Large job error

[rstudley]

One of the test jobs I ran yesterday failed.
JobGUID: 08c11991-1a6e-4cd0-9586-0b36626c8dd7
More info in the EE List of Tests.
NOTE: This was a very large job, 27,316 members of the treatment group, possibly representing 40% of the available cohort. I was able successfully to run a subset of this job: N=289, JobGUID=ca026078-ec4c-44ac-b52a-812899f72670.

[alexsmithRTI]

I deleted Roger's comment, and am testing if there is still a problem.

@markmfredrickson, did you see this?

[benthestatistician]

This sounds like an R side problem, not a web interface problem. Accordingly could you re-post it over in the (private) GH repo for that stuff, here, @rstudley ? (I'd offer to do this myself but that way we can find out whether you're already all set up to post issues there and assign them, or whether I'll need to up your access in order for that to occur.) I'd suggest closing this issue in the process, after adding a pointer to the new thread in a comment.

It's not too surprising to learn that we're still erroring out on a job of this size. I conjecture that our measures to avoid exceeding memory limits turned out to be not quite up to the task. That may be easy or difficult to fix. Once that's been addressed, we may find either that statistical performance is fine or that statistical performance is terrible on this job; either way I wouldn't be too surprised.

[rstudley]

I re-posted this issue as #38 on the Stat thread. Closing here.

[alexsmithRTI]

Add warning to upload page, < 10k student IDs for now.

[alexsmithRTI]

Also add a check for # of student IDs returned when report is run, for immediate feedback.

[rstudley]

For the WARNING, let's add a short paragraph under "Helpful Information" at the right-hand side of the Step1b page:
Treatment groups consisting of more than 10,000 students might not run successfully due to resource constraints.

ALSO, rather than a check for the # of student IDs upon job submission, we agreed instead to give feedback to the user if/when a job fails due to too many IDs submitted. Let's use this text:
Treatment group too large. Consider running separate analyses by grade level or contacting help@evaluationengine.org for assistance.