possible race condition writing disk.frame objects

[achubaty]

While running concurrent replicates, I occasionally see the following error:

[ENOTEMPTY] Failed to remove '<projectPath>/inputs/cohortDataFactorial55440/.metadata': directory not empty

Since useDiskFrame() is simply using the number of rows in cohortData/speciesTable when writing to disk, this may be leading to a race condition where one run tries to cleanup while the directory is still in use by another.

useDiskFrame <- function(sim) {
  setup_disk.frame(workers = 2) ## TODO: is there a better default? should this be user-specified?

  cdRows <- nrow(sim$cohortDataFactorial)
  ## the rows of a factorial object will determine whether it is unique in 99.9% of cases
  sim$cohortDataFactorial <- as.disk.frame(sim$cohortDataFactorial, overwrite = TRUE,
                                           outdir = file.path(inputPath(sim),
                                                              paste0("cohortDataFactorial", cdRows)))
  stRows <- nrow(sim$speciesTableFactorial)
  sim$speciesTableFactorial <- as.disk.frame(sim$speciesTableFactorial, overwrite = TRUE,
                                             outdir = file.path(inputPath(sim),
                                                                paste0("speciesTableFactorial", stRows)))
  ## NOTE: disk.frame objects can be converted to data.table with as.data.table
  gc(reset = TRUE)
  return(sim)
}

@ianmseddy any thoughts on how to mitigate this?

[achubaty]

If these directories are intended to be transient (and they are being cleaned up by something), then can we simply use a random string instead of the row-count?

(But why are they considered transient at all? Surely we are saving them so they can be inspected later?)

[achubaty]

these are now written to outputPath(sim), which avoids the race condition, but if the module is replicated will create multiple similar objects on disk. generally, these should not be replicated.