ClusterShell

Transport ClusterShell: worker and event handlers.

class cumin.transports.clustershell.AsyncEventHandler(target, commands, success_threshold=1.0, **kwargs)[source]

Bases: cumin.transports.clustershell.BaseEventHandler

Custom ClusterShell event handler class that execute commands asynchronously.

The implemented logic is:

  • execute on all nodes independently every command in a sequence, aborting the execution on that node if any command fails.
  • The success ratio is checked at each node completion (either because it completed all commands or aborted earlier), however nodes already scheduled for execution with ClusterShell will execute the commands anyway. The use of the batch_size allows to control this aspect.
  • if the success ratio is met, schedule the execution of all commands to the next node.

The typical use case is to execute read-only commands to gather the status of a fleet without any special need of orchestration between the nodes.

__init__(target, commands, success_threshold=1.0, **kwargs)[source]

Define a custom ClusterShell event handler to execute commands asynchronously between nodes.

Parameters:according to parent BaseEventHandler.__init__().
_commands_output_report(buffer_iterator, command=None)

inherited Print the commands output in a colored and tqdm-friendly way.

Parameters:
  • buffer_iterator (mixed) -- any ClusterShell object that implements iter_buffers() like ClusterShell.Task.Task and all the Worker objects.
  • command (str, optional) -- the command the output is referring to.
_failed_commands_report(filter_command_index=-1)

inherited Print the nodes that failed to execute commands in a colored and tqdm-friendly way.

Parameters:filter_command_index (int, optional) -- print only the nodes that failed to execute the command specified by this command index.
_get_log_message(num, message, nodes=None)

inherited Get a pre-formatted message suitable for logging or printing.

Parameters:
  • num (int) -- the number of affecte nodes.
  • message (str) -- the message to print.
  • nodes (list, optional) -- the list of nodes affected.
Returns:

a tuple of (logging message, NodeSet of the affected nodes).

Return type:

tuple

_get_short_command(command)

inherited Return a shortened representation of a command omitting the central part, if it's too long.

Parameters:command (str) -- the command to be shortened.
Returns:the short command.
Return type:str
_global_timeout_nodes_report()

inherited Print the nodes that were caught by the global timeout in a colored and tqdm-friendly way.

_print_report_line(message, color='\x1b[31m', nodes_string='')

inherited Print a tqdm-friendly colored status line with success/failure ratio and optional list of nodes.

Parameters:
  • message (str) -- the message to print.
  • color (str, optional) -- the message color.
  • nodes_string (str, optional) -- the string representation of the affected nodes.
_success_nodes_report(command=None)

inherited Print how many nodes succesfully executed all commands in a colored and tqdm-friendly way.

Parameters:command (str, optional) -- the command the report is referring to.
close(task)[source]

Concrete implementation of parent abstract method to print the nodes reports and close progress bars.

Parameters:according to parent cumin.transports.BaseEventHandler.close().
ev_hup(worker)[source]

Command execution completed on a node.

This callback is triggered by ClusterShell for each node when it completes the execution of a command. Enqueue the next command if the success criteria are met, track the failure otherwise. Update the progress bars accordingly.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_hup().
ev_pickup(worker)

inherited Command execution started on a node, remove the command from the node's queue.

This callback is triggered by the ClusterShell library for each node when it starts executing a command.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_pickup().
ev_read(worker)

inherited Worker has data to read from a specific node. Print it if running on a single host.

This callback is triggered by ClusterShell for each node when output is available.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_read().
ev_timeout(worker)

inherited Worker has timed out.

This callback is triggered by ClusterShell when the execution has timed out.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_timeout().
ev_timer(timer)[source]

Schedule the current command on the next node or the next command on the first batch of nodes.

This callback is triggered by ClusterShell when a scheduled Task.timer() goes off.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_timer().
on_timeout(task)

inherited Update the state of the nodes and the timeout counter.

Callback called by the ClusterShellWorker when a ClusterShell.Task.TimeoutError is raised. It means that the whole execution timed out.

Parameters:task (ClusterShell.Task.Task) -- a ClusterShell Task instance.
class cumin.transports.clustershell.BaseEventHandler(target, commands, success_threshold=1.0, **kwargs)[source]

Bases: ClusterShell.Event.EventHandler

ClusterShell event handler base class.

Inherit from ClusterShell.Event.EventHandler class and define a base EventHandler class to be used in Cumin. It can be subclassed to generate custom EventHandler classes while taking advantage of some common functionalities.

__init__(target, commands, success_threshold=1.0, **kwargs)[source]

Event handler ClusterShell extension constructor.

If subclasses defines a self.pbar_ko tqdm progress bar, it will be updated on timeout.

Parameters:
  • target (cumin.transports.Target) -- a Target instance.
  • commands (list) -- the list of Command objects that has to be executed on the nodes.
  • success_threshold (float, optional) -- the success threshold, a float between 0 and 1, to consider the execution successful.
  • **kwargs (optional) -- additional keyword arguments that might be used by derived classes.
_commands_output_report(buffer_iterator, command=None)[source]

Print the commands output in a colored and tqdm-friendly way.

Parameters:
  • buffer_iterator (mixed) -- any ClusterShell object that implements iter_buffers() like ClusterShell.Task.Task and all the Worker objects.
  • command (str, optional) -- the command the output is referring to.
_failed_commands_report(filter_command_index=-1)[source]

Print the nodes that failed to execute commands in a colored and tqdm-friendly way.

Parameters:filter_command_index (int, optional) -- print only the nodes that failed to execute the command specified by this command index.
_get_log_message(num, message, nodes=None)[source]

Get a pre-formatted message suitable for logging or printing.

Parameters:
  • num (int) -- the number of affecte nodes.
  • message (str) -- the message to print.
  • nodes (list, optional) -- the list of nodes affected.
Returns:

a tuple of (logging message, NodeSet of the affected nodes).

Return type:

tuple

_get_short_command(command)[source]

Return a shortened representation of a command omitting the central part, if it's too long.

Parameters:command (str) -- the command to be shortened.
Returns:the short command.
Return type:str
_global_timeout_nodes_report()[source]

Print the nodes that were caught by the global timeout in a colored and tqdm-friendly way.

_print_report_line(message, color='\x1b[31m', nodes_string='')[source]

Print a tqdm-friendly colored status line with success/failure ratio and optional list of nodes.

Parameters:
  • message (str) -- the message to print.
  • color (str, optional) -- the message color.
  • nodes_string (str, optional) -- the string representation of the affected nodes.
_success_nodes_report(command=None)[source]

Print how many nodes succesfully executed all commands in a colored and tqdm-friendly way.

Parameters:command (str, optional) -- the command the report is referring to.
close(task)[source]

Additional method called at the end of the whole execution, useful for reporting and final actions.

Parameters:task (ClusterShell.Task.Task) -- a ClusterShell Task instance.
ev_pickup(worker)[source]

Command execution started on a node, remove the command from the node's queue.

This callback is triggered by the ClusterShell library for each node when it starts executing a command.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_pickup().
ev_read(worker)[source]

Worker has data to read from a specific node. Print it if running on a single host.

This callback is triggered by ClusterShell for each node when output is available.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_read().
ev_timeout(worker)[source]

Worker has timed out.

This callback is triggered by ClusterShell when the execution has timed out.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_timeout().
on_timeout(task)[source]

Update the state of the nodes and the timeout counter.

Callback called by the ClusterShellWorker when a ClusterShell.Task.TimeoutError is raised. It means that the whole execution timed out.

Parameters:task (ClusterShell.Task.Task) -- a ClusterShell Task instance.
short_command_length = 35

int -- the length to which a command should be shortened in various outputs.

class cumin.transports.clustershell.ClusterShellWorker(config, target)[source]

Bases: cumin.transports.BaseWorker

It provides a Cumin worker for SSH using the ClusterShell library.

This transport uses the ClusterShell Python library to connect to the selected hosts and execute a list of commands. This transport accept the following customizations:

  • sync execution mode: given a list of commands, the first one will be executed on all the hosts, then, if the success ratio is reached, the second one will be executed on all hosts where the first one was successful, and so on.
  • async execution mode: given a list of commands, on each hosts the commands will be executed sequentially, interrupting the execution on any single host at the first command that fails. The execution on the hosts is independent between each other.
  • custom execution mode: can be achieved creating a custom event handler class that extends the BaseEventHandler class defined in cumin/transports/clustershell.py, implementing its abstract methods and setting to this class object the handler to the transport.
__init__(config, target)[source]

Worker ClusterShell constructor.

Parameters:according to parent cumin.transports.BaseWorker.__init__().
commands

inherited Commands for the current execution.

Getter:Returns the current command list or an empty list if not set.
Setter:list[Command], list[str]: a list of Command objects or str to be executed in the hosts. The elements are converted to Command automatically.
Raises:cumin.transports.WorkerError -- if trying to set it with invalid data.
execute()[source]

Execute the commands on all the targets using the handler.

Concrete implementation of parent abstract method.

Parameters:according to parent cumin.transports.BaseWorker.execute().
get_results()[source]

Get the results of the last task execution.

Concrete implementation of parent abstract method.

Parameters:according to parent cumin.transports.BaseWorker.get_results().
handler

Concrete implementation of parent abstract getter and setter.

Accepted values for the setter: * an instance of a custom handler class derived from BaseEventHandler. * a str with one of the available default handler listed in DEFAULT_HANDLERS.

The event handler is mandatory for this transport.

Parameters:according to parent cumin.transports.BaseWorker.handler.
success_threshold

inherited Success threshold for the current execution.

Getter:float: returns the current success_threshold or 1.0 (100%) if not set.
Setter:float, None: The success ratio threshold that must be reached to consider the run successful. A float between 0 and 1 or None to reset it. The specific meaning might change based on the chosen transport.
Raises:cumin.transports.WorkerError -- if trying to set it to an invalid value.
timeout

inherited Global timeout for the current execution.

Getter:int: returns the current timeout or 0 (no timeout) if not set.
Setter:int, None: timeout for the current execution in seconds. Must be a positive integer or None to reset it.
Raises:cumin.transports.WorkerError -- if trying to set it to an invalid value.
cumin.transports.clustershell.DEFAULT_HANDLERS = {'async': <class 'cumin.transports.clustershell.AsyncEventHandler'>, 'sync': <class 'cumin.transports.clustershell.SyncEventHandler'>}

dict -- mapping of available default event handlers for ClusterShellWorker.

class cumin.transports.clustershell.Node(name, commands)[source]

Bases: object

Node class to represent each target node.

__init__(name, commands)[source]

Node class constructor with default values.

Parameters:
class cumin.transports.clustershell.SyncEventHandler(target, commands, success_threshold=1.0, **kwargs)[source]

Bases: cumin.transports.clustershell.BaseEventHandler

Custom ClusterShell event handler class that execute commands synchronously.

The implemented logic is:

  • execute command #N on all nodes where command #`N-1` was successful according to batch_size.
  • the success ratio is checked at each command completion on every node, and will abort if not met, however nodes already scheduled for execution with ClusterShell will execute the command anyway. The use of the batch_size allow to control this aspect.
  • if the execution of command #N is completed and the success ratio is greater than the success threshold, re-start from the top with N=N+1.

The typical use case is to orchestrate some operation across a fleet, ensuring that each command is completed by enough nodes before proceeding with the next one.

__init__(target, commands, success_threshold=1.0, **kwargs)[source]

Define a custom ClusterShell event handler to execute commands synchronously.

Parameters:according to parent BaseEventHandler.__init__().
_commands_output_report(buffer_iterator, command=None)

inherited Print the commands output in a colored and tqdm-friendly way.

Parameters:
  • buffer_iterator (mixed) -- any ClusterShell object that implements iter_buffers() like ClusterShell.Task.Task and all the Worker objects.
  • command (str, optional) -- the command the output is referring to.
_failed_commands_report(filter_command_index=-1)

inherited Print the nodes that failed to execute commands in a colored and tqdm-friendly way.

Parameters:filter_command_index (int, optional) -- print only the nodes that failed to execute the command specified by this command index.
_get_log_message(num, message, nodes=None)

inherited Get a pre-formatted message suitable for logging or printing.

Parameters:
  • num (int) -- the number of affecte nodes.
  • message (str) -- the message to print.
  • nodes (list, optional) -- the list of nodes affected.
Returns:

a tuple of (logging message, NodeSet of the affected nodes).

Return type:

tuple

_get_short_command(command)

inherited Return a shortened representation of a command omitting the central part, if it's too long.

Parameters:command (str) -- the command to be shortened.
Returns:the short command.
Return type:str
_global_timeout_nodes_report()

inherited Print the nodes that were caught by the global timeout in a colored and tqdm-friendly way.

_print_report_line(message, color='\x1b[31m', nodes_string='')

inherited Print a tqdm-friendly colored status line with success/failure ratio and optional list of nodes.

Parameters:
  • message (str) -- the message to print.
  • color (str, optional) -- the message color.
  • nodes_string (str, optional) -- the string representation of the affected nodes.
_success_nodes_report(command=None)

inherited Print how many nodes succesfully executed all commands in a colored and tqdm-friendly way.

Parameters:command (str, optional) -- the command the report is referring to.
close(task)[source]

Concrete implementation of parent abstract method to print the success nodes report.

Parameters:according to parent cumin.transports.BaseEventHandler.close().
end_command()[source]

Command terminated, print the result and schedule the next command if criteria are met.

Executed at the end of each command inside a lock.

Returns:True if the next command should be scheduled, False otherwise.
Return type:bool
ev_hup(worker)[source]

Command execution completed on a node.

This callback is triggered by ClusterShell for each node when it completes the execution of a command. Update the progress bars and keep track of nodes based on the success/failure of the command's execution. Schedule a timer for further decisions.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_hup().
ev_pickup(worker)

inherited Command execution started on a node, remove the command from the node's queue.

This callback is triggered by the ClusterShell library for each node when it starts executing a command.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_pickup().
ev_read(worker)

inherited Worker has data to read from a specific node. Print it if running on a single host.

This callback is triggered by ClusterShell for each node when output is available.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_read().
ev_timeout(worker)

inherited Worker has timed out.

This callback is triggered by ClusterShell when the execution has timed out.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_timeout().
ev_timer(timer)[source]

Schedule the current command on the next node or the next command on the first batch of nodes.

This callback is triggered by ClusterShell when a scheduled Task.timer() goes off.

Parameters:according to parent ClusterShell.Event.EventHandler.ev_timer().
on_timeout(task)[source]

Override parent class on_timeout method to run end_command.

Parameters:according to parent BaseEventHandler.on_timeout().
start_command(schedule=False)[source]

Initialize progress bars and variables for this command execution.

Executed at the start of each command.

Parameters:schedule (bool, optional) -- whether the next command should be sent to ClusterShell for execution or not.
cumin.transports.clustershell.worker_class

Required by the transport auto-loader in cumin.transport.Transport.new().

alias of ClusterShellWorker