Capture and signal interrupts?

[HenrikBengtsson]

Interrupts (e.g. user interrupts/Ctrl-C, process terminate, ...) may be signaled while a future is evaluated. Should the future package make sure such interrupts are automatically caught and record them such that they are signaled when value() is called, cf. errors?

Details

Interrupts can be caught in R using tryCatch(..., interrupt=...), e.g.

res <- FALSE
res <- tryCatch({
  Sys.sleep(10)
  TRUE
}, interrupt = function(int) {
  str(int)
}, error = function(ex) {
  str(ex)
})

This is somewhat related to Issue #25, but since interrupts are quite different than warnings I think the two ideas should be kept separate.

PS. Added this issue based on a note to myself from 2015-06-18.

[HenrikBengtsson]

Test with multicore futures as they behave right now if there is an interrupt:

> library("future")

## Start a multicore process
> f <- multicore({ print(1); Sys.sleep(30); print(2); TRUE })
> [1] 1

## Let's find the PID (non-official; don't use in code)
> f$job$pid
[1] 47841

## Let's send an INT(errupt) signal
> system2("kill", c("-INT", f$job$pid))

## Check the value of the multicore future
> value(f)
[1] "fatal error in wrapper code"
attr(,"class")
[1] "try-error"
> traceback()
No traceback available

[HenrikBengtsson]

This is now detected and an error is generated;

> library("future")
> f <- multicore({ print(1); Sys.sleep(30); print(2); TRUE })
> system2("kill", c("-INT", f$job$pid))
[1] 1
> value(f)
Error in value.MulticoreFuture(f) :
Detected an error ('fatal error in wrapper code') by the 'parallel' package while trying
to retrieve the value of a MulticoreFuture ('{; print(1); Sys.sleep(30); print(2); TRUE; }').
This could be because the forked R process that evalutes the future was terminated
before it was completed.
>