Resit exams 15/08/2014

The morning's inf1-op exam was interrupted at around 09:45 by a fire alarm. Knowing no alarm tests were due, the exam labs were evacuated following existing procedures, and the candidates assembled outside. No attempt was undertaken to shepherd them or separate them from the other AT occupants. They were free to circulate and chat to each other.

Alistair Hill (ISS) immediately took action to contact registry and follow what advice they offered with regard to restarting or abandoning the exam. Registry were unable to provide a firm response one way or the other and despatched one of their staff to the Tower to assess the situation.

Fire stewards started letting people back in to the building at around 10:00, the invigilators and candidates finding their way back to their allocated rooms and PCs. Working without further instruction from registry the exam was resumed in the knowledge that it could be cancelled at short notice.

Registry then confirmed that the exam should be resumed. Candidates had no problem picking up where they left off, and extra time was agreed and added on to the finish time as compensation for the disruption.

The main exam was finished by around 13:00 and the West lab unlocked successfully. The North lab was then unsuccessfully unlocked. It seems that when the global #define DICE_STICK_WITH_SL64 definition was applied, the North lab, AT-4.14a, AT-5.01, AT-5.02 and AT-5.03 were already locked down, resulting in the following in each lab client's profile (with each client's ip address):

#define DICE_STICK_WITH_SL64
#define EXAM_LOCKDOWN 129.215.58.233
#define EXAM_COURSE inf1-op
#define EXAM_PAPER rpe
This resulted in a failure message from steen on running examlockdown-unlock:

FAILED (INCORRECT EXAM LOCKDOWN IN PROFILE)

Of course the West lab was only locked down in the morning, meaning that those clients' profiles looked like this (with each client's ip address):

#define EXAM_LOCKDOWN 129.215.58.233
#define EXAM_COURSE inf1-op
#define EXAM_PAPER rpe
#define DICE_STICK_WITH_SL64
Hence why the West lab unlocked. The fix for this was to change all exam clients' profiles that weren't in the West lab to the correct order so they could be unlocked. It is noteworthy that despite an lcfg branch being used as procedure dictates, the lab clients were not safe from this mass profile change. Branching does not protect in all circumstances.

The morning fire alarm and aftermath meant that the gap between the last of the extra-time candidates finishing inf1-op and the start of the lp exam was uncomfortably close. With little time to spare, the clients in AT-5.03 were locked to the afternoon's lp exam successfully. The other extra time clients were then unlocked unsuccessfully. Suspecting the #define ordering issue was interfering again an investigation was conducted. At this time it was discovered that a mass profile rebuild of well over 1000 clients was in progress, delaying the unlock significantly. Obviously an emergency as the emergency_overrides.h file had been changed.

It was only by good luck with timing that the relock of the AT-5.03 clients from inf1-op to lp went through and that there wasn't a 45 minute delay to the start of the afternoon's exam. Protocol in such a situation is unclear.

Richard Bell

-- AlisonDownie - 27 Aug 2014

Topic revision: r1 - 27 Aug 2014 - 14:05:28 - AlisonDownie
 
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback
This Wiki uses Cookies