Large number of collisions when placing instances with anti affinity rules
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Nova Cloud Controller Charm |
Fix Released
|
Medium
|
Tiago Pasqualini da Silva |
Bug Description
When deploying heat stacks with anti affinity server groups we are experiencing a large number of collisions when creating instances. Since a heat script is being used for instance creation requests occur in very short time frame increasing the chances of collisions.
This results in rescheduling of these instances increasing stack deployment times.
As the criteria for identical instances often results in identical host selection we would like to be able to configure the filter_scheduler flag "host_subset_size" to a value greater then one to reduce the probability of scheduling collisions and thus reduce rescheduling overall.
This should reduce heat stack deployment times.
This request is related to bug #1817877 which only helps if the placement scores of two or more instances is exactly identical which on our cloud is a rather rare occurrence.
Changed in charm-nova-cloud-controller: | |
milestone: | none → 19.10 |
importance: | Undecided → Medium |
status: | Confirmed → In Progress |
Changed in charm-nova-cloud-controller: | |
milestone: | 19.10 → 20.01 |
Changed in charm-nova-cloud-controller: | |
status: | Fix Committed → Fix Released |
The actual request is to provide the ability in the nova-controller charm to set the filter_ scheduler: host_subset_ size in nova.conf to an arbitrary number to reduce anti affinity collisions. This parameter would need to be tuneable to discover the optimal setting for a particular cloud.
Thanks,
Steven