A simple tool to test how kubernetes' topology manager behave, without need to run the real workload. This tool uses the very same packages from upstream kubernetes, to give the closest representation as possible as the real thing.
Usage example:
$ tmpolx -N 0-1 -P restricted \
'nvidia.com/gpu:[{01 true} {11 false}]' \
'openshift.io/intelsriov:[{10 true} {11 false}]' \
'cpu:[{01 true} {10 true} {11 false}]'
using policy "restricted"
. resource hints
. nvidia.com/gpu [{01 true} {11 false}]
. openshift.io/intelsriov [{10 true} {11 false}]
. cpu [{01 true} {10 true} {11 false}]
admit=false hint={01 false}
$ tmpolx -J -N 0-1 -P restricted \
'{"R":"cpu", "H":[{"M":"01","P":true},{"M":"10","P":true},{"M":"11","P":false}]}' \
'{"R":"nvidia.com/gpu", "H":[{"M":"01","P":true},{"M":"11","P":false}]}' \
'{"R":"openshift.io/intelsriov", "H":[{"M":"10","P":true},{"M":"11","P":false}]}'
using policy "restricted"
. resource hints
. nvidia.com/gpu [{01 true} {11 false}]
. openshift.io/intelsriov [{10 true} {11 false}]
. cpu [{01 true} {10 true} {11 false}]
admit=false hint={01 false}
$tmpolx expects to receive positional arguments which represent the topology manager hints.
{
"R": "something", # resource name
"H": [
# array of hints
{
"M": "1111", # bitmask
"P": true # preferred flag
}
]
}tmpolx accepts topology manager hints both in native go format (just copy/paste them from the kubelet logs) or in JSON format.
The go format is handier and simpler to use, but the support is still experimental. The JSON format is recommended to get
the maximum safety. The default is to use the go format.
Use the -J flag to enable this format.
cpu:[{01 true} {10 true} {11 false}]is
resource = "cpu"
hints =
{01 true}
{10 true}
{11 false}
# each hint is {mask preferred}
so we get
{
"R": "cpu",
"H":[
{
"M": "01",
"P": true
},
{
"M": "10",
"P": true
},
{
"M": "11",
"P": false
}
]
}(C) 2020 Red Hat Inc and licensed under the Apache License v2
just run
makesimilar tool to inspect cpumanager behaviour: https://github.com/fromanirh/cpumgrx