[xcompass]

Problem

DelayedJob::PushClassGrades inherits the 2-hour grab_for from DelayedJob::Worker. A whole-class push makes one AGS request per student per linked assignment, so on a very large course it can run longer than that. TheSchwartz then treats it as failed, and another worker starts a second copy while the first is still running.

GetClassMembership already overrides grab_for to 8 hours for exactly this reason (per its comment, a 5,100-student course took about 4 hours). UBC's largest course is 6,518 students with 56 linked assignments.

Change

PushClassGrades::grab_for returns $ENV{DELAYED_JOB_CLASS_GRADES_GRAB_FOR} // 8h, mirroring GetClassMembership and DELAYED_JOB_NRPS_GRAB_FOR.

• Why not per course: TheSchwartz reads grab_for from the worker class when grabbing ($worker_class->grab_for, TheSchwartz.pm:420), so it can't vary per course or job.
• Trade-off: if a worker genuinely dies mid-push, that class push waits up to 8 h before a retry instead of 2 h. The nightly class push is a catch-up/backstop, not the path new grades normally take. PushUserGrades (per-student, on submission) keeps its 2 h.
• Context: worker pods on UBC prod now carry karpenter.sh/do-not-disrupt (chart 0.3.17), so node consolidation no longer kills workers mid-job.

Verification

Loaded in the WeBWorK image (webwork-verf worker pod, patched file ahead of the installed one):

PushClassGrades grab_for=28800s (8.0h)
PushUserGrades  grab_for=7200s (2.0h)
GetClassMembership grab_for=28800s
with env override: 14400s

perltidy 20240903 with .perltidyrc: no changes.