Handle burst deployments with scheduler
Problem: Massive amount of VM is deployed on the same host because the lack of "predicted" monitor information.
Solutions:
- Wait for fresh monitor information after n deployment.
- Calculate predictive resource usage based on machines with deploy / wake up activity
- Neeeed Moar powaaa
Tune priority:
- The higher priority should mean higher probability (soft ratio)
- Count in recently scheduled vm-s with weight or with sort key
Reschedule:
- CPU threshold
- Reduce (trait etc) and sort VM's list
- Migrate VM
-
-
-
Assignee removed
Toggle commit list -
Please
register
or
sign in
to comment