topolvm-scheduler.md

topolvm-scheduler

topolvm-scheduler is a Kubernetes scheduler extender for TopoLVM.

It filters and prioritizes Nodes based on the amount of free space in their volume groups.

Scheduler Policy

topolvm-scheduler need to be configured in scheduler policy as follows:

{
    ...
    "extenders": [{
        "urlPrefix": "http://...",
        "filterVerb": "predicate",
        "prioritizeVerb": "prioritize",
        "managedResources":
        [{
          "name": "topolvm.io/capacity",
          "ignoredByScheduler": true
        }],
        "nodeCacheCapable": false
    }]
}

As shown above, only pods that request topolvm.io/capacity resource are managed by topolvm-scheduler.

Verbs

The extender provides two verbs:

• predicate to filter nodes
• prioritize to score nodes

predicate

This verb filters out nodes whose volume groups have not enough free space.

Volume group capacity is identified from the value of capacity.topolvm.io/<device-class> annotation.

prioritize

This verb scores nodes. The score of a node is calculated by this formula:

$$ \mathrm{min} \left( 10, \ \mathrm{max} \left( 0, \ \log_{2}{ \left( \mathrm{capacity} \gg 30 / \mathrm{divisor} \right) } \right) \right) $$

For example, the default of divisor is 1, then if a node has the free disk capacity more than 1024GiB, topolvm-scheduler scores the node as 10. divisor should be adjusted to suit each environment.

divisor can be given through the configuration file.

Command-line Flags

Name	Type	Default	Description
config	string	``	Config file path

Config File Format

The divisor parameter can be specified in YAML file:

default-divisor: 10
divisors:
  ssd: 5
  hdd: 10

Name	Type	Default	Description
listen	string	:8000	HTTP listening address
default-divisor	float64	1	A default value of the variable for node scoring.
divisors	map[string]float64	{}	A variable for node scoring per device-class.