moby--moby/container/monitor.go

400 lines
12 KiB
Go
Raw Normal View History

package container
import (
Remove static errors from errors package. Moving all strings to the errors package wasn't a good idea after all. Our custom implementation of Go errors predates everything that's nice and good about working with errors in Go. Take as an example what we have to do to get an error message: ```go func GetErrorMessage(err error) string { switch err.(type) { case errcode.Error: e, _ := err.(errcode.Error) return e.Message case errcode.ErrorCode: ec, _ := err.(errcode.ErrorCode) return ec.Message() default: return err.Error() } } ``` This goes against every good practice for Go development. The language already provides a simple, intuitive and standard way to get error messages, that is calling the `Error()` method from an error. Reinventing the error interface is a mistake. Our custom implementation also makes very hard to reason about errors, another nice thing about Go. I found several (>10) error declarations that we don't use anywhere. This is a clear sign about how little we know about the errors we return. I also found several error usages where the number of arguments was different than the parameters declared in the error, another clear example of how difficult is to reason about errors. Moreover, our custom implementation didn't really make easier for people to return custom HTTP status code depending on the errors. Again, it's hard to reason about when to set custom codes and how. Take an example what we have to do to extract the message and status code from an error before returning a response from the API: ```go switch err.(type) { case errcode.ErrorCode: daError, _ := err.(errcode.ErrorCode) statusCode = daError.Descriptor().HTTPStatusCode errMsg = daError.Message() case errcode.Error: // For reference, if you're looking for a particular error // then you can do something like : // import ( derr "github.com/docker/docker/errors" ) // if daError.ErrorCode() == derr.ErrorCodeNoSuchContainer { ... } daError, _ := err.(errcode.Error) statusCode = daError.ErrorCode().Descriptor().HTTPStatusCode errMsg = daError.Message default: // This part of will be removed once we've // converted everything over to use the errcode package // FIXME: this is brittle and should not be necessary. // If we need to differentiate between different possible error types, // we should create appropriate error types with clearly defined meaning errStr := strings.ToLower(err.Error()) for keyword, status := range map[string]int{ "not found": http.StatusNotFound, "no such": http.StatusNotFound, "bad parameter": http.StatusBadRequest, "conflict": http.StatusConflict, "impossible": http.StatusNotAcceptable, "wrong login/password": http.StatusUnauthorized, "hasn't been activated": http.StatusForbidden, } { if strings.Contains(errStr, keyword) { statusCode = status break } } } ``` You can notice two things in that code: 1. We have to explain how errors work, because our implementation goes against how easy to use Go errors are. 2. At no moment we arrived to remove that `switch` statement that was the original reason to use our custom implementation. This change removes all our status errors from the errors package and puts them back in their specific contexts. IT puts the messages back with their contexts. That way, we know right away when errors used and how to generate their messages. It uses custom interfaces to reason about errors. Errors that need to response with a custom status code MUST implementent this simple interface: ```go type errorWithStatus interface { HTTPErrorStatusCode() int } ``` This interface is very straightforward to implement. It also preserves Go errors real behavior, getting the message is as simple as using the `Error()` method. I included helper functions to generate errors that use custom status code in `errors/errors.go`. By doing this, we remove the hard dependency we have eeverywhere to our custom errors package. Yes, you can use it as a helper to generate error, but it's still very easy to generate errors without it. Please, read this fantastic blog post about errors in Go: http://dave.cheney.net/2014/12/24/inspecting-errors Signed-off-by: David Calavera <david.calavera@gmail.com>
2016-02-25 15:53:35 +00:00
"fmt"
"io"
"os/exec"
"strings"
"sync"
"syscall"
"time"
"github.com/Sirupsen/logrus"
"github.com/docker/docker/daemon/execdriver"
"github.com/docker/docker/pkg/promise"
"github.com/docker/docker/pkg/stringid"
"github.com/docker/engine-api/types/container"
)
const (
defaultTimeIncrement = 100
loggerCloseTimeout = 10 * time.Second
)
// supervisor defines the interface that a supervisor must implement
type supervisor interface {
// LogContainerEvent generates events related to a given container
LogContainerEvent(*Container, string)
// Cleanup ensures that the container is properly unmounted
Cleanup(*Container)
// StartLogging starts the logging driver for the container
StartLogging(*Container) error
// Run starts a container
Run(c *Container, pipes *execdriver.Pipes, startCallback execdriver.DriverCallback) (execdriver.ExitStatus, error)
// IsShuttingDown tells whether the supervisor is shutting down or not
IsShuttingDown() bool
}
// containerMonitor monitors the execution of a container's main process.
// If a restart policy is specified for the container the monitor will ensure that the
// process is restarted based on the rules of the policy. When the container is finally stopped
// the monitor will reset and cleanup any of the container resources such as networking allocations
// and the rootfs
type containerMonitor struct {
mux sync.Mutex
// supervisor keeps track of the container and the events it generates
supervisor supervisor
// container is the container being monitored
container *Container
// restartPolicy is the current policy being applied to the container monitor
restartPolicy container.RestartPolicy
// failureCount is the number of times the container has failed to
// start in a row
failureCount int
// shouldStop signals the monitor that the next time the container exits it is
// either because docker or the user asked for the container to be stopped
shouldStop bool
// startSignal is a channel that is closes after the container initially starts
startSignal chan struct{}
// stopChan is used to signal to the monitor whenever there is a wait for the
// next restart so that the timeIncrement is not honored and the user is not
// left waiting for nothing to happen during this time
stopChan chan struct{}
// timeIncrement is the amount of time to wait between restarts
// this is in milliseconds
timeIncrement int
// lastStartTime is the time which the monitor last exec'd the container's process
lastStartTime time.Time
}
// StartMonitor initializes a containerMonitor for this container with the provided supervisor and restart policy
// and starts the container's process.
func (container *Container) StartMonitor(s supervisor) error {
container.monitor = &containerMonitor{
supervisor: s,
container: container,
restartPolicy: container.HostConfig.RestartPolicy,
timeIncrement: defaultTimeIncrement,
stopChan: make(chan struct{}),
startSignal: make(chan struct{}),
}
return container.monitor.wait()
}
// wait starts the container and wait until
// we either receive an error from the initial start of the container's
// process or until the process is running in the container
func (m *containerMonitor) wait() error {
select {
case <-m.startSignal:
case err := <-promise.Go(m.start):
return err
}
return nil
}
// Stop signals to the container monitor that it should stop monitoring the container
// for exits the next time the process dies
func (m *containerMonitor) ExitOnNext() {
m.mux.Lock()
// we need to protect having a double close of the channel when stop is called
// twice or else we will get a panic
if !m.shouldStop {
m.shouldStop = true
close(m.stopChan)
}
m.mux.Unlock()
}
// Close closes the container's resources such as networking allocations and
// unmounts the container's root filesystem
func (m *containerMonitor) Close() error {
// Cleanup networking and mounts
m.supervisor.Cleanup(m.container)
// FIXME: here is race condition between two RUN instructions in Dockerfile
// because they share same runconfig and change image. Must be fixed
// in builder/builder.go
if err := m.container.ToDisk(); err != nil {
logrus.Errorf("Error dumping container %s state to disk: %s", m.container.ID, err)
return err
}
return nil
}
// Start starts the containers process and monitors it according to the restart policy
func (m *containerMonitor) start() error {
var (
err error
exitStatus execdriver.ExitStatus
// this variable indicates where we in execution flow:
// before Run or after
afterRun bool
)
// ensure that when the monitor finally exits we release the networking and unmount the rootfs
defer func() {
if afterRun {
m.container.Lock()
defer m.container.Unlock()
m.container.SetStopped(&exitStatus)
}
m.Close()
}()
// reset stopped flag
if m.container.HasBeenManuallyStopped {
m.container.HasBeenManuallyStopped = false
}
// reset the restart count
m.container.RestartCount = -1
for {
m.container.RestartCount++
if err := m.supervisor.StartLogging(m.container); err != nil {
m.resetContainer(false)
return err
}
pipes := execdriver.NewPipes(m.container.Stdin(), m.container.Stdout(), m.container.Stderr(), m.container.Config.OpenStdin)
m.logEvent("start")
m.lastStartTime = time.Now()
if exitStatus, err = m.supervisor.Run(m.container, pipes, m.callback); err != nil {
// if we receive an internal error from the initial start of a container then lets
// return it instead of entering the restart loop
// set to 127 for container cmd not found/does not exist)
if strings.Contains(err.Error(), "executable file not found") ||
strings.Contains(err.Error(), "no such file or directory") ||
strings.Contains(err.Error(), "system cannot find the file specified") {
if m.container.RestartCount == 0 {
m.container.ExitCode = 127
m.resetContainer(false)
Remove static errors from errors package. Moving all strings to the errors package wasn't a good idea after all. Our custom implementation of Go errors predates everything that's nice and good about working with errors in Go. Take as an example what we have to do to get an error message: ```go func GetErrorMessage(err error) string { switch err.(type) { case errcode.Error: e, _ := err.(errcode.Error) return e.Message case errcode.ErrorCode: ec, _ := err.(errcode.ErrorCode) return ec.Message() default: return err.Error() } } ``` This goes against every good practice for Go development. The language already provides a simple, intuitive and standard way to get error messages, that is calling the `Error()` method from an error. Reinventing the error interface is a mistake. Our custom implementation also makes very hard to reason about errors, another nice thing about Go. I found several (>10) error declarations that we don't use anywhere. This is a clear sign about how little we know about the errors we return. I also found several error usages where the number of arguments was different than the parameters declared in the error, another clear example of how difficult is to reason about errors. Moreover, our custom implementation didn't really make easier for people to return custom HTTP status code depending on the errors. Again, it's hard to reason about when to set custom codes and how. Take an example what we have to do to extract the message and status code from an error before returning a response from the API: ```go switch err.(type) { case errcode.ErrorCode: daError, _ := err.(errcode.ErrorCode) statusCode = daError.Descriptor().HTTPStatusCode errMsg = daError.Message() case errcode.Error: // For reference, if you're looking for a particular error // then you can do something like : // import ( derr "github.com/docker/docker/errors" ) // if daError.ErrorCode() == derr.ErrorCodeNoSuchContainer { ... } daError, _ := err.(errcode.Error) statusCode = daError.ErrorCode().Descriptor().HTTPStatusCode errMsg = daError.Message default: // This part of will be removed once we've // converted everything over to use the errcode package // FIXME: this is brittle and should not be necessary. // If we need to differentiate between different possible error types, // we should create appropriate error types with clearly defined meaning errStr := strings.ToLower(err.Error()) for keyword, status := range map[string]int{ "not found": http.StatusNotFound, "no such": http.StatusNotFound, "bad parameter": http.StatusBadRequest, "conflict": http.StatusConflict, "impossible": http.StatusNotAcceptable, "wrong login/password": http.StatusUnauthorized, "hasn't been activated": http.StatusForbidden, } { if strings.Contains(errStr, keyword) { statusCode = status break } } } ``` You can notice two things in that code: 1. We have to explain how errors work, because our implementation goes against how easy to use Go errors are. 2. At no moment we arrived to remove that `switch` statement that was the original reason to use our custom implementation. This change removes all our status errors from the errors package and puts them back in their specific contexts. IT puts the messages back with their contexts. That way, we know right away when errors used and how to generate their messages. It uses custom interfaces to reason about errors. Errors that need to response with a custom status code MUST implementent this simple interface: ```go type errorWithStatus interface { HTTPErrorStatusCode() int } ``` This interface is very straightforward to implement. It also preserves Go errors real behavior, getting the message is as simple as using the `Error()` method. I included helper functions to generate errors that use custom status code in `errors/errors.go`. By doing this, we remove the hard dependency we have eeverywhere to our custom errors package. Yes, you can use it as a helper to generate error, but it's still very easy to generate errors without it. Please, read this fantastic blog post about errors in Go: http://dave.cheney.net/2014/12/24/inspecting-errors Signed-off-by: David Calavera <david.calavera@gmail.com>
2016-02-25 15:53:35 +00:00
return fmt.Errorf("Container command not found or does not exist.")
}
}
// set to 126 for container cmd can't be invoked errors
if strings.Contains(err.Error(), syscall.EACCES.Error()) {
if m.container.RestartCount == 0 {
m.container.ExitCode = 126
m.resetContainer(false)
Remove static errors from errors package. Moving all strings to the errors package wasn't a good idea after all. Our custom implementation of Go errors predates everything that's nice and good about working with errors in Go. Take as an example what we have to do to get an error message: ```go func GetErrorMessage(err error) string { switch err.(type) { case errcode.Error: e, _ := err.(errcode.Error) return e.Message case errcode.ErrorCode: ec, _ := err.(errcode.ErrorCode) return ec.Message() default: return err.Error() } } ``` This goes against every good practice for Go development. The language already provides a simple, intuitive and standard way to get error messages, that is calling the `Error()` method from an error. Reinventing the error interface is a mistake. Our custom implementation also makes very hard to reason about errors, another nice thing about Go. I found several (>10) error declarations that we don't use anywhere. This is a clear sign about how little we know about the errors we return. I also found several error usages where the number of arguments was different than the parameters declared in the error, another clear example of how difficult is to reason about errors. Moreover, our custom implementation didn't really make easier for people to return custom HTTP status code depending on the errors. Again, it's hard to reason about when to set custom codes and how. Take an example what we have to do to extract the message and status code from an error before returning a response from the API: ```go switch err.(type) { case errcode.ErrorCode: daError, _ := err.(errcode.ErrorCode) statusCode = daError.Descriptor().HTTPStatusCode errMsg = daError.Message() case errcode.Error: // For reference, if you're looking for a particular error // then you can do something like : // import ( derr "github.com/docker/docker/errors" ) // if daError.ErrorCode() == derr.ErrorCodeNoSuchContainer { ... } daError, _ := err.(errcode.Error) statusCode = daError.ErrorCode().Descriptor().HTTPStatusCode errMsg = daError.Message default: // This part of will be removed once we've // converted everything over to use the errcode package // FIXME: this is brittle and should not be necessary. // If we need to differentiate between different possible error types, // we should create appropriate error types with clearly defined meaning errStr := strings.ToLower(err.Error()) for keyword, status := range map[string]int{ "not found": http.StatusNotFound, "no such": http.StatusNotFound, "bad parameter": http.StatusBadRequest, "conflict": http.StatusConflict, "impossible": http.StatusNotAcceptable, "wrong login/password": http.StatusUnauthorized, "hasn't been activated": http.StatusForbidden, } { if strings.Contains(errStr, keyword) { statusCode = status break } } } ``` You can notice two things in that code: 1. We have to explain how errors work, because our implementation goes against how easy to use Go errors are. 2. At no moment we arrived to remove that `switch` statement that was the original reason to use our custom implementation. This change removes all our status errors from the errors package and puts them back in their specific contexts. IT puts the messages back with their contexts. That way, we know right away when errors used and how to generate their messages. It uses custom interfaces to reason about errors. Errors that need to response with a custom status code MUST implementent this simple interface: ```go type errorWithStatus interface { HTTPErrorStatusCode() int } ``` This interface is very straightforward to implement. It also preserves Go errors real behavior, getting the message is as simple as using the `Error()` method. I included helper functions to generate errors that use custom status code in `errors/errors.go`. By doing this, we remove the hard dependency we have eeverywhere to our custom errors package. Yes, you can use it as a helper to generate error, but it's still very easy to generate errors without it. Please, read this fantastic blog post about errors in Go: http://dave.cheney.net/2014/12/24/inspecting-errors Signed-off-by: David Calavera <david.calavera@gmail.com>
2016-02-25 15:53:35 +00:00
return fmt.Errorf("Container command could not be invoked.")
}
}
if m.container.RestartCount == 0 {
m.container.ExitCode = -1
m.resetContainer(false)
Remove static errors from errors package. Moving all strings to the errors package wasn't a good idea after all. Our custom implementation of Go errors predates everything that's nice and good about working with errors in Go. Take as an example what we have to do to get an error message: ```go func GetErrorMessage(err error) string { switch err.(type) { case errcode.Error: e, _ := err.(errcode.Error) return e.Message case errcode.ErrorCode: ec, _ := err.(errcode.ErrorCode) return ec.Message() default: return err.Error() } } ``` This goes against every good practice for Go development. The language already provides a simple, intuitive and standard way to get error messages, that is calling the `Error()` method from an error. Reinventing the error interface is a mistake. Our custom implementation also makes very hard to reason about errors, another nice thing about Go. I found several (>10) error declarations that we don't use anywhere. This is a clear sign about how little we know about the errors we return. I also found several error usages where the number of arguments was different than the parameters declared in the error, another clear example of how difficult is to reason about errors. Moreover, our custom implementation didn't really make easier for people to return custom HTTP status code depending on the errors. Again, it's hard to reason about when to set custom codes and how. Take an example what we have to do to extract the message and status code from an error before returning a response from the API: ```go switch err.(type) { case errcode.ErrorCode: daError, _ := err.(errcode.ErrorCode) statusCode = daError.Descriptor().HTTPStatusCode errMsg = daError.Message() case errcode.Error: // For reference, if you're looking for a particular error // then you can do something like : // import ( derr "github.com/docker/docker/errors" ) // if daError.ErrorCode() == derr.ErrorCodeNoSuchContainer { ... } daError, _ := err.(errcode.Error) statusCode = daError.ErrorCode().Descriptor().HTTPStatusCode errMsg = daError.Message default: // This part of will be removed once we've // converted everything over to use the errcode package // FIXME: this is brittle and should not be necessary. // If we need to differentiate between different possible error types, // we should create appropriate error types with clearly defined meaning errStr := strings.ToLower(err.Error()) for keyword, status := range map[string]int{ "not found": http.StatusNotFound, "no such": http.StatusNotFound, "bad parameter": http.StatusBadRequest, "conflict": http.StatusConflict, "impossible": http.StatusNotAcceptable, "wrong login/password": http.StatusUnauthorized, "hasn't been activated": http.StatusForbidden, } { if strings.Contains(errStr, keyword) { statusCode = status break } } } ``` You can notice two things in that code: 1. We have to explain how errors work, because our implementation goes against how easy to use Go errors are. 2. At no moment we arrived to remove that `switch` statement that was the original reason to use our custom implementation. This change removes all our status errors from the errors package and puts them back in their specific contexts. IT puts the messages back with their contexts. That way, we know right away when errors used and how to generate their messages. It uses custom interfaces to reason about errors. Errors that need to response with a custom status code MUST implementent this simple interface: ```go type errorWithStatus interface { HTTPErrorStatusCode() int } ``` This interface is very straightforward to implement. It also preserves Go errors real behavior, getting the message is as simple as using the `Error()` method. I included helper functions to generate errors that use custom status code in `errors/errors.go`. By doing this, we remove the hard dependency we have eeverywhere to our custom errors package. Yes, you can use it as a helper to generate error, but it's still very easy to generate errors without it. Please, read this fantastic blog post about errors in Go: http://dave.cheney.net/2014/12/24/inspecting-errors Signed-off-by: David Calavera <david.calavera@gmail.com>
2016-02-25 15:53:35 +00:00
return fmt.Errorf("Cannot start container %s: %v", m.container.ID, err)
}
logrus.Errorf("Error running container: %s", err)
}
// here container.Lock is already lost
afterRun = true
m.resetMonitor(err == nil && exitStatus.ExitCode == 0)
if m.shouldRestart(exitStatus.ExitCode) {
m.container.SetRestartingLocking(&exitStatus)
m.logEvent("die")
m.resetContainer(true)
// sleep with a small time increment between each restart to help avoid issues cased by quickly
// restarting the container because of some types of errors ( networking cut out, etc... )
m.waitForNextRestart()
// we need to check this before reentering the loop because the waitForNextRestart could have
// been terminated by a request from a user
if m.shouldStop {
return err
}
continue
}
m.logEvent("die")
m.resetContainer(true)
return err
}
}
// resetMonitor resets the stateful fields on the containerMonitor based on the
// previous runs success or failure. Regardless of success, if the container had
// an execution time of more than 10s then reset the timer back to the default
func (m *containerMonitor) resetMonitor(successful bool) {
executionTime := time.Now().Sub(m.lastStartTime).Seconds()
if executionTime > 10 {
m.timeIncrement = defaultTimeIncrement
} else {
// otherwise we need to increment the amount of time we wait before restarting
// the process. We will build up by multiplying the increment by 2
m.timeIncrement *= 2
}
// the container exited successfully so we need to reset the failure counter
if successful {
m.failureCount = 0
} else {
m.failureCount++
}
}
// waitForNextRestart waits with the default time increment to restart the container unless
// a user or docker asks for the container to be stopped
func (m *containerMonitor) waitForNextRestart() {
select {
case <-time.After(time.Duration(m.timeIncrement) * time.Millisecond):
case <-m.stopChan:
}
}
// shouldRestart checks the restart policy and applies the rules to determine if
// the container's process should be restarted
func (m *containerMonitor) shouldRestart(exitCode int) bool {
m.mux.Lock()
defer m.mux.Unlock()
// do not restart if the user or docker has requested that this container be stopped
if m.shouldStop {
m.container.HasBeenManuallyStopped = !m.supervisor.IsShuttingDown()
return false
}
switch {
case m.restartPolicy.IsAlways(), m.restartPolicy.IsUnlessStopped():
return true
case m.restartPolicy.IsOnFailure():
// the default value of 0 for MaximumRetryCount means that we will not enforce a maximum count
if max := m.restartPolicy.MaximumRetryCount; max != 0 && m.failureCount > max {
logrus.Debugf("stopping restart of container %s because maximum failure could of %d has been reached",
stringid.TruncateID(m.container.ID), max)
return false
}
return exitCode != 0
}
return false
}
// callback ensures that the container's state is properly updated after we
// received ack from the execution drivers
func (m *containerMonitor) callback(processConfig *execdriver.ProcessConfig, pid int, chOOM <-chan struct{}) error {
go func() {
for range chOOM {
m.logEvent("oom")
}
}()
if processConfig.Tty {
// The callback is called after the process start()
// so we are in the parent process. In TTY mode, stdin/out/err is the PtySlave
// which we close here.
if c, ok := processConfig.Stdout.(io.Closer); ok {
c.Close()
}
}
m.container.SetRunning(pid)
// signal that the process has started
// close channel only if not closed
select {
case <-m.startSignal:
default:
close(m.startSignal)
}
if err := m.container.ToDiskLocking(); err != nil {
logrus.Errorf("Error saving container to disk: %v", err)
}
return nil
}
// resetContainer resets the container's IO and ensures that the command is able to be executed again
// by copying the data into a new struct
// if lock is true, then container locked during reset
func (m *containerMonitor) resetContainer(lock bool) {
container := m.container
if lock {
container.Lock()
defer container.Unlock()
}
if err := container.CloseStreams(); err != nil {
logrus.Errorf("%s: %s", container.ID, err)
}
if container.Command != nil && container.Command.ProcessConfig.Terminal != nil {
if err := container.Command.ProcessConfig.Terminal.Close(); err != nil {
logrus.Errorf("%s: Error closing terminal: %s", container.ID, err)
}
}
// Re-create a brand new stdin pipe once the container exited
if container.Config.OpenStdin {
container.NewInputPipes()
}
if container.LogDriver != nil {
if container.LogCopier != nil {
exit := make(chan struct{})
go func() {
container.LogCopier.Wait()
close(exit)
}()
select {
case <-time.After(loggerCloseTimeout):
logrus.Warnf("Logger didn't exit in time: logs may be truncated")
container.LogCopier.Close()
// always waits for the LogCopier to finished before closing
<-exit
case <-exit:
}
}
container.LogDriver.Close()
container.LogCopier = nil
container.LogDriver = nil
}
c := container.Command.ProcessConfig.Cmd
container.Command.ProcessConfig.Cmd = exec.Cmd{
Stdin: c.Stdin,
Stdout: c.Stdout,
Stderr: c.Stderr,
Path: c.Path,
Env: c.Env,
ExtraFiles: c.ExtraFiles,
Args: c.Args,
Dir: c.Dir,
SysProcAttr: c.SysProcAttr,
}
}
func (m *containerMonitor) logEvent(action string) {
m.supervisor.LogContainerEvent(m.container, action)
}