Lose less work to errors in RStudio Jobs

A user reported running a 14-hour job but an error on the very last line terminated the job and nothing was saved/returned. Here’s a few tricks to ameliorate such situations:

Wrap code in try()

job::job() returns everything that’s left in the job session upon successful completion. So we just need to make the script run to the end and that’s where try() comes in handy. Say we want to do model comparisons between two models. The following code returns everything before the erroring line:

job::job({
  try({
    # Fit models
    fit_cyl = lm(mpg ~ cyl, mtcars)
    fit_wt = lm(mpg ~ wt, mtcars)
    
    # Compare - with spelling mistake. Oops.
    result = anova(fit_cyl, fit_wtt)
  })
})

Note that your job will be listed as “Success” in the RStudio jobs pane because it encountered no hard errors.

Save to file(s)

You can also save intermediate results to a file which can be loaded from the main session. Modifying the example above:

job::job({
    # Fit models
    fit_cyl = lm(mpg ~ cyl, mtcars)
    saveRDS(fit_cyl, "fit_cyl.rds")  # Ensure no data loss
    
    fit_wt = lm(mpg ~ wt, mtcars)
    saveRDS(fit_wt, "fit_wt.rds")  # Ensure no data loss
    
    # Compare - with spelling mistake. Oops.
    result = anova(fit_cyl, fit_wtt)
})

You can read more about exporting to files in the article on controlling exports using job::export().