Environment variables for resize notification command
All environment variables that are set for a job are also set when a job is resized.
The following (additional) environment variables apply only to the resize notification
command environment (when using resizable jobs):
- LSB_RESIZE_NOTIFY_OK
- A notification command should exit with this variable if the allocation resize notification
command succeeds.
LSF updates the job allocation to reflect the new allocation.
- LSB_RESIZE_NOTIFY_FAIL
- A notification command should exit with this variable if the allocation resize notification
command fails.
For an allocation grow event, LSF schedules the pending allocation request.
For an allocation shrink event, LSF fails the release request.
- LSB_RESIZE_EVENT = grow | shrink
- Indicates why the notification command was called. Specifying grow means add more resources to an existing allocation. Specifying shrink means remove some resources from existing allocation.
- LSB_RESIZE_HOSTS = host1 num1 host2 num2 … hostZ numZ
- Lists the additional slots for a grow event, or the released slots for a shrink event.
- LSB_RESIZE_GPUS=hostname:[[task1:GPUID1,GPUID2][task2:GPUID1]];hostname:[[task1:GPUID1,GPUID2][task2:GPUID1]]]
- Lists the additional GPUs for a grow event, or the released GPUs for a shrink event.
- LSB_ RESIZE_TIME=timestamp
- Timestamp for the resize action, which helps identify the exact resized GPU allocation for changed tasks. Export this environment variable for the blaunch command before new tasks grow or shrink. If not set, when the blaunch re-sizes tasks, LSF uses the latest resized GPU allocation.