ENH: Error messages in failed scripts should be the job log

[patr1ckm]

job <- scripts_post_r(name = "test", source = "stop('err')")
run <- scripts_post_r_runs(job$id)
await(scripts_get_r_runs, job$id, run$id)

Produces:

 Error: scripts_get_r_runs(id = 10602571, NA = 77083111): 

The object is:

scripts_get_r_runs(job$id, run$id)
<civis_api>
List of 7
 $ id               : int 77083111
 $ rId              : int 10602571
 $ state            : chr "failed"
 $ isCancelRequested: logi FALSE
 $ startedAt        : chr "2018-03-30T15:33:59.000Z"
 $ finishedAt       : chr "2018-03-30T15:34:37.000Z"
 $ error            : NULL
 - attr(*, "class")= chr [1:2] "civis_api" "list"

[keithing]

@stephen-hoover confirmed that this field has only ever returned Platform errors. This means queries to Redshift will return a useful error, but user's code in container jobs has never returned this error field.

[patr1ckm]

Got it, I was definitely misunderstanding how this worked when I wrote the error handling in await.

I think the right thing to do is on an error to fetch the job log (whatever the job is) and surface the log as an error message to the user. I haven't tested, but I'm assuming this will work for queries as well.

[keithing]

I think the right thing to do is on an error to fetch the job log (whatever the job is) and surface the log as an error message to the user.

If we can do that for a general case, that would be awesome.

I'm assuming this will work for queries as well

For queries we do get the error back in result[["error"]], so we should be okay there as well.