In normal Shiny reactive code, whenever an observer, calc, or output is busy computing, it blocks the current session from receiving any inputs or attempting to proceed with any other computation related to that session.
The ExtendedTask class allows you to have an expensive operation that is
started by a reactive effect, and whose (eventual) results can be accessed
by a regular observer, calc, or output; but during the course of the
operation, the current session is completely unblocked, allowing the user
to continue using the rest of the app while the operation proceeds in the
background.
Note that each ExtendedTask object does not represent a single
invocation of its long-running function. Rather, it's an object that is
used to invoke the function with different arguments, keeps track of
whether an invocation is in progress, and provides ways to get at the
current status or results of the operation. A single ExtendedTask object
does not permit overlapping invocations: if the invoke() method is called
before the previous invoke() is completed, the new invocation will not
begin until the previous invocation has completed.
ExtendedTask versus asynchronous reactivesShiny has long supported using {promises} to write asynchronous observers, calcs, or outputs. You may be wondering what the differences are between those techniques and this class.
Asynchronous observers, calcs, and outputs are not–and have never been–designed to let a user start a long-running operation, while keeping that very same (browser) session responsive to other interactions. Instead, they unblock other sessions, so you can take a long-running operation that would normally bring the entire R process to a halt and limit the blocking to just the session that started the operation. (For more details, see the section on "The Flush Cycle".)
ExtendedTask, on the other hand, invokes an asynchronous function (that
is, a function that quickly returns a promise) and allows even that very
session to immediately unblock and carry on with other user interactions.
new()Creates a new ExtendedTask object. ExtendedTask should generally be
created either at the top of a server function, or at the top of a module
server function.
ExtendedTask$new(func)funcThe long-running operation to execute. This should be an
asynchronous function, meaning, it should use the
{promises} package, most
likely in conjuction with the
{future}
package. (In short, the return value of func should be a
Future object, or a promise, or something else
that promises::as.promise() understands.)
It's also important that this logic does not read from any
reactive inputs/sources, as inputs may change after the function is
invoked; instead, if the function needs to access reactive inputs, it
should take parameters and the caller of the invoke() method should
read reactive inputs and pass them as arguments.
invoke()Starts executing the long-running operation. If this ExtendedTask is
already running (meaning, a previous call to invoke() is not yet
complete) then enqueues this invocation until after the current
invocation, and any already-enqueued invocation, completes.
status()This is a reactive read that invalidates the caller when the task's status changes.
Returns one of the following values:
"initial": This ExtendedTask has not yet been invoked
"running": An invocation is currently running
"success": An invocation completed successfully, and a value can be
retrieved via the result() method
"error": An invocation completed with an error, which will be
re-thrown if you call the result() method
result()Attempts to read the results of the most recent invocation. This is a reactive read that invalidates as the task's status changes.
The actual behavior differs greatly depending on the current status of the task:
"initial": Throws a silent error (like req(FALSE)). If
this happens during output rendering, the output will be blanked out.
"running": Throws a special silent error that, if it happens during
output rendering, makes the output appear "in progress" until further
notice.
"success": Returns the return value of the most recent invocation.
"error": Throws whatever error was thrown by the most recent
invocation.
This method is intended to be called fairly naively by any output or
reactive expression that cares about the output–you just have to be
aware that if the result isn't ready for whatever reason, processing will
stop in much the same way as req(FALSE) does, but when the result is
ready you'll get invalidated, and when you run again the result should be
there.
Note that the result() method is generally not meant to be used with
observeEvent(), eventReactive(), bindEvent(), or isolate() as the
invalidation will be ignored.