R6
class for controller groups.
See also
Other controller_group:
crew_controller_group()
Active bindings
controllers
List of
R6
controller objects.relay
Relay object for event-driven programming on a downstream condition variable.
Methods
Method new()
Multi-controller constructor.
Usage
crew_class_controller_group$new(controllers = NULL, relay = NULL)
Arguments
controllers
List of
R6
controller objects.relay
Relay object for event-driven programming on a downstream condition variable.
Examples
if (identical(Sys.getenv("CREW_EXAMPLES"), "true")) {
persistent <- crew_controller_local(name = "persistent")
transient <- crew_controller_local(
name = "transient",
tasks_max = 1L
)
group <- crew_controller_group(persistent, transient)
group$start()
group$push(name = "task", command = sqrt(4), controller = "transient")
group$wait()
group$pop()
group$terminate()
}
Method empty()
See if the controllers are empty.
Method nonempty()
Check if the controller group is nonempty.
Method resolved()
Number of resolved mirai()
tasks.
Method saturated()
Check if a controller is saturated.
Arguments
collect
Deprecated in version 0.5.0.9003 (2023-10-02). Not used.
throttle
Deprecated in version 0.5.0.9003 (2023-10-02). Not used.
controller
Character vector of length 1 with the controller name. Set to
NULL
to select the default controller thatpush()
would choose.
Method start()
Start one or more controllers.
Method launch()
Launch one or more workers on one or more controllers.
Method scale()
Automatically scale up the number of workers if needed in one or more controller objects.
Arguments
throttle
TRUE
to skip auto-scaling if it already happened within the last polling interval.FALSE
to auto-scale every timescale()
is called. Throttling avoids overburdening themirai
dispatcher and other resources.controllers
Character vector of controller names. Set to
NULL
to select all controllers.
Details
See the scale()
method in individual controller classes.
Method autoscale()
Run worker auto-scaling in a later
loop.
Usage
crew_class_controller_group$autoscale(
loop = later::current_loop(),
controllers = NULL
)
Method crashes()
Report the number of consecutive crashes of a task, summed over all selected controllers in the group.
Arguments
name
Character string, name of the task to check.
controllers
Not used. Included to ensure the signature is compatible with the analogous method of controller groups.
Details
See the crashes_max
argument of crew_controller()
.
Method push()
Push a task to the head of the task list.
Arguments
command
Language object with R code to run.
data
Named list of local data objects in the evaluation environment.
globals
Named list of objects to temporarily assign to the global environment for the task. See the
reset_globals
argument ofcrew_controller_local()
.substitute
Logical of length 1, whether to call
base::substitute()
on the supplied value of thecommand
argument. IfTRUE
(default) thencommand
is quoted literally as you write it, e.g.push(command = your_function_call())
. IfFALSE
, thencrew
assumescommand
is a language object and you are passing its value, e.g.push(command = quote(your_function_call()))
.substitute = TRUE
is appropriate for interactive use, whereassubstitute = FALSE
is meant for automated R programs that invokecrew
controllers.seed
Integer of length 1 with the pseudo-random number generator seed to set for the evaluation of the task. Passed to the
seed
argument ofset.seed()
if notNULL
. Ifalgorithm
andseed
are bothNULL
, then the random number generator defaults to the widely spaced worker-specific L'Ecuyer streams as supported bymirai::nextstream()
. Seevignette("parallel", package = "parallel")
for details.algorithm
Integer of length 1 with the pseudo-random number generator algorithm to set for the evaluation of the task. Passed to the
kind
argument ofRNGkind()
if notNULL
. Ifalgorithm
andseed
are bothNULL
, then the random number generator defaults to the recommended widely spaced worker-specific L'Ecuyer streams as supported bymirai::nextstream()
. Seevignette("parallel", package = "parallel")
for details.packages
Character vector of packages to load for the task.
library
Library path to load the packages. See the
lib.loc
argument ofrequire()
.seconds_timeout
Optional task timeout passed to the
.timeout
argument ofmirai::mirai()
(after converting to milliseconds).scale
Logical, whether to automatically scale workers to meet demand. See the
scale
argument of thepush()
method of ordinary single controllers.throttle
TRUE
to skip auto-scaling if it already happened within the last polling interval.FALSE
to auto-scale every timescale()
is called. Throttling avoids overburdening themirai
dispatcher and other resources.name
Character string, name of the task. If
NULL
, a random name is automatically generated. The task name must not conflict with an existing task in the controller where it is submitted. To reuse the name, wait for the existing task to finish, then eitherpop()
orcollect()
it to remove it from its controller.save_command
Deprecated on 2025-01-22 (
crew
version 0.10.2.9004).controller
Character of length 1, name of the controller to submit the task. If
NULL
, the controller defaults to the first controller in the list.
Returns
Invisibly return the mirai
object of the pushed task.
This allows you to interact with the task directly, e.g.
to create a promise object with promises::as.promise()
.
Method walk()
Apply a single command to multiple inputs, and return control to the user without waiting for any task to complete.
Usage
crew_class_controller_group$walk(
command,
iterate,
data = list(),
globals = list(),
substitute = TRUE,
seed = NULL,
algorithm = NULL,
packages = character(0),
library = NULL,
seconds_timeout = NULL,
names = NULL,
save_command = NULL,
verbose = interactive(),
scale = TRUE,
throttle = TRUE,
controller = NULL
)
Arguments
command
Language object with R code to run.
iterate
Named list of vectors or lists to iterate over. For example, to run function calls
f(x = 1, y = "a")
andf(x = 2, y = "b")
, setcommand
tof(x, y)
, and setiterate
tolist(x = c(1, 2), y = c("a", "b"))
. The individual function calls are evaluated asf(x = iterate$x[[1]], y = iterate$y[[1]])
andf(x = iterate$x[[2]], y = iterate$y[[2]])
. All the elements ofiterate
must have the same length. If there are any name conflicts betweeniterate
anddata
,iterate
takes precedence.data
Named list of constant local data objects in the evaluation environment. Objects in this list are treated as single values and are held constant for each iteration of the map.
globals
Named list of constant objects to temporarily assign to the global environment for each task. This list should include any functions you previously defined in the global environment which are required to run tasks. See the
reset_globals
argument ofcrew_controller_local()
. Objects in this list are treated as single values and are held constant for each iteration of the map.substitute
Logical of length 1, whether to call
base::substitute()
on the supplied value of thecommand
argument. IfTRUE
(default) thencommand
is quoted literally as you write it, e.g.push(command = your_function_call())
. IfFALSE
, thencrew
assumescommand
is a language object and you are passing its value, e.g.push(command = quote(your_function_call()))
.substitute = TRUE
is appropriate for interactive use, whereassubstitute = FALSE
is meant for automated R programs that invokecrew
controllers.seed
Integer of length 1 with the pseudo-random number generator seed to set for the evaluation of the task. Passed to the
seed
argument ofset.seed()
if notNULL
. Ifalgorithm
andseed
are bothNULL
, then the random number generator defaults to the recommended widely spaced worker-specific L'Ecuyer streams as supported bymirai::nextstream()
. Seevignette("parallel", package = "parallel")
for details.algorithm
Integer of length 1 with the pseudo-random number generator algorithm to set for the evaluation of the task. Passed to the
kind
argument ofRNGkind()
if notNULL
. Ifalgorithm
andseed
are bothNULL
, then the random number generator defaults to the recommended widely spaced worker-specific L'Ecuyer streams as supported bymirai::nextstream()
. Seevignette("parallel", package = "parallel")
for details.packages
Character vector of packages to load for the task.
library
Library path to load the packages. See the
lib.loc
argument ofrequire()
.seconds_timeout
Optional task timeout passed to the
.timeout
argument ofmirai::mirai()
(after converting to milliseconds).names
Optional character of length 1, name of the element of
iterate
with names for the tasks. Ifnames
is supplied, theniterate[[names]]
must be a character vector.save_command
Deprecated on 2025-01-22 (
crew
version 0.10.2.9004).verbose
Logical of length 1, whether to print to a progress bar when pushing tasks.
scale
Logical, whether to automatically scale workers to meet demand. See also the
throttle
argument.throttle
TRUE
to skip auto-scaling if it already happened within the last polling interval.FALSE
to auto-scale every timescale()
is called. Throttling avoids overburdening themirai
dispatcher and other resources.controller
Character of length 1, name of the controller to submit the tasks. If
NULL
, the controller defaults to the first controller in the list.
Method map()
Apply a single command to multiple inputs.
Usage
crew_class_controller_group$map(
command,
iterate,
data = list(),
globals = list(),
substitute = TRUE,
seed = NULL,
algorithm = NULL,
packages = character(0),
library = NULL,
seconds_interval = NULL,
seconds_timeout = NULL,
names = NULL,
save_command = NULL,
error = "stop",
warnings = TRUE,
verbose = interactive(),
scale = TRUE,
throttle = TRUE,
controller = NULL
)
Arguments
command
Language object with R code to run.
iterate
Named list of vectors or lists to iterate over. For example, to run function calls
f(x = 1, y = "a")
andf(x = 2, y = "b")
, setcommand
tof(x, y)
, and setiterate
tolist(x = c(1, 2), y = c("a", "b"))
. The individual function calls are evaluated asf(x = iterate$x[[1]], y = iterate$y[[1]])
andf(x = iterate$x[[2]], y = iterate$y[[2]])
. All the elements ofiterate
must have the same length. If there are any name conflicts betweeniterate
anddata
,iterate
takes precedence.data
Named list of constant local data objects in the evaluation environment. Objects in this list are treated as single values and are held constant for each iteration of the map.
globals
Named list of constant objects to temporarily assign to the global environment for each task. This list should include any functions you previously defined in the global environment which are required to run tasks. See the
reset_globals
argument ofcrew_controller_local()
. Objects in this list are treated as single values and are held constant for each iteration of the map.substitute
Logical of length 1, whether to call
base::substitute()
on the supplied value of thecommand
argument. IfTRUE
(default) thencommand
is quoted literally as you write it, e.g.push(command = your_function_call())
. IfFALSE
, thencrew
assumescommand
is a language object and you are passing its value, e.g.push(command = quote(your_function_call()))
.substitute = TRUE
is appropriate for interactive use, whereassubstitute = FALSE
is meant for automated R programs that invokecrew
controllers.seed
Integer of length 1 with the pseudo-random number generator seed to set for the evaluation of the task. Passed to the
seed
argument ofset.seed()
if notNULL
. Ifalgorithm
andseed
are bothNULL
, then the random number generator defaults to the recommended widely spaced worker-specific L'Ecuyer streams as supported bymirai::nextstream()
. Seevignette("parallel", package = "parallel")
for details.algorithm
Integer of length 1 with the pseudo-random number generator algorithm to set for the evaluation of the task. Passed to the
kind
argument ofRNGkind()
if notNULL
. Ifalgorithm
andseed
are bothNULL
, then the random number generator defaults to the recommended widely spaced worker-specific L'Ecuyer streams as supported bymirai::nextstream()
. Seevignette("parallel", package = "parallel")
for details.packages
Character vector of packages to load for the task.
library
Library path to load the packages. See the
lib.loc
argument ofrequire()
.seconds_interval
Deprecated on 2025-01-17 (
crew
version 0.10.2.9003). Instead, theseconds_interval
argument passed tocrew_controller_group()
is used asseconds_max
in acrew_throttle()
object which orchestrates exponential backoff.seconds_timeout
Optional task timeout passed to the
.timeout
argument ofmirai::mirai()
(after converting to milliseconds).names
Optional character of length 1, name of the element of
iterate
with names for the tasks. Ifnames
is supplied, theniterate[[names]]
must be a character vector.save_command
Deprecated on 2025-01-22 (
crew
version 0.10.2.9004).error
Character vector of length 1, choice of action if a task has an error. Possible values:
"stop"
: throw an error in the main R session instead of returning a value. In case of an error, the results from the last erroredmap()
are in theerror
field of the controller, e.g.controller_object$error
. To reduce memory consumption, setcontroller_object$error <- NULL
after you are finished troubleshooting."warn"
: throw a warning. This allows the return value with all the error messages and tracebacks to be generated."silent"
: do nothing special.
warnings
Logical of length 1, whether to throw a warning in the interactive session if at least one task encounters an error.
verbose
Logical of length 1, whether to print progress messages.
scale
Logical, whether to automatically scale workers to meet demand. See also the
throttle
argument.throttle
TRUE
to skip auto-scaling if it already happened within the last polling interval.FALSE
to auto-scale every timescale()
is called. Throttling avoids overburdening themirai
dispatcher and other resources.controller
Character of length 1, name of the controller to submit the tasks. If
NULL
, the controller defaults to the first controller in the list.
Method pop()
Pop a completed task from the results data frame.
Usage
crew_class_controller_group$pop(
scale = TRUE,
collect = NULL,
throttle = TRUE,
error = NULL,
controllers = NULL
)
Arguments
scale
Logical, whether to automatically scale workers to meet demand. See the
scale
argument of thepop()
method of ordinary single controllers.collect
Deprecated in version 0.5.0.9003 (2023-10-02). Not used.
throttle
TRUE
to skip auto-scaling if it already happened within the last polling interval.FALSE
to auto-scale every timescale()
is called. Throttling avoids overburdening themirai
dispatcher and other resources.error
NULL
or character of length 1, choice of action if the popped task threw an error. Possible values:"stop"
: throw an error in the main R session instead of returning a value."warn"
: throw a warning.NULL
or"silent"
: do not react to errors.
controllers
Character vector of controller names. Set to
NULL
to select all controllers.
Method collect()
Pop all available task results and return them in a tidy
tibble
.
Usage
crew_class_controller_group$collect(
scale = TRUE,
throttle = TRUE,
error = NULL,
controllers = NULL
)
Arguments
scale
Logical of length 1, whether to automatically call
scale()
to auto-scale workers to meet the demand of the task load.throttle
TRUE
to skip auto-scaling if it already happened within the last polling interval.FALSE
to auto-scale every timescale()
is called. Throttling avoids overburdening themirai
dispatcher and other resources.error
NULL
or character of length 1, choice of action if the popped task threw an error. Possible values:"stop"
: throw an error in the main R session instead of returning a value."warn"
: throw a warning.NULL
or"silent"
: do not react to errors.
controllers
Character vector of controller names. Set to
NULL
to select all controllers.
Method wait()
Wait for tasks.
Usage
crew_class_controller_group$wait(
mode = "all",
seconds_interval = NULL,
seconds_timeout = Inf,
scale = TRUE,
throttle = TRUE,
controllers = NULL
)
Arguments
mode
Character string, name of the waiting condition.
wait(mode = "all")
waits until all tasks in themirai
compute profile resolve, andwait(mode = "one")
waits until at least one task is available topush()
orcollect()
from the controller. The former still works if the controller is not the only means of submitting tasks to the compute profile, whereas the latter assumes only the controller submits tasks.seconds_interval
Deprecated on 2025-01-17 (
crew
version 0.10.2.9003). Instead, theseconds_interval
argument passed tocrew_controller_group()
is used asseconds_max
in acrew_throttle()
object which orchestrates exponential backoff.seconds_timeout
Timeout length in seconds waiting for results to become available.
scale
Logical of length 1, whether to call
scale_later()
on each selected controller to schedule auto-scaling. See thescale
argument of thewait()
method of ordinary single controllers.throttle
TRUE
to skip auto-scaling if it already happened within the last polling interval.FALSE
to auto-scale every timescale()
is called. Throttling avoids overburdening themirai
dispatcher and other resources.controllers
Character vector of controller names. Set to
NULL
to select all controllers.
Details
The wait()
method blocks the calling R session
until the condition in the mode
argument is met.
During the wait, wait()
iteratively auto-scales the workers.
Returns
A logical of length 1, invisibly.
wait(mode = "all")
returns TRUE
if all tasks in the mirai
compute profile have resolved (FALSE
otherwise).
wait(mode = "one")
returns TRUE
if the controller is ready
to pop or collect at least one resolved task (FALSE
otherwise).
wait(mode = "one")
assumes all
tasks were submitted through the controller and not by other means.
Method push_backlog()
Push the name of a task to the backlog.
Method pop_backlog()
Pop the task names from the head of the backlog which can be pushed without saturating the controller.
Method summary()
Summarize the workers of one or more controllers.
Returns
A data frame of aggregated worker summary statistics
of all the selected controllers. It has one row per worker,
and the rows are grouped by controller.
See the documentation of the summary()
method of the controller
class for specific information about the columns in the output.
Examples
if (identical(Sys.getenv("CREW_EXAMPLES"), "true")) {
persistent <- crew_controller_local(name = "persistent")
transient <- crew_controller_local(
name = "transient",
tasks_max = 1L
)
group <- crew_controller_group(persistent, transient)
group$start()
group$push(name = "task", command = sqrt(4), controller = "transient")
group$wait()
group$pop()
group$terminate()
}
## ------------------------------------------------
## Method `crew_class_controller_group$new`
## ------------------------------------------------
if (identical(Sys.getenv("CREW_EXAMPLES"), "true")) {
persistent <- crew_controller_local(name = "persistent")
transient <- crew_controller_local(
name = "transient",
tasks_max = 1L
)
group <- crew_controller_group(persistent, transient)
group$start()
group$push(name = "task", command = sqrt(4), controller = "transient")
group$wait()
group$pop()
group$terminate()
}