvRuntime = &hcsshim.HvRuntime{
ImagePath: spec.Windows.HvRuntime.ImagePath,
}
}
if configuration.HvPartition {
configuration.SandboxPath = filepath.Dir(spec.Windows.LayerFolder)
} else {
configuration.VolumePath = spec.Root.Path
configuration.LayerFolderPath = spec.Windows.LayerFolder
}
for _, option := range options {
if s, ok := option.(*ServicingOption); ok {
configuration.Servicing = s.IsServicing
break
}
if s, ok := option.(*FlushOption); ok {
configuration.IgnoreFlushesDuringBoot = s.IgnoreFlushesDuringBoot
break
}
}
for _, layerPath := range spec.Windows.LayerPaths {
_, filename := filepath.Split(layerPath)
g, err := hcsshim.NameToGuid(filename)
if err != nil {
return err
}
configuration.Layers = append(configuration.Layers, hcsshim.Layer{
ID: g.ToString(),
Path: layerPath,
})
}
// Add the mounts (volumes, bind mounts etc) to the structure
mds := make([]hcsshim.MappedDir, len(spec.Mounts))
for i, mount := range spec.Mounts {
mds[i] = hcsshim.MappedDir{
HostPath: mount.Source,
ContainerPath: mount.Destination,
ReadOnly: false,
}
for _, o := range mount.Options {
if strings.ToLower(o) == "ro" {
mds[i].ReadOnly = true
}
}
}
configuration.MappedDirectories = mds
hcsContainer, err := hcsshim.CreateContainer(containerID, configuration)
if err != nil {
return err
}
// Construct a container object for calling start on it.
container := &container{
containerCommon: containerCommon{
process: process{
processCommon: processCommon{
containerID: containerID,
client: clnt,
friendlyName: InitFriendlyName,
},
commandLine: strings.Join(spec.Process.Args, " "),
},
processes: make(map[string]*process),
},
ociSpec: spec,
hcsContainer: hcsContainer,
}
container.options = options
for _, option := range options {
if err := option.Apply(container); err != nil {
logrus.Errorf("libcontainerd: %v", err)
}
}
// Call start, and if it fails, delete the container from our
// internal structure, start will keep HCS in sync by deleting the
// container there.
logrus.Debugf("libcontainerd: Create() id=%s, Calling start()", containerID)
if err := container.start(); err != nil {
clnt.deleteContainer(containerID)
return err
}
logrus.Debugf("libcontainerd: Create() id=%s completed successfully", containerID)
return nil
}
// AddProcess is the handler for adding a process to an already running
// container. It's called through docker exec.
func (clnt *client) AddProcess(ctx context.Context, containerID, processFriendlyName string, procToAdd Process) error {
clnt.lock(containerID)
defer clnt.unlock(containerID)
container, err := clnt.getContainer(containerID)
if err != nil {
return err
}
// Note we always tell HCS to
// create stdout as it's required regardless of '-i' or '-t' options, so that
// docker can always grab the output through logs. We also tell HCS to always
// create stdin, even if it's not used - it will be closed shortly. Stderr
// is only created if it we're not -t.
createProcessParms := hcsshim.ProcessConfig{
EmulateConsole: procToAdd.Terminal,
ConsoleSize: procToAdd.InitialConsoleSize,
CreateStdInPipe: true,
CreateStdOutPipe: true,
CreateStdErrPipe: !procToAdd.Terminal,
}
// Take working directory from the process to add if it is defined,
// otherwise take from the first process.
if procToAdd.Cwd != "" {
createProcessParms.WorkingDirectory = procToAdd.Cwd
} else {
createProcessParms.WorkingDirectory = container.ociSpec.Process.Cwd
}
// Configure the environment for the process
createProcessParms.Environment = setupEnvironmentVariables(procToAdd.Env)
createProcessParms.CommandLine = strings.Join(procToAdd.Args, " ")
logrus.Debugf("libcontainerd: commandLine: %s", createProcessParms.CommandLine)
// Start the command running in the container.
var stdout, stderr io.ReadCloser
var stdin io.WriteCloser
newProcess, err := container.hcsContainer.CreateProcess(&createProcessParms)
if err != nil {
logrus.Errorf("libcontainerd: AddProcess(%s) CreateProcess() failed %s", containerID, err)
return err
}
stdin, stdout, stderr, err = newProcess.Stdio()
if err != nil {
logrus.Errorf("libcontainerd: %s getting std pipes failed %s", containerID, err)
return err
}
iopipe := &IOPipe{Terminal: procToAdd.Terminal}
iopipe.Stdin = createStdInCloser(stdin, newProcess)
// Convert io.ReadClosers to io.Readers
if stdout != nil {
iopipe.Stdout = openReaderFromPipe(stdout)
}
if stderr != nil {
iopipe.Stderr = openReaderFromPipe(stderr)
}
pid := newProcess.Pid()
proc := &process{
processCommon: processCommon{
containerID: containerID,
friendlyName: processFriendlyName,
client: clnt,
systemPid: uint32(pid),
},
commandLine: createProcessParms.CommandLine,
hcsProcess: newProcess,
}
// Add the process to the container's list of processes
container.processes[processFriendlyName] = proc
// Make sure the lock is not held while calling back into the daemon
clnt.unlock(containerID)
// Tell the engine to attach streams back to the client
if err := clnt.backend.AttachStreams(processFriendlyName, *iopipe); err != nil {
return err
}
// Lock again so that the defer unlock doesn't fail. (I really don't like this code)
clnt.lock(containerID)
// Spin up a go routine waiting for exit to handle cleanup
go container.waitExit(proc, false)
return nil
}
// Signal handles `docker stop` on Windows. While Linux has support for
// the full range of signals, signals aren't really implemented on Windows.
// We fake supporting regular stop and -9 to force kill.
func (clnt *client) Signal(containerID string, sig int) error {
var (
cont *container
err error
)
// Get the container as we need it to find the pid of the process.
clnt.lock(containerID)
defer clnt.unlock(containerID)
if cont, err = clnt.getContainer(containerID); err != nil {
return err
}
cont.manualStopRequested = true
logrus.Debugf("libcontainerd: Signal() containerID=%s sig=%d pid=%d", containerID, sig, cont.systemPid)
if syscall.Signal(sig) == syscall.SIGKILL {
// Terminate the compute system
if err := cont.hcsContainer.Terminate(); err != nil {
if !hcsshim.IsPending(err) {
logrus.Errorf("libcontainerd: failed to terminate %s - %q", containerID, err)
}
}
} else {
// Terminate Process
if err := cont.hcsProcess.Kill(); err != nil && !hcsshim.IsAlreadyStopped(err) {
// ignore errors
logrus.Warnf("libcontainerd: failed to terminate pid %d in %s: %q", cont.systemPid, containerID, err)
}
}
return nil
}
Add support for user-defined healthchecks This PR adds support for user-defined health-check probes for Docker containers. It adds a `HEALTHCHECK` instruction to the Dockerfile syntax plus some corresponding "docker run" options. It can be used with a restart policy to automatically restart a container if the check fails. The `HEALTHCHECK` instruction has two forms: * `HEALTHCHECK [OPTIONS] CMD command` (check container health by running a command inside the container) * `HEALTHCHECK NONE` (disable any healthcheck inherited from the base image) The `HEALTHCHECK` instruction tells Docker how to test a container to check that it is still working. This can detect cases such as a web server that is stuck in an infinite loop and unable to handle new connections, even though the server process is still running. When a container has a healthcheck specified, it has a _health status_ in addition to its normal status. This status is initially `starting`. Whenever a health check passes, it becomes `healthy` (whatever state it was previously in). After a certain number of consecutive failures, it becomes `unhealthy`. The options that can appear before `CMD` are: * `--interval=DURATION` (default: `30s`) * `--timeout=DURATION` (default: `30s`) * `--retries=N` (default: `1`) The health check will first run **interval** seconds after the container is started, and then again **interval** seconds after each previous check completes. If a single run of the check takes longer than **timeout** seconds then the check is considered to have failed. It takes **retries** consecutive failures of the health check for the container to be considered `unhealthy`. There can only be one `HEALTHCHECK` instruction in a Dockerfile. If you list more than one then only the last `HEALTHCHECK` will take effect. The command after the `CMD` keyword can be either a shell command (e.g. `HEALTHCHECK CMD /bin/check-running`) or an _exec_ array (as with other Dockerfile commands; see e.g. `ENTRYPOINT` for details). The command's exit status indicates the health status of the container. The possible values are: - 0: success - the container is healthy and ready for use - 1: unhealthy - the container is not working correctly - 2: starting - the container is not ready for use yet, but is working correctly If the probe returns 2 ("starting") when the container has already moved out of the "starting" state then it is treated as "unhealthy" instead. For example, to check every five minutes or so that a web-server is able to serve the site's main page within three seconds: HEALTHCHECK --interval=5m --timeout=3s \ CMD curl -f http://localhost/ || exit 1 To help debug failing probes, any output text (UTF-8 encoded) that the command writes on stdout or stderr will be stored in the health status and can be queried with `docker inspect`. Such output should be kept short (only the first 4096 bytes are stored currently). When the health status of a container changes, a `health_status` event is generated with the new status. The health status is also displayed in the `docker ps` output. Signed-off-by: Thomas Leonard <thomas.leonard@docker.com> Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2016-04-18 09:48:13 +00:00
// While Linux has support for the full range of signals, signals aren't really implemented on Windows.
// We try to terminate the specified process whatever signal is requested.
func (clnt *client) SignalProcess(containerID string, processFriendlyName string, sig int) error {
clnt.lock(containerID)
defer clnt.unlock(containerID)
cont, err := clnt.getContainer(containerID)
if err != nil {
return err
}
for _, p := range cont.processes {
if p.friendlyName == processFriendlyName {
return p.hcsProcess.Kill()
Add support for user-defined healthchecks This PR adds support for user-defined health-check probes for Docker containers. It adds a `HEALTHCHECK` instruction to the Dockerfile syntax plus some corresponding "docker run" options. It can be used with a restart policy to automatically restart a container if the check fails. The `HEALTHCHECK` instruction has two forms: * `HEALTHCHECK [OPTIONS] CMD command` (check container health by running a command inside the container) * `HEALTHCHECK NONE` (disable any healthcheck inherited from the base image) The `HEALTHCHECK` instruction tells Docker how to test a container to check that it is still working. This can detect cases such as a web server that is stuck in an infinite loop and unable to handle new connections, even though the server process is still running. When a container has a healthcheck specified, it has a _health status_ in addition to its normal status. This status is initially `starting`. Whenever a health check passes, it becomes `healthy` (whatever state it was previously in). After a certain number of consecutive failures, it becomes `unhealthy`. The options that can appear before `CMD` are: * `--interval=DURATION` (default: `30s`) * `--timeout=DURATION` (default: `30s`) * `--retries=N` (default: `1`) The health check will first run **interval** seconds after the container is started, and then again **interval** seconds after each previous check completes. If a single run of the check takes longer than **timeout** seconds then the check is considered to have failed. It takes **retries** consecutive failures of the health check for the container to be considered `unhealthy`. There can only be one `HEALTHCHECK` instruction in a Dockerfile. If you list more than one then only the last `HEALTHCHECK` will take effect. The command after the `CMD` keyword can be either a shell command (e.g. `HEALTHCHECK CMD /bin/check-running`) or an _exec_ array (as with other Dockerfile commands; see e.g. `ENTRYPOINT` for details). The command's exit status indicates the health status of the container. The possible values are: - 0: success - the container is healthy and ready for use - 1: unhealthy - the container is not working correctly - 2: starting - the container is not ready for use yet, but is working correctly If the probe returns 2 ("starting") when the container has already moved out of the "starting" state then it is treated as "unhealthy" instead. For example, to check every five minutes or so that a web-server is able to serve the site's main page within three seconds: HEALTHCHECK --interval=5m --timeout=3s \ CMD curl -f http://localhost/ || exit 1 To help debug failing probes, any output text (UTF-8 encoded) that the command writes on stdout or stderr will be stored in the health status and can be queried with `docker inspect`. Such output should be kept short (only the first 4096 bytes are stored currently). When the health status of a container changes, a `health_status` event is generated with the new status. The health status is also displayed in the `docker ps` output. Signed-off-by: Thomas Leonard <thomas.leonard@docker.com> Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
2016-04-18 09:48:13 +00:00
}
}
return fmt.Errorf("SignalProcess could not find process %s in %s", processFriendlyName, containerID)
}
// Resize handles a CLI event to resize an interactive docker run or docker exec
// window.
func (clnt *client) Resize(containerID, processFriendlyName string, width, height int) error {
// Get the libcontainerd container object
clnt.lock(containerID)
defer clnt.unlock(containerID)
cont, err := clnt.getContainer(containerID)
if err != nil {
return err
}
h, w := uint16(height), uint16(width)
if processFriendlyName == InitFriendlyName {
logrus.Debugln("libcontainerd: resizing systemPID in", containerID, cont.process.systemPid)
return cont.process.hcsProcess.ResizeConsole(w, h)
}
for _, p := range cont.processes {
if p.friendlyName == processFriendlyName {
logrus.Debugln("libcontainerd: resizing exec'd process", containerID, p.systemPid)
return p.hcsProcess.ResizeConsole(w, h)
}
}
return fmt.Errorf("Resize could not find containerID %s to resize", containerID)
}
// Pause handles pause requests for containers
func (clnt *client) Pause(containerID string) error {
return errors.New("Windows: Containers cannot be paused")
}
// Resume handles resume requests for containers
func (clnt *client) Resume(containerID string) error {
return errors.New("Windows: Containers cannot be paused")
}
// Stats handles stats requests for containers
func (clnt *client) Stats(containerID string) (*Stats, error) {
// Get the libcontainerd container object
clnt.lock(containerID)
defer clnt.unlock(containerID)
container, err := clnt.getContainer(containerID)
if err != nil {
return nil, err
}
s, err := container.hcsContainer.Statistics()
if err != nil {
return nil, err
}
st := Stats(s)
return &st, nil
}
// Restore is the handler for restoring a container
func (clnt *client) Restore(containerID string, unusedOnWindows ...CreateOption) error {
// TODO Windows: Implement this. For now, just tell the backend the container exited.
logrus.Debugf("libcontainerd: Restore(%s)", containerID)
return clnt.backend.StateChanged(containerID, StateInfo{
CommonStateInfo: CommonStateInfo{
State: StateExit,
ExitCode: 1 << 31,
}})
}
// GetPidsForContainer returns a list of process IDs running in a container.
// Although implemented, this is not used in Windows.
func (clnt *client) GetPidsForContainer(containerID string) ([]int, error) {
var pids []int
clnt.lock(containerID)
defer clnt.unlock(containerID)
cont, err := clnt.getContainer(containerID)
if err != nil {
return nil, err
}
// Add the first process
pids = append(pids, int(cont.containerCommon.systemPid))
// And add all the exec'd processes
for _, p := range cont.processes {
pids = append(pids, int(p.processCommon.systemPid))
}
return pids, nil
}
// Summary returns a summary of the processes running in a container.
// This is present in Windows to support docker top. In linux, the
// engine shells out to ps to get process information. On Windows, as
// the containers could be Hyper-V containers, they would not be
// visible on the container host. However, libcontainerd does have
// that information.
func (clnt *client) Summary(containerID string) ([]Summary, error) {
// Get the libcontainerd container object
clnt.lock(containerID)
defer clnt.unlock(containerID)
container, err := clnt.getContainer(containerID)
if err != nil {
return nil, err
}
p, err := container.hcsContainer.ProcessList()
if err != nil {
return nil, err
}
pl := make([]Summary, len(p))
for i := range p {
pl[i] = Summary(p[i])
}
return pl, nil
}
// UpdateResources updates resources for a running container.
func (clnt *client) UpdateResources(containerID string, resources Resources) error {
// Updating resource isn't supported on Windows
// but we should return nil for enabling updating container
return nil
}
func (clnt *client) CreateCheckpoint(containerID string, checkpointID string, checkpointDir string, exit bool) error {
return errors.New("Windows: Containers do not support checkpoints")
}
func (clnt *client) DeleteCheckpoint(containerID string, checkpointID string, checkpointDir string) error {
return errors.New("Windows: Containers do not support checkpoints")
}
func (clnt *client) ListCheckpoints(containerID string, checkpointDir string) (*Checkpoints, error) {
return nil, errors.New("Windows: Containers do not support checkpoints")
}