Package org.apache.hadoop.yarn.server.nodemanager

Class ContainerExecutor

java.lang.Object

org.apache.hadoop.yarn.server.nodemanager.ContainerExecutor

All Implemented Interfaces:

org.apache.hadoop.conf.Configurable

Direct Known Subclasses:

DefaultContainerExecutor, LinuxContainerExecutor

public abstract class ContainerExecutor
extends java.lang.Object
implements org.apache.hadoop.conf.Configurable

This class is abstraction of the mechanism used to launch a container on the underlying OS. All executor implementations must extend ContainerExecutor.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static class	ContainerExecutor.DelayedProcessKiller	This class will signal a target container after a specified delay.
static class	ContainerExecutor.ExitCode	The container exit code.
static class	ContainerExecutor.Signal	The constants for the signals.

Field Summary

Fields

Modifier and Type	Field	Description
static java.lang.String	DIRECTORY_CONTENTS	The relative path to which debug information will be written.
static org.apache.hadoop.fs.permission.FsPermission	TASK_LAUNCH_SCRIPT_PERMISSION	The permissions to use when creating the launch script.
static java.lang.String	TOKEN_FILE_NAME_FMT
protected static java.lang.String	WILDCARD

Constructor Summary

Constructors

Constructor	Description
ContainerExecutor()

Method Summary

Modifier and Type	Method	Description
void	activateContainer(org.apache.hadoop.yarn.api.records.ContainerId containerId, org.apache.hadoop.fs.Path pidFilePath)	Mark the container as active.
void	cleanupBeforeRelaunch(Container container)	Perform any cleanup before the next launch of the container.
void	deactivateContainer(org.apache.hadoop.yarn.api.records.ContainerId containerId)	Mark the container as inactive.
abstract void	deleteAsUser(DeletionAsUserContext ctx)	Delete specified directories as a given user.
abstract org.apache.hadoop.hdfs.protocol.datatransfer.IOStreamPair	execContainer(ContainerExecContext ctx)	Perform interactive docker command into running container.
org.apache.hadoop.conf.Configuration	getConf()
java.lang.String	getExposedPorts(Container container)
java.lang.String[]	getIpAndHost(Container container)
static java.lang.String[]	getLocalIpAndHost(Container container)
java.util.Map<java.lang.String,org.apache.hadoop.yarn.api.records.LocalResource>	getLocalResources(Container container)
protected java.lang.String	getNMEnvVar(java.lang.String varname)
protected org.apache.hadoop.fs.Path	getPidFilePath(org.apache.hadoop.yarn.api.records.ContainerId containerId)	Get the pidFile of the container.
java.lang.String	getProcessId(org.apache.hadoop.yarn.api.records.ContainerId containerID)	Get the process-identifier for the container.
protected java.lang.String[]	getRunCommand(java.lang.String command, java.lang.String groupId, java.lang.String userName, org.apache.hadoop.fs.Path pidFile, org.apache.hadoop.conf.Configuration config)	Return a command line to execute the given command in the OS shell.
protected java.lang.String[]	getRunCommand(java.lang.String command, java.lang.String groupId, java.lang.String userName, org.apache.hadoop.fs.Path pidFile, org.apache.hadoop.conf.Configuration config, org.apache.hadoop.yarn.api.records.Resource resource)	Return a command line to execute the given command in the OS shell.
protected java.lang.String[]	getRunCommandForOther(java.lang.String command, org.apache.hadoop.conf.Configuration config)	Return a command line to execute the given command in the OS shell.
protected java.lang.String[]	getRunCommandForWindows(java.lang.String command, java.lang.String groupId, java.lang.String userName, org.apache.hadoop.fs.Path pidFile, org.apache.hadoop.conf.Configuration config, org.apache.hadoop.yarn.api.records.Resource resource)	Return a command line to execute the given command in the OS shell.
abstract void	init(Context nmContext)	Run the executor initialization steps.
protected boolean	isContainerActive(org.apache.hadoop.yarn.api.records.ContainerId containerId)	Return whether the container is still active.
abstract boolean	isContainerAlive(ContainerLivenessContext ctx)	Check if a container is alive.
abstract int	launchContainer(ContainerStartContext ctx)	Launch the container on the node.
org.apache.hadoop.fs.Path	localizeClasspathJar(org.apache.hadoop.fs.Path jarPath, org.apache.hadoop.fs.Path target, java.lang.String owner)	This function localizes the JAR file on-demand.
protected void	logOutput(java.lang.String output)	Log each line of the output string as INFO level log messages.
void	pauseContainer(Container container)	Pause the container.
void	prepareContainer(ContainerPrepareContext ctx)	Prepare the container prior to the launch environment being written.
int	reacquireContainer(ContainerReacquisitionContext ctx)	Recover an already existing container.
protected java.io.File[]	readDirAsUser(java.lang.String user, org.apache.hadoop.fs.Path dir)	Return the files in the target directory.
abstract boolean	reapContainer(ContainerReapContext ctx)	Perform the steps necessary to reap the container.
abstract int	relaunchContainer(ContainerStartContext ctx)	Relaunch the container on the node.
void	resumeContainer(Container container)	Resume the container from pause state.
void	setConf(org.apache.hadoop.conf.Configuration conf)
abstract boolean	signalContainer(ContainerSignalContext ctx)	Signal container with the specified signal.
void	start()
abstract void	startLocalizer(LocalizerStartContext ctx)	Prepare the environment for containers in this application to execute.
void	stop()
abstract void	symLink(java.lang.String target, java.lang.String symlink)	Create a symlink file which points to the target.
abstract void	updateYarnSysFS(Context ctx, java.lang.String user, java.lang.String appId, java.lang.String spec)	Update cluster information inside container.
void	writeLaunchEnv(java.io.OutputStream out, java.util.Map<java.lang.String,java.lang.String> environment, java.util.Map<org.apache.hadoop.fs.Path,java.util.List<java.lang.String>> resources, java.util.List<java.lang.String> command, org.apache.hadoop.fs.Path logDir, java.lang.String user, java.lang.String outFilename, java.util.LinkedHashSet<java.lang.String> nmVars)	This method writes out the launch environment of a container to a specified path.
void	writeLaunchEnv(java.io.OutputStream out, java.util.Map<java.lang.String,java.lang.String> environment, java.util.Map<org.apache.hadoop.fs.Path,java.util.List<java.lang.String>> resources, java.util.List<java.lang.String> command, org.apache.hadoop.fs.Path logDir, java.lang.String user, java.util.LinkedHashSet<java.lang.String> nmVars)	This method writes out the launch environment of a container to the default container launch script.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

WILDCARD

protected static final java.lang.String WILDCARD

See Also:

Constant Field Values

TOKEN_FILE_NAME_FMT

public static final java.lang.String TOKEN_FILE_NAME_FMT

See Also:

Constant Field Values

TASK_LAUNCH_SCRIPT_PERMISSION

public static final org.apache.hadoop.fs.permission.FsPermission TASK_LAUNCH_SCRIPT_PERMISSION

The permissions to use when creating the launch script.

DIRECTORY_CONTENTS

public static final java.lang.String DIRECTORY_CONTENTS

The relative path to which debug information will be written.

See Also:

ContainerLaunch.ShellScriptBuilder.listDebugInformation(org.apache.hadoop.fs.Path), Constant Field Values

Constructor Detail

ContainerExecutor

public ContainerExecutor()

Method Detail

setConf

public void setConf(org.apache.hadoop.conf.Configuration conf)

Specified by:

setConf in interface org.apache.hadoop.conf.Configurable

getConf

public org.apache.hadoop.conf.Configuration getConf()

Specified by:

getConf in interface org.apache.hadoop.conf.Configurable

init

public abstract void init(Context nmContext)
                   throws java.io.IOException

Run the executor initialization steps. Verify that the necessary configs and permissions are in place.

Parameters:

nmContext - Context of NM

Throws:

java.io.IOException - if initialization fails

start

public void start()

stop

public void stop()

localizeClasspathJar

public org.apache.hadoop.fs.Path localizeClasspathJar(org.apache.hadoop.fs.Path jarPath,
                                                      org.apache.hadoop.fs.Path target,
                                                      java.lang.String owner)
                                               throws java.io.IOException

This function localizes the JAR file on-demand. On Windows the ContainerLaunch creates a temporary special JAR manifest of other JARs to workaround the CLASSPATH length. In a secure cluster this JAR must be localized so that the container has access to it. The default implementation returns the classpath passed to it, which is expected to have been created in the node manager's fprivate folder, which will not work with secure Windows clusters.

Parameters:

jarPath - the path to the JAR to localize

target - the directory where the JAR file should be localized

owner - the name of the user who should own the localized file

Returns:

the path to the localized JAR file

Throws:

java.io.IOException - if localization fails

startLocalizer

public abstract void startLocalizer(LocalizerStartContext ctx)
                             throws java.io.IOException,
                                    java.lang.InterruptedException

Prepare the environment for containers in this application to execute.

 For $x in local.dirs
   create $x/$user/$appId
 Copy $nmLocal/appTokens -> $N/$user/$appId
 For $rsrc in private resources
   Copy $rsrc -> $N/$user/filecache/[idef]
 For $rsrc in job resources
   Copy $rsrc -> $N/$user/$appId/filecache/idef
 

Parameters:

ctx - LocalizerStartContext that encapsulates necessary information for starting a localizer.

Throws:

java.io.IOException - for most application init failures

java.lang.InterruptedException - if application init thread is halted by NM

prepareContainer

public void prepareContainer(ContainerPrepareContext ctx)
                      throws java.io.IOException

Prepare the container prior to the launch environment being written.

Parameters:

ctx - Encapsulates information necessary for launching containers.

Throws:

java.io.IOException - if errors occur during container preparation

launchContainer

public abstract int launchContainer(ContainerStartContext ctx)
                             throws java.io.IOException,
                                    org.apache.hadoop.yarn.exceptions.ConfigurationException

Launch the container on the node. This is a blocking call and returns only when the container exits.

Parameters:

ctx - Encapsulates information necessary for launching containers.

Returns:

the return status of the launch

Throws:

java.io.IOException - if the container launch fails

org.apache.hadoop.yarn.exceptions.ConfigurationException - if config error was found

relaunchContainer

public abstract int relaunchContainer(ContainerStartContext ctx)
                               throws java.io.IOException,
                                      org.apache.hadoop.yarn.exceptions.ConfigurationException

Relaunch the container on the node. This is a blocking call and returns only when the container exits.

Parameters:

ctx - Encapsulates information necessary for relaunching containers.

Returns:

the return status of the relaunch

Throws:

java.io.IOException - if the container relaunch fails

org.apache.hadoop.yarn.exceptions.ConfigurationException - if config error was found

signalContainer

public abstract boolean signalContainer(ContainerSignalContext ctx)
                                 throws java.io.IOException

Signal container with the specified signal.

Parameters:

ctx - Encapsulates information necessary for signaling containers.

Returns:

returns true if the operation succeeded

Throws:

java.io.IOException - if signaling the container fails

reapContainer

public abstract boolean reapContainer(ContainerReapContext ctx)
                               throws java.io.IOException

Perform the steps necessary to reap the container.

Parameters:

ctx - Encapsulates information necessary for reaping containers.

Returns:

returns true if the operation succeeded.

Throws:

java.io.IOException - if reaping the container fails.

execContainer

public abstract org.apache.hadoop.hdfs.protocol.datatransfer.IOStreamPair execContainer(ContainerExecContext ctx)
                                                                                 throws ContainerExecutionException

Perform interactive docker command into running container.

Parameters:

ctx - Encapsulates information necessary for exec containers.

Returns:

return input/output stream if the operation succeeded.

Throws:

ContainerExecutionException - if container exec fails.

deleteAsUser

public abstract void deleteAsUser(DeletionAsUserContext ctx)
                           throws java.io.IOException,
                                  java.lang.InterruptedException

Delete specified directories as a given user.

Parameters:

ctx - Encapsulates information necessary for deletion.

Throws:

java.io.IOException - if delete fails

java.lang.InterruptedException - if interrupted while waiting for the deletion operation to complete

symLink

public abstract void symLink(java.lang.String target,
                             java.lang.String symlink)
                      throws java.io.IOException

Create a symlink file which points to the target.

Parameters:

target - The target for symlink

symlink - the symlink file

Throws:

java.io.IOException - Error when creating symlinks

isContainerAlive

public abstract boolean isContainerAlive(ContainerLivenessContext ctx)
                                  throws java.io.IOException

Check if a container is alive.

Parameters:

ctx - Encapsulates information necessary for container liveness check.

Returns:

true if container is still alive

Throws:

java.io.IOException - if there is a failure while checking the container status

getLocalResources

public java.util.Map<java.lang.String,org.apache.hadoop.yarn.api.records.LocalResource> getLocalResources(Container container)
                                                                                                         throws java.io.IOException

Throws:

java.io.IOException

updateYarnSysFS

public abstract void updateYarnSysFS(Context ctx,
                                     java.lang.String user,
                                     java.lang.String appId,
                                     java.lang.String spec)
                              throws java.io.IOException

Update cluster information inside container.

Parameters:

ctx - ContainerRuntimeContext

user - Owner of application

appId - YARN application ID

spec - Service Specification

Throws:

java.io.IOException - if there is a failure while writing spec to disk

reacquireContainer

public int reacquireContainer(ContainerReacquisitionContext ctx)
                       throws java.io.IOException,
                              java.lang.InterruptedException

Recover an already existing container. This is a blocking call and returns only when the container exits. Note that the container must have been activated prior to this call.

Parameters:

ctx - encapsulates information necessary to reacquire container

Returns:

The exit code of the pre-existing container

Throws:

java.io.IOException - if there is a failure while reacquiring the container

java.lang.InterruptedException - if interrupted while waiting to reacquire the container

writeLaunchEnv

public void writeLaunchEnv(java.io.OutputStream out,
                           java.util.Map<java.lang.String,java.lang.String> environment,
                           java.util.Map<org.apache.hadoop.fs.Path,java.util.List<java.lang.String>> resources,
                           java.util.List<java.lang.String> command,
                           org.apache.hadoop.fs.Path logDir,
                           java.lang.String user,
                           java.util.LinkedHashSet<java.lang.String> nmVars)
                    throws java.io.IOException

This method writes out the launch environment of a container to the default container launch script. For the default container script path see ContainerLaunch.CONTAINER_SCRIPT.

Parameters:

out - the output stream to which the environment is written (usually a script file which will be executed by the Launcher)

environment - the environment variables and their values

resources - the resources which have been localized for this container. Symlinks will be created to these localized resources

command - the command that will be run

logDir - the log dir to which to copy debugging information

user - the username of the job owner

nmVars - the set of environment vars that are explicitly set by NM

Throws:

java.io.IOException - if any errors happened writing to the OutputStream, while creating symlinks

writeLaunchEnv

@VisibleForTesting
public void writeLaunchEnv(java.io.OutputStream out,
                           java.util.Map<java.lang.String,java.lang.String> environment,
                           java.util.Map<org.apache.hadoop.fs.Path,java.util.List<java.lang.String>> resources,
                           java.util.List<java.lang.String> command,
                           org.apache.hadoop.fs.Path logDir,
                           java.lang.String user,
                           java.lang.String outFilename,
                           java.util.LinkedHashSet<java.lang.String> nmVars)
                    throws java.io.IOException

This method writes out the launch environment of a container to a specified path.

Parameters:

out - the output stream to which the environment is written (usually a script file which will be executed by the Launcher)

environment - the environment variables and their values

resources - the resources which have been localized for this container. Symlinks will be created to these localized resources

command - the command that will be run

logDir - the log dir to which to copy debugging information

user - the username of the job owner

outFilename - the path to which to write the launch environment

nmVars - the set of environment vars that are explicitly set by NM

Throws:

java.io.IOException - if any errors happened writing to the OutputStream, while creating symlinks

readDirAsUser

protected java.io.File[] readDirAsUser(java.lang.String user,
                                       org.apache.hadoop.fs.Path dir)

Return the files in the target directory. If retrieving the list of files requires specific access rights, that access will happen as the specified user. The list will not include entries for "." or "..".

Parameters:

user - the user as whom to access the target directory

dir - the target directory

Returns:

a list of files in the target directory

logOutput

protected void logOutput(java.lang.String output)

Log each line of the output string as INFO level log messages.

Parameters:

output - the output string to log

getPidFilePath

protected org.apache.hadoop.fs.Path getPidFilePath(org.apache.hadoop.yarn.api.records.ContainerId containerId)

Get the pidFile of the container.

Parameters:

containerId - the container ID

Returns:

the path of the pid-file for the given containerId.

getRunCommand

protected java.lang.String[] getRunCommand(java.lang.String command,
                                           java.lang.String groupId,
                                           java.lang.String userName,
                                           org.apache.hadoop.fs.Path pidFile,
                                           org.apache.hadoop.conf.Configuration config)

Return a command line to execute the given command in the OS shell. On Windows, the {code}groupId{code} parameter can be used to launch and associate the given GID with a process group. On non-Windows hosts, the {code}groupId{code} parameter is ignored.

Parameters:

command - the command to execute

groupId - the job owner's GID

userName - the job owner's username

pidFile - the path to the container's PID file

config - the configuration

Returns:

the command line to execute

getRunCommand

protected java.lang.String[] getRunCommand(java.lang.String command,
                                           java.lang.String groupId,
                                           java.lang.String userName,
                                           org.apache.hadoop.fs.Path pidFile,
                                           org.apache.hadoop.conf.Configuration config,
                                           org.apache.hadoop.yarn.api.records.Resource resource)

Return a command line to execute the given command in the OS shell. On Windows, the {code}groupId{code} parameter can be used to launch and associate the given GID with a process group. On non-Windows hosts, the {code}groupId{code} parameter is ignored.

Parameters:

command - the command to execute

groupId - the job owner's GID for Windows. On other operating systems it is ignored.

userName - the job owner's username for Windows. On other operating systems it is ignored.

pidFile - the path to the container's PID file on Windows. On other operating systems it is ignored.

config - the configuration

resource - on Windows this parameter controls memory and CPU limits. If null, no limits are set. On other operating systems it is ignored.

Returns:

the command line to execute

getRunCommandForWindows

protected java.lang.String[] getRunCommandForWindows(java.lang.String command,
                                                     java.lang.String groupId,
                                                     java.lang.String userName,
                                                     org.apache.hadoop.fs.Path pidFile,
                                                     org.apache.hadoop.conf.Configuration config,
                                                     org.apache.hadoop.yarn.api.records.Resource resource)

Return a command line to execute the given command in the OS shell. The {code}groupId{code} parameter can be used to launch and associate the given GID with a process group.

Parameters:

command - the command to execute

groupId - the job owner's GID

userName - the job owner's username

pidFile - the path to the container's PID file

config - the configuration

resource - this parameter controls memory and CPU limits. If null, no limits are set.

Returns:

the command line to execute

getRunCommandForOther

protected java.lang.String[] getRunCommandForOther(java.lang.String command,
                                                   org.apache.hadoop.conf.Configuration config)

Return a command line to execute the given command in the OS shell.

Parameters:

command - the command to execute

config - the configuration

Returns:

the command line to execute

isContainerActive

protected boolean isContainerActive(org.apache.hadoop.yarn.api.records.ContainerId containerId)

Return whether the container is still active.

Parameters:

containerId - the target container's ID

Returns:

true if the container is active

getNMEnvVar

@VisibleForTesting
protected java.lang.String getNMEnvVar(java.lang.String varname)

activateContainer

public void activateContainer(org.apache.hadoop.yarn.api.records.ContainerId containerId,
                              org.apache.hadoop.fs.Path pidFilePath)

Mark the container as active.

Parameters:

containerId - the container ID

pidFilePath - the path where the executor should write the PID of the launched process

getIpAndHost

public java.lang.String[] getIpAndHost(Container container)
                                throws ContainerExecutionException

Throws:

ContainerExecutionException

getLocalIpAndHost

public static java.lang.String[] getLocalIpAndHost(Container container)

deactivateContainer

public void deactivateContainer(org.apache.hadoop.yarn.api.records.ContainerId containerId)

Mark the container as inactive. For inactive containers this method has no effect.

Parameters:

containerId - the container ID

pauseContainer

public void pauseContainer(Container container)

Pause the container. The default implementation is to raise a kill event. Specific executor implementations can override this behavior.

Parameters:

container - the Container

resumeContainer

public void resumeContainer(Container container)

Resume the container from pause state. The default implementation ignores this event. Specific implementations can override this behavior.

Parameters:

container - the Container

cleanupBeforeRelaunch

public void cleanupBeforeRelaunch(Container container)
                           throws java.io.IOException,
                                  java.lang.InterruptedException

Perform any cleanup before the next launch of the container.

Parameters:

container - container

Throws:

java.io.IOException

java.lang.InterruptedException

getProcessId

public java.lang.String getProcessId(org.apache.hadoop.yarn.api.records.ContainerId containerID)

Get the process-identifier for the container.

Parameters:

containerID - the container ID

Returns:

the process ID of the container if it has already launched, or null otherwise

getExposedPorts

public java.lang.String getExposedPorts(Container container)
                                 throws ContainerExecutionException

Throws:

ContainerExecutionException