Dejan Zele Pejchev
Dejan Zele Pejchev
Currently, the executor (i.e., the in-cluster component of Armada) only reports free resources to the Armada server (the central component of Armada responsible for assigning jobs to executors). Available resources...
Once we have added proto messages fields necessary to support preemption in https://github.com/G-Research/armada/issues/1356 we need to update the executor to update these fields.
Currently, the Armada server only assigns jobs to an executor if the worker cluster managed by that executor has resources not assigned to any pod. To support preemption, the server...
Similar to how we show failed jobs, we should have a label for preempted jobs.
We need to notify users if their job was killed due to preemption. To that end, we should add a specific failure message for preemption. Similar to how OOM errors...