I have been experimenting a bit with how a cluster behaves in terms of migrations when a node is removed from an HA group. As far as I can tell, when the removed node is evacuated by HA, the target node seems to be selected based on which remaining node has the lowest CPU usage at the time of selection. Is it possible to set it to select based on the lowest memory usage instead?
HA migration selection critera?
- Thread starter konaya
- Start date
Not exactly, the HA manager is using something far simpler at the moment (to avoid high dynamic changes triggering unstable behavior), it simply balances the recovered service to the remaining nodes by active service count.
A more elaborate target selection is on the long-term roadmap, but nothing currently actively being worked on.
Does active service count mean number of VPSes/containers or simply the number of active processes?
Would you mind pointing me in the right direction to find where in the code this check happens? I'd like to study it and see if we can find a workaround in-house.
Would you mind pointing me in the right direction to find where in the code this check happens? I'd like to study it and see if we can find a workaround in-house.