Would be nice if the high priority wu fell inline when an open thread becomes next available.

Let's set some facts straight.

1) The default is [still] 60 seconds, one minutes for WtD intervals. It's under review [you don't want this on 4-8-12-16 and more core devices], but when that review is translating to application... don't know [could be hard coded in clients in case there's no global_prefs.xml].

2) The HCC-GPU jobs have no checkpoint at all... too short on average to bother and complicate the program... risk resume integrity issues. Anticipate this earliest to happen when 2 HCC tasks are packaged together [planned for soon].

Consider, if no-one comes in the correct, the incorrect gets repeated.

BTW, Write to disk "at most" with emphasis, is a limiter, not an enforcer. If the application has no checkpoint point [where the intermediate data point is small, thus if saved not interfering with user, it wont write one.

Or, high-priority should mean 'the right to be next in line' and not 'the right to interrupt anything because I can'. If anything else, there needs to be a limit imposed on high-priority WUs before they are allowed to interrupt an in-progress WU.
;