Resume computation with partial results

Description

Identifies unsuccessful results returned from bplapply, bpmapply, bpvec, bpaggregate or bpvectorize.

Usage

  bpok(x, type = bperrorTypes())

  bperrorTypes()

  bpresult(x)

Arguments

Details

bpok() returns a logical() vector: FALSE for any jobs that resulted in an error. x is the result list output by bplapply, bpmapply, bpvec, bpaggregate or bpvectorize.

bperrorTypes() returns a character() vector of possible error types generated during parallel evaluation. Types are:

• bperror: Any of the following errors. This is the default value for bpok().

• remote_error: An R error occurring while evaluating FUN(), e.g., taking the square root of a character vector, sqrt("One").

• unevaluated_error: When *Param(stop.on.error = TRUE) (default), a remote error halts evaluation of other tasks assigned to the same worker. The return value for these unevaluated elements is an error of type unevaluated_error.

• not_available_error: Only produced by DoparParam() when a remote error occurs during evaluation of an element of X – DoparParam() sets all values after the remote error to this class.

• worker_comm_error: An error occurring while trying to communicate with workers, e.g., when a worker quits unexpectedly. when this type of error occurs, the length of the result may differ from the length of the input X.

bpresult() when applied to an object with a class of one of the error types returns the list of tasks results.

Author(s)

Michel Lang, Martin Morgan, Valerie Obenchain, and Jiefei Wang

See Also

tryCatch

Examples

## -----------------------------------------------------------------------
## Catch errors: 
## -----------------------------------------------------------------------

## By default 'stop.on.error' is TRUE in BiocParallelParam objects.  If
## 'stop.on.error' is TRUE an ill-fated bplapply() simply stops,
## displaying the error message.
param <- SnowParam(workers = 2, stop.on.error = TRUE)
result <- tryCatch({
    bplapply(list(1, "two", 3), sqrt, BPPARAM = param)
}, error=identity)
result
class(result)
bpresult(result)

## If 'stop.on.error' is FALSE then the computation continues. Errors
## are signalled but the full evaluation can be retrieved
param <- SnowParam(workers = 2, stop.on.error = FALSE)
X <- list(1, "two", 3)
result <- bptry(bplapply(X, sqrt, BPPARAM = param))
result

## Check for errors:
fail <- !bpok(result)
fail

## Access the traceback with attr():
tail(attr(result[[2]], "traceback"), 5)

## -----------------------------------------------------------------------
## Resume calculations: 
## -----------------------------------------------------------------------

## The 'resume' mechanism is triggered by supplying a list of partial
## results as 'BPREDO'. Data elements that failed are rerun and merged
## with previous results.

## A call of sqrt() on the character "2" returns an error. Fix the input
## data by changing the character "2" to a numeric 2:
X_mod <- list(1, 2, 3)
bplapply(X_mod, sqrt, BPPARAM = param , BPREDO = result)