Rush Controller

Rush is the controller in a centralized rush network. The controller starts and stops the workers, pushes tasks to the workers and fetches results.

Value

Object of class R6::R6Class and Rush with controller methods.

Local Workers

A local worker runs on the same machine as the controller. Local workers are spawned with the $start_local_workers() method via the processx` package.

Remote Workers

A remote worker runs on a different machine than the controller. Remote workers are started manually with the $create_worker_script() and $start_remote_workers() methods. Remote workers can be started on any system as long as the system has access to Redis and all required packages are installed. Only a heartbeat process can kill remote workers. The heartbeat process also monitors the remote workers for crashes.

Stopping Workers

Local and remote workers can be terminated with the $stop_workers(type = "terminate") method. The workers evaluate the currently running task and then terminate. The option type = "kill" stops the workers immediately. Killing a local worker is done with the processx package. Remote workers are killed by pushing a kill signal to the heartbeat process. Without a heartbeat process a remote worker cannot be killed (see section heartbeat).

Heartbeat

The heartbeat process periodically signals that a worker is still alive. This is implemented by setting a timeout on the heartbeat key. Furthermore, the heartbeat process can kill the worker.

Data Structure

Tasks are stored in Redis hashes. Hashes are collections of field-value pairs. The key of the hash identifies the task in Redis and rush.

key : xs | ys | xs_extra

The field-value pairs are written by different methods, e.g. $push_tasks() writes xs and $push_results() writes ys. The values of the fields are serialized lists or atomic values e.g. unserializing xs gives list(x1 = 1, x2 = 2) This data structure allows quick converting of a hash into a row and joining multiple hashes into a table.

| key | x1 | x2 | y | timestamp |
| 1.. |  3 |  4 | 7 |  12:04:11 |
| 2.. |  1 |  4 | 5 |  12:04:12 |
| 3.. |  1 |  1 | 2 |  12:04:13 |

When the value of a field is a named list, the field can store the cells of multiple columns of the table. When the value of a field is an atomic value, the field stores a single cell of a column named after the field. The methods $push_tasks() and $push_results() write into multiple hashes. For example, $push_tasks(xss = list(list(x1 = 1, x2 = 2), list(x1 = 2, x2 = 2)) writes xs in two hashes.

Task States

A task can go through four states "queued", "running", "finished" or "failed". Internally, the keys of the tasks are pushed through Redis lists and sets to keep track of their state. Queued tasks are waiting to be evaluated. A worker pops a task from the queue and changes the state to "running" while evaluating the task. When the task is finished, the state is changed to "finished" and the result is written to the data base. If the task fails, the state is changed to "failed"instead of"finished"`.

Queues

Rush uses a shared queue and a queue for each worker. The shared queue is used to push tasks to the workers. The first worker that pops a task from the shared queue evaluates the task. The worker queues are used to push tasks to specific workers.

Fetch Tasks and Results

The $fetch_*() methods retrieve data from the Redis database. A matching method is defined for each task state e.g. $fetch_running_tasks() and $fetch_finished_tasks(). The methods $fetch_new_tasks() and $fetch_finished_tasks() cache the already queried data. The $wait_for_finished_tasks() variant wait until a new result is available.

Error Handling

When evaluating tasks in a distributed system, many things can go wrong. Simple R errors in the worker loop are caught and written to the archive. The task is marked as "failed". If the connection to a worker is lost, it looks like a task is "running" forever. The method $detect_lost_workers() identifies lost workers. Running this method periodically adds a small overhead.

Logging

The worker logs all messages written with the lgr package to the data base. The lgr_thresholds argument defines the logging level for each logger e.g. c(rush = "debug"). Saving log messages adds a small overhead but is useful for debugging. By default, no log messages are stored.

Seed

Setting a seed is important for reproducibility. The tasks can be evaluated with a specific L'Ecuyer-CMRG seed. If an initial seed is passed, the seed is used to generate L'Ecuyer-CMRG seeds for each task. Each task is then evaluated with a separate RNG stream. See parallel::nextRNGStream for more details.

Public fields

network_id: (character(1))
Identifier of the rush network.
config: (redux::redis_config)
Redis configuration options.
connector: (redux::redis_api)
Returns a connection to Redis.
processes: (processx::process)
List of processes started with $start_local_workers().

Active bindings

n_workers: (integer(1))
Number of workers.
n_running_workers: (integer(1))
Number of running workers.
n_terminated_workers: (integer(1))
Number of terminated workers.
n_killed_workers: (integer(1))
Number of killed workers.
n_lost_workers: (integer(1))
Number of lost workers. Run $detect_lost_workers() to update the number of lost workers.
n_pre_workers: (integer(1))
Number of workers that are not yet completely started.
worker_ids: (character())
Ids of workers.
running_worker_ids: (character())
Ids of running workers.
terminated_worker_ids: (character())
Ids of terminated workers.
killed_worker_ids: (character())
Ids of killed workers.
lost_worker_ids: (character())
Ids of lost workers.
pre_worker_ids: (character())
Ids of workers that are not yet completely started.
tasks: (character())
Keys of all tasks.
queued_tasks: (character())
Keys of queued tasks.
running_tasks: (character())
Keys of running tasks.
finished_tasks: (character())
Keys of finished tasks.
failed_tasks: (character())
Keys of failed tasks.
n_queued_tasks: (integer(1))
Number of queued tasks.
n_queued_priority_tasks: (integer(1))
Number of queued priority tasks.
n_running_tasks: (integer(1))
Number of running tasks.
n_finished_tasks: (integer(1))
Number of finished tasks.
n_failed_tasks: (integer(1))
Number of failed tasks.
n_tasks: (integer(1))
Number of all tasks.
worker_info: (data.table::data.table())
Contains information about the workers.
worker_states: (data.table::data.table())
Contains the states of the workers.
all_workers_terminated: (logical(1))
Whether all workers are terminated.
all_workers_lost: (logical(1))
Whether all workers are lost. Runs $detect_lost_workers() to detect lost workers.
priority_info: (data.table::data.table)
Contains the number of tasks in the priority queues.
snapshot_schedule: (character())
Set a snapshot schedule to periodically save the data base on disk. For example, c(60, 1000) saves the data base every 60 seconds if there are at least 1000 changes. Overwrites the redis configuration file. Set to NULL to disable snapshots. For more details see redis.io.
redis_info: (list())
Information about the Redis server.

Methods

Method `new()`

Creates a new instance of this R6 class.

Usage

Rush$new(network_id = NULL, config = NULL, seed = NULL)

Arguments

network_id: (character(1))
Identifier of the rush network. Controller and workers must have the same instance id. Keys in Redis are prefixed with the instance id.
config: (redux::redis_config)
Redis configuration options. If NULL, configuration set by rush_plan() is used. If rush_plan() has not been called, the REDIS_URL environment variable is parsed. If REDIS_URL is not set, a default configuration is used. See redux::redis_config for details.
seed: (integer())
Initial seed for the random number generator. Either a L'Ecuyer-CMRG seed (integer(7)) or a regular RNG seed (integer(1)). The later is converted to a L'Ecuyer-CMRG seed. If NULL, no seed is used for the random number generator.

Method `format()`

Helper for print outputs.

Usage

Rush$format(...)

Arguments

...: (ignored).

Returns

(character()).

Method `print()`

Print method.

Usage

Rush$print()

Returns

(character()).

Method `reconnect()`

Reconnect to Redis. The connection breaks when the Rush object is saved to disk. Call this method to reconnect after loading the object.

Usage

Rush$reconnect()

Method `start_local_workers()`

Start workers locally with processx. The processx::process are stored in $processes. Alternatively, use $create_worker_script() to create a script for starting workers on remote machines. By default, worker_loop_default() is used as worker loop. This function takes the arguments fun and optionally constants which are passed in ....

Usage

Rush$start_local_workers(
  n_workers = NULL,
  wait_for_workers = TRUE,
  timeout = Inf,
  globals = NULL,
  packages = NULL,
  heartbeat_period = NULL,
  heartbeat_expire = NULL,
  lgr_thresholds = NULL,
  lgr_buffer_size = 0,
  supervise = TRUE,
  worker_loop = worker_loop_default,
  ...
)

Arguments

n_workers: (integer(1))
Number of workers to be started.
wait_for_workers: (logical(1))
Whether to wait until all workers are available.
timeout: (numeric(1))
Timeout to wait for workers in seconds.
globals: (character())
Global variables to be loaded to the workers global environment.
packages: (character())
Packages to be loaded by the workers.
heartbeat_period: (integer(1))
Period of the heartbeat in seconds.
heartbeat_expire: (integer(1))
Time to live of the heartbeat in seconds.
lgr_thresholds: (named character() | named numeric())
Logger threshold on the workers e.g. c(rush = "debug").
lgr_buffer_size: (integer(1))
By default (lgr_buffer_size = 0), the log messages are directly saved in the Redis data store. If lgr_buffer_size > 0, the log messages are buffered and saved in the Redis data store when the buffer is full. This improves the performance of the logging.
supervise: (logical(1))
Whether to kill the workers when the main R process is shut down.
worker_loop: (function)
Loop run on the workers. Defaults to worker_loop_default which is called with fun. Pass fun in .... Use worker_loop_callr to run fun in an external callr session.
...: (any)
Arguments passed to worker_loop.

Method `restart_local_workers()`

Restart local workers. If the worker is is still running, it is killed and restarted.

Usage

Rush$restart_local_workers(worker_ids, supervise = TRUE)

Arguments

worker_ids: (character())
Worker ids to be restarted.
supervise: (logical(1))
Whether to kill the workers when the main R process is shut down.

Method `create_worker_script()`

Create script to remote start workers. Run these command to pre-start a worker. The worker will wait until the start arguments are pushed with $start_remote_workers().

Usage

Rush$create_worker_script()

Method `start_remote_workers()`

Push start arguments to remote workers. Remote workers must be pre-started with $create_worker_script().

Usage

Rush$start_remote_workers(
  globals = NULL,
  packages = NULL,
  heartbeat_period = NULL,
  heartbeat_expire = NULL,
  lgr_thresholds = NULL,
  lgr_buffer_size = 0,
  worker_loop = worker_loop_default,
  ...
)

Arguments

globals: (character())
Global variables to be loaded to the workers global environment.
packages: (character())
Packages to be loaded by the workers.
heartbeat_period: (integer(1))
Period of the heartbeat in seconds.
heartbeat_expire: (integer(1))
Time to live of the heartbeat in seconds.
lgr_thresholds: (named character() | named numeric())
Logger threshold on the workers e.g. c(rush = "debug").
lgr_buffer_size: (integer(1))
By default (lgr_buffer_size = 0), the log messages are directly saved in the Redis data store. If lgr_buffer_size > 0, the log messages are buffered and saved in the Redis data store when the buffer is full. This improves the performance of the logging.
worker_loop: (function)
Loop run on the workers. Defaults to worker_loop_default which is called with fun. Pass fun in .... Use worker_loop_callr to run fun in an external callr session.
...: (any)
Arguments passed to worker_loop.

Method `wait_for_workers()`

Wait until n workers are available.

Usage

Rush$wait_for_workers(n, timeout = Inf)

Arguments

n: (integer(1))
Number of workers to wait for.
timeout: (numeric(1))
Timeout in seconds. Default is Inf.

Method `stop_workers()`

Stop workers.

Usage

Rush$stop_workers(type = "terminate", worker_ids = NULL)

Arguments

type: (character(1))
Type of stopping. Either "terminate" or "kill". If "terminate" the workers evaluate the currently running task and then terminate. If "kill" the workers are stopped immediately.
worker_ids: (character())
Worker ids to be stopped. If NULL all workers are stopped.

Method `detect_lost_workers()`

Detect lost workers. The state of the worker is changed to "lost". Local workers without a heartbeat are checked by their process id. Checking local workers on unix systems only takes a few microseconds per worker. But checking local workers on windows might be very slow. Workers with a heartbeat process are checked with the heartbeat. Lost tasks are marked as "lost".

Usage

Rush$detect_lost_workers(restart_local_workers = FALSE)

Arguments

restart_local_workers: (logical(1))
Whether to restart lost workers.

Method `reset()`

Stop workers and delete data stored in redis.

Usage

Rush$reset(type = "kill")

Arguments

type: (character(1))
Type of stopping. Either "terminate" or "kill". If "terminate" the workers evaluate the currently running task and then terminate. If "kill" the workers are stopped immediately.

Method `read_log()`

Read log messages written with the lgr package from a worker.

Usage

Rush$read_log(worker_ids = NULL)

Arguments

worker_ids: (character(1))
Worker ids. If NULL all worker ids are used.

Method `print_log()`

Print log messages written with the lgr package from a worker.

Usage

Rush$print_log()

Method `push_tasks()`

Pushes a task to the queue. Task is added to queued tasks.

Usage

Rush$push_tasks(
  xss,
  extra = NULL,
  seeds = NULL,
  timeouts = NULL,
  max_retries = NULL,
  terminate_workers = FALSE
)

Arguments

xss: (list of named list())
Lists of arguments for the function e.g. list(list(x1, x2), list(x1, x2))).
extra: (list())
List of additional information stored along with the task e.g. list(list(timestamp), list(timestamp))).
seeds: (list())
List of L'Ecuyer-CMRG seeds for each task e.g list(list(c(104071, 490840688, 1690070564, -495119766, 503491950, 1801530932, -1629447803))). If NULL but an initial seed is set, L'Ecuyer-CMRG seeds are generated from the initial seed. If NULL and no initial seed is set, no seeds are used for the random number generator.
timeouts: (integer())
Timeouts for each task in seconds e.g. c(10, 15). A single number is used as the timeout for all tasks. If NULL no timeout is set.
max_retries: (integer())
Number of retries for each task. A single number is used as the number of retries for all tasks. If NULL tasks are not retried.
terminate_workers: (logical(1))
Whether to stop the workers after evaluating the tasks.

Returns

(character())
Keys of the tasks.

Method `push_priority_tasks()`

Pushes a task to the queue of a specific worker. Task is added to queued priority tasks. A worker evaluates the tasks in the priority queue before the shared queue. If priority is NA the task is added to the shared queue. If the worker is lost or worker id is not known, the task is added to the shared queue.

Usage

Rush$push_priority_tasks(xss, extra = NULL, priority = NULL)

Arguments

xss: (list of named list())
Lists of arguments for the function e.g. list(list(x1, x2), list(x1, x2))).
extra: (list)
List of additional information stored along with the task e.g. list(list(timestamp), list(timestamp))).
priority: (character())
Worker ids to which the tasks should be pushed.

Returns

(character())
Keys of the tasks.

Method `push_failed()`

Pushes failed tasks to the data base.

Usage

Rush$push_failed(keys, conditions)

Arguments

keys: (character(1))
Keys of the associated tasks.
conditions: (named list())
List of lists of conditions.

Method `retry_tasks()`

Retry failed tasks.

Usage

Rush$retry_tasks(keys, ignore_max_retries = FALSE, next_seed = FALSE)

Arguments

keys: (character())
Keys of the tasks to be retried.
ignore_max_retries: (logical(1))
Whether to ignore the maximum number of retries.
next_seed: (logical(1))
Whether to change the seed of the task.

Method `fetch_queued_tasks()`

Fetch queued tasks from the data base.

Usage

Rush$fetch_queued_tasks(
  fields = c("xs", "xs_extra"),
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra").
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Table of queued tasks.

Method `fetch_priority_tasks()`

Fetch queued priority tasks from the data base.

Usage

Rush$fetch_priority_tasks(
  fields = c("xs", "xs_extra"),
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra").
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Table of queued priority tasks.

Method `fetch_running_tasks()`

Fetch running tasks from the data base.

Usage

Rush$fetch_running_tasks(
  fields = c("xs", "xs_extra", "worker_extra"),
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra", "worker_extra").
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Table of running tasks.

Method `fetch_finished_tasks()`

Fetch finished tasks from the data base. Finished tasks are cached.

Usage

Rush$fetch_finished_tasks(
  fields = c("xs", "ys", "xs_extra", "worker_extra", "ys_extra", "condition"),
  reset_cache = FALSE,
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra", "worker_extra", "ys", "ys_extra").
reset_cache: (logical(1))
Whether to reset the cache.
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Table of finished tasks.

Method `wait_for_finished_tasks()`

Block process until a new finished task is available. Returns all finished tasks or NULL if no new task is available after timeout seconds.

Usage

Rush$wait_for_finished_tasks(
  fields = c("xs", "ys", "xs_extra", "worker_extra", "ys_extra"),
  timeout = Inf,
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra", "worker_extra", "ys", "ys_extra").
timeout: (numeric(1))
Time to wait for a result in seconds.
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Table of finished tasks.

Method `fetch_new_tasks()`

Fetch finished tasks from the data base that finished after the last fetch. Updates the cache of the finished tasks.

Usage

Rush$fetch_new_tasks(
  fields = c("xs", "ys", "xs_extra", "worker_extra", "ys_extra", "condition"),
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes.
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Latest results.

Method `wait_for_new_tasks()`

Block process until a new finished task is available. Returns new tasks or NULL if no new task is available after timeout seconds.

Usage

Rush$wait_for_new_tasks(
  fields = c("xs", "ys", "xs_extra", "worker_extra", "ys_extra", "condition"),
  timeout = Inf,
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra", "worker_extra", "ys", "ys_extra").
timeout: (numeric(1))
Time to wait for new result in seconds.
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table() | list().

Method `fetch_failed_tasks()`

Fetch failed tasks from the data base.

Usage

Rush$fetch_failed_tasks(
  fields = c("xs", "worker_extra", "condition"),
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra", "worker_extra", "condition".
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Table of failed tasks.

Method `fetch_tasks()`

Fetch all tasks from the data base.

Usage

Rush$fetch_tasks(
  fields = c("xs", "ys", "xs_extra", "worker_extra", "ys_extra", "condition"),
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "xs_extra", "worker_extra", "ys", "ys_extra", "condition", "state").
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Returns

data.table()
Table of all tasks.

Method `fetch_tasks_with_state()`

Fetch tasks with different states from the data base. If tasks with different states are to be queried at the same time, this function prevents tasks from appearing twice. This could be the case if a worker changes the state of a task while the tasks are being fetched. Finished tasks are cached.

Usage

Rush$fetch_tasks_with_state(
  fields = c("xs", "ys", "xs_extra", "worker_extra", "ys_extra", "condition"),
  states = c("queued", "running", "finished", "failed"),
  reset_cache = FALSE,
  data_format = "data.table"
)

Arguments

fields: (character())
Fields to be read from the hashes. Defaults to c("xs", "ys", "xs_extra", "worker_extra", "ys_extra").
states: (character())
States of the tasks to be fetched. Defaults to c("queued", "running", "finished", "failed").
reset_cache: (logical(1))
Whether to reset the cache of the finished tasks.
data_format: (character())
Returned data format. Choose "data.table" or "list". The default is "data.table" but "list" is easier when list columns are present.

Method `wait_for_tasks()`

Wait until tasks are finished. The function also unblocks when no worker is running or all tasks failed.

Usage

Rush$wait_for_tasks(keys, detect_lost_workers = FALSE)

Arguments

keys: (character())
Keys of the tasks to wait for.
detect_lost_workers: (logical(1))
Whether to detect failed tasks. Comes with an overhead.

Method `write_hashes()`

Writes R objects to Redis hashes. The function takes the vectors in ... as input and writes each element as a field-value pair to a new hash. The name of the argument defines the field into which the serialized element is written. For example, xs = list(list(x1 = 1, x2 = 2), list(x1 = 3, x2 = 4)) writes serialize(list(x1 = 1, x2 = 2)) at field xs into a hash and serialize(list(x1 = 3, x2 = 4)) at field xs into another hash. The function can iterate over multiple vectors simultaneously. For example, xs = list(list(x1 = 1, x2 = 2), list(x1 = 3, x2 = 4)), ys = list(list(y = 3), list(y = 7)) creates two hashes with the fields xs and ys. The vectors are recycled to the length of the longest vector. Both lists and atomic vectors are supported. Arguments that are NULL are ignored.

Usage

Rush$write_hashes(..., .values = list(), keys = NULL)

Arguments

...: (named list())
Lists to be written to the hashes. The names of the arguments are used as fields.
.values: (named list())
Lists to be written to the hashes. The names of the list are used as fields.
keys: (character())
Keys of the hashes. If NULL new keys are generated.

Returns

(character())
Keys of the hashes.

Method `read_hashes()`

Reads R Objects from Redis hashes. The function reads the field-value pairs of the hashes stored at keys. The values of a hash are deserialized and combined to a list. If flatten is TRUE, the values are flattened to a single list e.g. list(xs = list(x1 = 1, x2 = 2), ys = list(y = 3)) becomes list(x1 = 1, x2 = 2, y = 3). The reading functions combine the hashes to a table where the names of the inner lists are the column names. For example, xs = list(list(x1 = 1, x2 = 2), list(x1 = 3, x2 = 4)), ys = list(list(y = 3), list(y = 7)) becomes data.table(x1 = c(1, 3), x2 = c(2, 4), y = c(3, 7)).

Usage

Rush$read_hashes(keys, fields, flatten = TRUE)

Arguments

keys: (character())
Keys of the hashes.
fields: (character())
Fields to be read from the hashes.
flatten: (logical(1))
Whether to flatten the list.

Returns

(list of list())
The outer list contains one element for each key. The inner list is the combination of the lists stored at the different fields.

Method `read_hash()`

Reads a single Redis hash and returns the values as a list named by the fields.

Usage

Rush$read_hash(key, fields)

Arguments

key: (character(1))
Key of the hash.
fields: (character())
Fields to be read from the hash.

Returns

(list of list())
The outer list contains one element for each key. The inner list is the combination of the lists stored at the different fields.

Method `is_running_task()`

Checks whether tasks have the status "running".

Usage

Rush$is_running_task(keys)

Arguments

keys: (character())
Keys of the tasks.

Method `is_failed_task()`

Checks whether tasks have the status "failed".

Usage

Rush$is_failed_task(keys)

Arguments

keys: (character())
Keys of the tasks.

Method `tasks_with_state()`

Returns keys of requested states.

Usage

Rush$tasks_with_state(states)

Arguments

states: (character())
States of the tasks.

Returns

(Named list of character()).

Method `clone()`

The objects of this class are cloneable with this method.

Usage

Rush$clone(deep = FALSE)

Arguments

deep: Whether to make a deep clone.

Examples

# This example is not executed since Redis must be installed
# \donttest{
   config_local = redux::redis_config()
   rush = rsh(network_id = "test_network", config = config_local)
#> Error in initialize(...): Can't connect to Redis. Check the configuration.
   rush
#> Error: object 'rush' not found
# }

Value

Local Workers

Remote Workers

Stopping Workers

Heartbeat

Data Structure

Task States

Queues

Fetch Tasks and Results

Error Handling

Logging

Seed

Public fields

Active bindings

Methods

Public methods

Method new()

Usage

Arguments

Method format()

Usage

Arguments

Returns

Method print()

Usage

Returns

Method reconnect()

Usage

Method start_local_workers()

Usage

Arguments

Method restart_local_workers()

Usage

Arguments

Method create_worker_script()

Usage

Method start_remote_workers()

Usage

Arguments

Method wait_for_workers()

Usage

Arguments

Method stop_workers()

Usage

Arguments

Method detect_lost_workers()

Usage

Arguments

Method reset()

Usage

Arguments

Method read_log()

Usage

Arguments

Method print_log()

Usage

Method push_tasks()

Usage

Arguments

Returns

Method push_priority_tasks()

Usage

Arguments

Returns

Method push_failed()

Usage

Arguments

Method retry_tasks()

Usage

Arguments

Method fetch_queued_tasks()

Usage

Arguments

Returns

Method fetch_priority_tasks()

Usage

Arguments

Returns

Method fetch_running_tasks()

Method `new()`

Method `format()`

Method `print()`

Method `reconnect()`

Method `start_local_workers()`

Method `restart_local_workers()`

Method `create_worker_script()`

Method `start_remote_workers()`

Method `wait_for_workers()`

Method `stop_workers()`

Method `detect_lost_workers()`

Method `reset()`

Method `read_log()`

Method `print_log()`

Method `push_tasks()`

Method `push_priority_tasks()`

Method `push_failed()`

Method `retry_tasks()`

Method `fetch_queued_tasks()`

Method `fetch_priority_tasks()`

Method `fetch_running_tasks()`

Method `fetch_finished_tasks()`

Method `wait_for_finished_tasks()`

Method `fetch_new_tasks()`

Method `wait_for_new_tasks()`

Method `fetch_failed_tasks()`

Method `fetch_tasks()`

Method `fetch_tasks_with_state()`

Method `wait_for_tasks()`

Method `write_hashes()`

Method `read_hashes()`

Method `read_hash()`

Method `is_running_task()`

Method `is_failed_task()`

Method `tasks_with_state()`

Method `clone()`