Maintenance and downtime

Planned maintenance

Planned maintenance is scheduled periodically on BEAR's services to allow for upgrades to the hardware and software, ensuring the systems remain in a healthy state. The dates for maintenance periods are normally known several weeks or months in advance and are communicated to users via email.

Occasionally maintenance may be scheduled at short notice, for example where a security fix is required. In this case, users will be notified via email as soon as possible.

Emergency downtime

Emergency downtime is unscheduled downtime that could be due to hardware or software failures, or due to environmental issues such as high temperatures in the data centre due to weather conditions.
In the event of emergency downtime, users will be notified via email as soon as possible.

Maintenance and emergency downtime on BlueBEAR

In the event of planned maintenance or expected emergency downtime (e.g. high temperatures), the cluster will have a Slurm "reservation" applied to it. This reservation will prevent jobs from starting on the cluster after a specified time but note that the duration of the reservation will normally be significantly longer than is required for the maintenance or downtime, as this ensures the ARC Systems Team have sufficient time to complete work and deal with any issues that might arise.

In the period prior to a reservation being active, the cluster will continue to accept jobs but any job that has a wall-time that would extend it beyond the start of the reservation will be in a "Pending" state until the reservation has been removed. This is to ensure that jobs don't start and then fail due to a system shutdown. Also, in this situation, the reported "start time" of a job (see the section on "When will my job start?") will likely be "N/A" as the scheduler will not be able to determine accurately when the job will start due to the reservation.