Resource providers - Introduce resource classes¶
https://blueprints.launchpad.net/nova/+spec/resource-classes
This blueprint introduces a mechanism for representing the types of quantitative resources that may be provided by the system.
Problem description¶
The types of quantitative resources that are exposed in Nova are currently hard-coded into various places in the code base. For instance, on the Instance object, we store the amount of requested vCPUs in the vcpus field and the amount of RAM in the memory_mb field. We have a separate instance_pci_devices table in the database for PCI resource types. We store the NUMA-related resource information in a field in a different table, etc.
Whenever we come up with a new class of resources that we wish to provide, we end up creating either a new field in a table or a whole new database table for this new resource type. Any time we make changes to the database schema, we introduce some amount of downtime to the system. These changes, however, are not necessary. We should be able to add new resource classes to the system without changes to the database schema, and this series of blueprints lay the groundwork for doing just that.
This work will also enable us to have a generic resource pool system that does not hard-code resource classes into database field names. Instead of a system where we have a table tracking just shared disk storage, we can have a database table that can be used for tracking many different types of resources, and the schema of this table will not need to change when we add support for more types of shared resources.
Use Cases¶
As a cloud deployer, I wish to reduce the downtime experienced by database schema changes when a new type of resource (or a new way of handling an existing type of resource) is added to the system.
Proposed change¶
We propose to add a new Nova object field type called ResourceClassField that derives from the nova.objects.fields.EnumType object. The ALLOWED values of this field will be a set of agreed-upon constants. We do not propose adding operator extensibility of this list of constants, because we do not want to encourage situations where two OpenStack clouds have different definitions of the same-named resource class.
Alternatives¶
We could continue doing things the way we have been doing, adding new schema fields or new schema tables each time we add a new class of resources that the system provides.
Data model impact¶
A new nova object field type for resource classes will be introduced.
Initially, however, we can populate the enum with the known resource classes in the system:
VCPU
MEMORY_MB
DISK_GB
PCI_DEVICE
SRIOV_NET_VF
NUMA_SOCKET
NUMA_CORE
NUMA_THREAD
NUMA_MEMORY_MB
IPV4_ADDRESS
REST API impact¶
None at this time. In the future, we may wish to allow a cloud user to query the resource classes via HTTP, but initially I don’t believe this is necessary.
Security impact¶
None.
Notifications impact¶
None.
Other end user impact¶
Future work can introduce a nova resource-class-list command, however this is not particularly important for this blueprint spec.
Performance Impact¶
None introduced in this blueprint since we are only adding a set of constants in an Enum-like field type.
Other deployer impact¶
None.
Developer impact¶
None.
Implementation¶
Assignee(s)¶
- Primary assignee:
cdent
- Other contributors:
jaypipes
Work Items¶
Create new nova.objects.fields.ResourceClassField nova field
Dependencies¶
None.
Testing¶
Unit testing should be sufficient for this small blueprint spec. Functional and integration testing is more appropriate for the specs like generic-resource-pools that build on top of this work.
Documentation Impact¶
None.
References¶
generic-resource-pools specification:
History¶
Release Name |
Description |
---|---|
Mitaka |
Introduced |