Expose host capabilities
Nova has API and implementation to expose capabilities of each host (i.e. cpu, memory, etc.). First, introduce the compute_host table. Then introduce a periodically task to populate the table. Then, expose host info via REST API.
Whiteboard
In this spec, we are referring to creation of a table very similar to the compute_nodes table in Nova[1].
Nova currently populates this table via it's compute drivers. For linux, it uses the libvirt APIs to find out about the current host capabilities. This is also done periodically to ensure that the right data is presented to the user.
In between this, there's a resource tracker, that tracks the resources from an OpenStack view per say. We won't implement a resource tracker in zun to start with.
We could start with the following work items:
1. Create a os-capability package in zun. This package can have drivers for various operating systems to obtain the data for cpu/mem/numa nodes etc.
I am also in favor of seeing this package released as a library in OpenStack, as opposed to residing completely in zun's code. But since we are in rapid development mode, we will just keep the design as such that the packages can be split out anytime, we want to create a library out of it.
2. For Linux Operating systems, use the following commands for cpu and mem stats:
For cpu - use lscpu and parse the o/p. This will give us the NUMA nodes value as well. I have the code ready for this.
For memory - use /proc/meminfo and parse the o/p for the details on memory.
3. Create a host_capabilities (or compute_nodes) table that would have the following entries to begin with:
- ID
- hostname
- total cpu
- used_cpu
- no_of_containers
- total memory (can be further divided into):
MemTotal: 263636352 kB
MemFree: 180490624 kB
MemAvailable: 214005120 kB
Buffers: 3328 kB
Cached: 35926272 kB
- NumaTopology Object, consisting of cpuset details.
4. Implement a periodic task in the driver that would refresh this host capabilities data periodically.
Gerrit topic: https:/
Addressed by: https:/
[WIP] Add resource provider table in DB
Addressed by: https:/
DB: Add inventory table
Addressed by: https:/
DB: Add resource class table
Addressed by: https:/
DB: Add allocation table
Addressed by: https:/
Add resource provider objects
Addressed by: https:/
Add resource class object
Addressed by: https:/
Add tests for custom object fields
Addressed by: https:/
Define a ResourceClassField in object
Addressed by: https:/
[WIP] Add inventory object
Addressed by: https:/
Implement etcd db API for resource_provider
Addressed by: https:/
[WIP] Add resource_providers API endpoint
Addressed by: https:/
Eagerly load resource_provider from foreignkey
Addressed by: https:/
Implement etcd API for ResourceClass
Addressed by: https:/
Add memory info to compute node.
Addressed by: https:/
Add container number info to compute node.
Addressed by: https:/
Add cpu, cpu_used for compute node
Addressed by: https:/
Add host operating system info to compute node
Addressed by: https:/
Add labels info to compute node
@hongbin, I guess the host capability is almost there, could this BP marked as complete? Any other capability that you prefer to be added?
Sorry, seems this BP has been implemented long time ago.
Work Items
Dependency tree
* Blueprints in grey have been implemented.
