Resource are not overcommitted when cpu_allocation_ratio and ram_allocation _ratio value is set in nova.conf

[Rahul]

Brief Description
-----------------
Issue:
After deployment, created maximum numbers of VM's and then changed cpu_allocation_ratio and ram_allocation value in nova.conf of both compute nodes to achieve overcommit of CPU and RAM resources but Memory value doesn't reach maximum value according to overcommitted value.

#Physical CPU and RAM on compute0
[root@Ctrl0 ~]# openstack host show Comp0
+-------+----------------------------------+-----+-----------+---------+
| Host | Project | CPU | Memory MB | Disk GB |
+-------+----------------------------------+-----+-----------+---------+
| Comp0 | (total) | 40 | 322094 | 54387 |
+-------+----------------------------------+-----+-----------+---------+

#Physical CPU on compute1
[root@Ctrl0 ~]# openstack host show Comp1
+-------+----------------------------------+-----+-----------+---------+
| Host | Project | CPU | Memory MB | Disk GB |
+-------+----------------------------------+-----+-----------+---------+
| Comp1 | (total) | 40 | 322099 | 54387 |
+-------+----------------------------------+-----+-----------+---------+

#Changed cpu_allocation_ratio and ram_allocation_ratio values in nova.conf of both compute nodes
[root@Comp0 ~]# grep "cpu_allocation_ratio" /etc/kolla/nova-compute/nova.conf
cpu_allocation_ratio = 8.0
[root@Comp0 ~]# grep "ram_allocation_ratio" /etc/kolla/nova-compute/nova.conf
ram_allocation_ratio = 16.0
[root@Comp0 ~]#

[root@Comp1 ~]# grep "cpu_allocation_ratio" /etc/kolla/nova-compute/nova.conf
cpu_allocation_ratio = 8.0
[root@Comp1 ~]# grep "ram_allocation_ratio" /etc/kolla/nova-compute/nova.conf
ram_allocation_ratio = 16.0
[root@Comp1 ~]#

#Current vCPU and Memory consumption in system
Hostname Type VCPUs (used) VCPUs (total) RAM (used) RAM (total) Local Storage (used) Local Storage (total) Instances
Comp0 QEMU 130 40 360.5GB 314.5GB 4.1TB 53.1TB 39
Comp1 QEMU 118 40 340.5GB 314.5GB 3.2TB 53.1TB 34

VCPU Usage
Used 248 of 80

Memory Usage
Used 701GB of 629.1GB

#After Creating multiple instance, we will be able to reach maximum vCPU and RAM consumption to,

VCPU Usage
Used 312 of 80

Memory Usage
Used 813GB of 629.1GB

According to arch design mentioned in below link
https://docs.openstack.org/arch-design/design-compute/design-compute-overcommit.html

Overcommit of resources are not done for ram_allocation_ratio.. After 813GB usage of RAM we are not able to make more instance of 8GB of RAM.
---------------------------------------------------------------------------------------------------------------------------------------------

#Physical Memory of Both nodes
[root@Comp0 ~]# free -g
              total used free shared buff/cache available
Mem: 314 306 6 0 2 0
Swap: 31 31 0

[root@Comp1 ~]# free -g
              total used free shared buff/cache available
Mem: 314 282 16 0 15 19
Swap: 31 23 8
[root@Comp1 ~]#
----------------------------------------------------------------------------------------------------------------------------------------------

#Schedular filters in the system
scheduler_weight_classes = nova.scheduler.weights.all_weighers

[filter_scheduler]
available_filters = nova.scheduler.filters.all_filters
enabled_filters = AvailabilityZoneFilter,ComputeFilter,ComputeCapabilitiesFilter,ImagePropertiesFilter
----------------------------------------------------------------------------------------------------------------------------------------------

Kindly please share your valuable inputs regarding this overcommit ratio settings in the system.
Also please correct us to achieve overcommit of vCPU and RAM resources.

[Rahul (rahulxp22)]

Hi
Any update on this bug..

Please look into this as it hampers some testing scenarios.

[Artom Lifshitz (notartom)]

If this is a remotely recent version of OpenStack, the Placement service will be the source of truth for CPU and RAM inventories. Start with a `openstack resource provider inventory list <compute node UUID>` to see the inventory for your compute nodes, including the allocation ratio.

[Launchpad Janitor (janitor)]

[Expired for OpenStack Compute (nova) because there has been no activity for 60 days.]