make_cluster: Make Mirai Cluster

In mirai: Minimalist Async Evaluation Framework for R

Make Mirai Cluster

Description

make_cluster creates a cluster of type 'miraiCluster', which may be used as a cluster object for any function in the parallel base package such as parallel::clusterApply() or parallel::parLapply().

stop_cluster stops a cluster created by make_cluster.

Usage

make_cluster(n, url = NULL, remote = NULL, ...)

stop_cluster(cl)

Arguments

n	integer number of nodes (automatically launched on the local machine unless url is supplied).
url	[default NULL] (specify for remote nodes) the character URL on the host for remote nodes to dial into, including a port accepting incoming connections, e.g. 'tcp://10.75.37.40:5555'. Specify a URL with the scheme 'tls+tcp://' to use secure TLS connections.
remote	[default NULL] (specify to launch remote nodes) a remote launch configuration generated by remote_config() or ssh_config(). If not supplied, nodes may be deployed manually on remote resources.
...	additional arguments passed onto daemons().
cl	a 'miraiCluster'.

Details

For R version 4.5 or newer, parallel::makeCluster() specifying type = "MIRAI" is equivalent to this function.

Value

For make_cluster: An object of class 'miraiCluster' and 'cluster'. Each 'miraiCluster' has an automatically assigned ID and n nodes of class 'miraiNode'. If url is supplied but not remote, the shell commands for deployment of nodes on remote resources are printed to the console.

For stop_cluster: invisible NULL.

Remote Nodes

Specify url and n to set up a host connection for remote nodes to dial into. n defaults to one if not specified.

Also specify remote to launch the nodes using a configuration generated by remote_config() or ssh_config(). In this case, the number of nodes is inferred from the configuration provided and n is disregarded.

If remote is not supplied, the shell commands for deploying nodes manually on remote resources are automatically printed to the console.

launch_remote() may be called at any time on a 'miraiCluster' to return the shell commands for deployment of all nodes, or on a 'miraiNode' to return the command for a single node.

Status

Call status() on a 'miraiCluster' to check the number of currently active connections as well as the host URL.

Errors

Errors are thrown by the parallel package mechanism if one or more nodes failed (quit unexpectedly). The resulting 'errorValue' returned is 19 (Connection reset). Other types of error, e.g. in evaluation, result in the usual 'miraiError' being returned.

Note

The default behaviour of clusters created by this function is designed to map as closely as possible to clusters created by the parallel package. However, ... arguments are passed onto daemons() for additional customisation if desired, although resultant behaviour may not always be supported.

Examples

cl <- make_cluster(2)
cl
cl[[1L]]

Sys.sleep(0.5)
status(cl)

stop_cluster(cl)

mirai documentation built on April 4, 2025, 12:27 a.m.