Tasks

Core.TaskType
Task(func)

Create a Task (i.e. coroutine) to execute the given function func (which must be callable with no arguments). The task exits when this function returns.

Examples

julia> a() = sum(i for i in 1:1000);

julia> b = Task(a);

In this example, b is a runnable Task that hasn't started yet.

source
Base.istaskdoneFunction
istaskdone(t::Task) -> Bool

Determine whether a task has exited.

Examples

julia> a2() = sum(i for i in 1:1000);

julia> b = Task(a2);

julia> istaskdone(b)
false

julia> schedule(b);

julia> yield();

julia> istaskdone(b)
true
source
Base.istaskstartedFunction
istaskstarted(t::Task) -> Bool

Determine whether a task has started executing.

Examples

julia> a3() = sum(i for i in 1:1000);

julia> b = Task(a3);

julia> istaskstarted(b)
false
source
Base.yieldFunction
yield()

Switch to the scheduler to allow another scheduled task to run. A task that calls this function is still runnable, and will be restarted immediately if there are no other runnable tasks.

source
yield(t::Task, arg = nothing)

A fast, unfair-scheduling version of schedule(t, arg); yield() which immediately yields to t before calling the scheduler.

source
Base.yieldtoFunction
yieldto(t::Task, arg = nothing)

Switch to the given task. The first time a task is switched to, the task's function is called with no arguments. On subsequent switches, arg is returned from the task's last call to yieldto. This is a low-level call that only switches tasks, not considering states or scheduling in any way. Its use is discouraged.

source
Base.task_local_storageMethod
task_local_storage(body, key, value)

Call the function body with a modified task-local storage, in which value is assigned to key; the previous value of key, or lack thereof, is restored afterwards. Useful for emulating dynamic scoping.

source
Base.ConditionType
Condition()

Create an edge-triggered event source that tasks can wait for. Tasks that call wait on a Condition are suspended and queued. Tasks are woken up when notify is later called on the Condition. Edge triggering means that only tasks waiting at the time notify is called can be woken up. For level-triggered notifications, you must keep extra state to keep track of whether a notification has happened. The Channel and Threads.Event types do this, and can be used for level-triggered events.

This object is NOT thread-safe. See Threads.Condition for a thread-safe version.

source
Base.notifyFunction
notify(condition, val=nothing; all=true, error=false)

Wake up tasks waiting for a condition, passing them val. If all is true (the default), all waiting tasks are woken, otherwise only one is. If error is true, the passed value is raised as an exception in the woken tasks.

Return the count of tasks woken up. Return 0 if no tasks are waiting on condition.

source
Base.scheduleFunction
schedule(t::Task, [val]; error=false)

Add a Task to the scheduler's queue. This causes the task to run constantly when the system is otherwise idle, unless the task performs a blocking operation such as wait.

If a second argument val is provided, it will be passed to the task (via the return value of yieldto) when it runs again. If error is true, the value is raised as an exception in the woken task.

Examples

julia> a5() = sum(i for i in 1:1000);

julia> b = Task(a5);

julia> istaskstarted(b)
false

julia> schedule(b);

julia> yield();

julia> istaskstarted(b)
true

julia> istaskdone(b)
true
source
Base.@taskMacro
@task

Wrap an expression in a Task without executing it, and return the Task. This only creates a task, and does not run it.

Examples

julia> a1() = sum(i for i in 1:1000);

julia> b = @task a1();

julia> istaskstarted(b)
false

julia> schedule(b);

julia> yield();

julia> istaskdone(b)
true
source
Base.sleepFunction
sleep(seconds)

Block the current task for a specified number of seconds. The minimum sleep time is 1 millisecond or input of 0.001.

source
Base.ChannelType
Channel{T=Any}(size::Int=0)

Constructs a Channel with an internal buffer that can hold a maximum of size objects of type T. put! calls on a full channel block until an object is removed with take!.

Channel(0) constructs an unbuffered channel. put! blocks until a matching take! is called. And vice-versa.

Other constructors:

  • Channel(): default constructor, equivalent to Channel{Any}(0)
  • Channel(Inf): equivalent to Channel{Any}(typemax(Int))
  • Channel(sz): equivalent to Channel{Any}(sz)
Julia 1.3

The default constructor Channel() and default size=0 were added in Julia 1.3.

source
Base.put!Method
put!(c::Channel, v)

Append an item v to the channel c. Blocks if the channel is full.

For unbuffered channels, blocks until a take! is performed by a different task.

Julia 1.1

v now gets converted to the channel's type with convert as put! is called.

source
Base.take!Method
take!(c::Channel)

Remove and return a value from a Channel. Blocks until data is available.

For unbuffered channels, blocks until a put! is performed by a different task.

source
Base.isreadyMethod
isready(c::Channel)

Determine whether a Channel has a value stored to it. Returns immediately, does not block.

For unbuffered channels returns true if there are tasks waiting on a put!.

source
Base.fetchMethod
fetch(c::Channel)

Wait for and get the first available item from the channel. Does not remove the item. fetch is unsupported on an unbuffered (0-size) channel.

source
Base.closeMethod
close(c::Channel[, excp::Exception])

Close a channel. An exception (optionally given by excp), is thrown by:

source
Base.bindMethod
bind(chnl::Channel, task::Task)

Associate the lifetime of chnl with a task. Channel chnl is automatically closed when the task terminates. Any uncaught exception in the task is propagated to all waiters on chnl.

The chnl object can be explicitly closed independent of task termination. Terminating tasks have no effect on already closed Channel objects.

When a channel is bound to multiple tasks, the first task to terminate will close the channel. When multiple channels are bound to the same task, termination of the task will close all of the bound channels.

Examples

julia> c = Channel(0);

julia> task = @async foreach(i->put!(c, i), 1:4);

julia> bind(c,task);

julia> for i in c
           @show i
       end;
i = 1
i = 2
i = 3
i = 4

julia> isopen(c)
false
julia> c = Channel(0);

julia> task = @async (put!(c,1);error("foo"));

julia> bind(c,task);

julia> take!(c)
1

julia> put!(c,1);
ERROR: foo
Stacktrace:
[...]
source
Base.asyncmapFunction
asyncmap(f, c...; ntasks=0, batch_size=nothing)

Uses multiple concurrent tasks to map f over a collection (or multiple equal length collections). For multiple collection arguments, f is applied elementwise.

ntasks specifies the number of tasks to run concurrently. Depending on the length of the collections, if ntasks is unspecified, up to 100 tasks will be used for concurrent mapping.

ntasks can also be specified as a zero-arg function. In this case, the number of tasks to run in parallel is checked before processing every element and a new task started if the value of ntasks_func is less than the current number of tasks.

If batch_size is specified, the collection is processed in batch mode. f must then be a function that must accept a Vector of argument tuples and must return a vector of results. The input vector will have a length of batch_size or less.

The following examples highlight execution in different tasks by returning the objectid of the tasks in which the mapping function is executed.

First, with ntasks undefined, each element is processed in a different task.

julia> tskoid() = objectid(current_task());

julia> asyncmap(x->tskoid(), 1:5)
5-element Array{UInt64,1}:
 0x6e15e66c75c75853
 0x440f8819a1baa682
 0x9fb3eeadd0c83985
 0xebd3e35fe90d4050
 0x29efc93edce2b961

julia> length(unique(asyncmap(x->tskoid(), 1:5)))
5

With ntasks=2 all elements are processed in 2 tasks.

julia> asyncmap(x->tskoid(), 1:5; ntasks=2)
5-element Array{UInt64,1}:
 0x027ab1680df7ae94
 0xa23d2f80cd7cf157
 0x027ab1680df7ae94
 0xa23d2f80cd7cf157
 0x027ab1680df7ae94

julia> length(unique(asyncmap(x->tskoid(), 1:5; ntasks=2)))
2

With batch_size defined, the mapping function needs to be changed to accept an array of argument tuples and return an array of results. map is used in the modified mapping function to achieve this.

julia> batch_func(input) = map(x->string("args_tuple: ", x, ", element_val: ", x[1], ", task: ", tskoid()), input)
batch_func (generic function with 1 method)

julia> asyncmap(batch_func, 1:5; ntasks=2, batch_size=2)
5-element Array{String,1}:
 "args_tuple: (1,), element_val: 1, task: 9118321258196414413"
 "args_tuple: (2,), element_val: 2, task: 4904288162898683522"
 "args_tuple: (3,), element_val: 3, task: 9118321258196414413"
 "args_tuple: (4,), element_val: 4, task: 4904288162898683522"
 "args_tuple: (5,), element_val: 5, task: 9118321258196414413"
Note

Currently, all tasks in Julia are executed in a single OS thread co-operatively. Consequently, asyncmap is beneficial only when the mapping function involves any I/O - disk, network, remote worker invocation, etc.

source
Base.asyncmap!Function
asyncmap!(f, results, c...; ntasks=0, batch_size=nothing)

Like asyncmap, but stores output in results rather than returning a collection.

source