In scenarios such as node pool replacement and rolling node upgrade, an old resource pool needs to be replaced with a new one. To prevent the node pool replacement from affecting services, enable soft affinity to schedule service pods to the new node pool. Soft affinity scheduling tries to schedule newly created pods or rescheduled pods to the new node pool. If the pods cannot be scheduled to the new node pool, for example, due to insufficient resources, the pods can also be scheduled to the old node pool. Since a node pool replacement should not affect services, the node affinity configuration is not declared in service workloads. Use soft affinity in cluster scheduling to schedule pods to new node pools when a pool replacement is triggered.
Volcano aims to soft schedule service pods to specified nodes when node soft affinity is not configured on service workloads.
Soft affinity scheduling of a node pool is implemented based on labels in the node pool. Each node in the node pool is scored to select the optimal one for pod scheduling.
The rule is to schedule pods to nodes with specified labels as far as possible.
The formula for scoring a node is as follows:
Node score = Weight x MaxNodeScore x haveLabel
Parameters:
...
"default_scheduler_conf": {
"actions": "allocate, backfill, preempt",
"tiers": [
{
"plugins": [
{
"name": "priority"
},
{
"name": "gang"
},
{
"name": "conformance"
}
]
},
{
"plugins": [
{
"name": "drf"
},
{
"name": "predicates"
},
{
"name": "nodeorder"
}
]
},
{
"plugins": [
{
"name": "cce-gpu-topology-predicate"
},
{
"name": "cce-gpu-topology-priority"
},
{
"name": "cce-gpu"
},
{
// Enable node pool affinity scheduling.
"name": "nodepoolaffinity",
// Configure the affinity scheduling weight and labels of the node pool.
"arguments": {
"nodepoolaffinity.weight": 10000,
"nodepoolaffinity.label": "nodepool1=nodepool1"
}
}
]
},
{
"plugins": [
{
"name": "nodelocalvolume"
},
{
"name": "nodeemptydirvolume"
},
{
"name": "nodeCSIscheduling"
},
{
"name": "networkresource"
}
]
}
]
},
...