libmuscle.native_instantiator.agent_manager module

class libmuscle.native_instantiator.agent_manager.AgentManager(agent_dir: Path)

Bases: IAgentManager

Manage the node agents.

Each node of our allocated resources gets an agent, which launches and monitors processes or that node. This class launches those agents across the nodes, and communicates with them.

The AgentManager sits in between the NativeInstantiator and the MAPServer. It gets called by NativeInstantiator with requests for resources and commands to start and cancel processes on nodes, and it gets called by MAPServer with requests from the agents.

cancel_all() → None

Cancel all processes.

This tells the agents to stop all running processes they’ve started.

Called by NativeInstantiator.

get_finished() → list[tuple[str, int]]

Returns names and exit codes of finished processes.

This returns all processes that have finished running since the previous call; each started process will be returned exactly once. The names are the ones passed to start().

Called by NativeInstantiator.

get_resources() → Resources

Return detected resources.

This returns a list of sets of logical hwthread ids per core, per node.

Called by NativeInstantiator.

report_resources(resources: OnNodeResources) → None

Report resources found on a node.

Called by MAPServer from a server thread.

Parameters:

resources – Description of a node’s resources

report_result(names_exit_codes: list[tuple[str, int]]) → None

Report results of finished processes.

Called by MAPServer from a server thread.

Parameters:

names_exit_codes – A list of names and exit codes of finished processes.

shutdown() → None

Shut down the manager and its agents.

start(node_name: str, name: str, work_dir: Path, args: list[str], env: dict[str, str], stdout: Path, stderr: Path) → None

Start a process on a node.

The files that the output is directed to will be overwritten if they already exist.

Parameters:

• node_name – Name of the node to run the process on

• name – Name under which this process will be known

• work_dir – Working directory in which to start

• args – Executable and arguments to run

• env – Environment variables to set

• stdout – File to redirect stdout to

• stderr – File to redirect stderr to