How to configure cluster nodes

Your cluster or multi-node appliance might contain nodes with different resources and different capacity. Therefore, each node can be configured separately.

Show node configuration

To display the current node configuration for a node (for example, lxd0), enter the following command:

amc node show <node>

This command will return output similar to the following:

name: lxd0
status: online
disk:
    size: 100GB
network:
    address: 10.119.216.34
    bridge-mtu: 1500
config:
    public-address: 10.119.216.34
    use-port-forwarding: true
    cpu-cores: 8
    cpu-allocation-rate: 4
    memory: 16GB
    memory-allocation-rate: 2
    gpu-slots: 10
    gpu-encoder-slots: 0
    tags: []

Configure allocation rates

AMS allows over-committing available resources on a node. This mechanism improves resource usage, because usually, instances don’t use 100% of their dedicated resources all of the time.

By default, AMS uses a CPU allocation rate of 4 and a memory allocation rate of 2. See Over-committing for more information.

You can configure the allocation rates with the cpu-allocation-rate and memory-allocation-rate configuration items.

Use the following commands to set the allocation rates on a node (for example, lxd0):

amc node set <node> cpu-allocation-rate <value>
amc node set <node> memory-allocation-rate <value>

Configure if a node can accept new instances

You can configure a node to stop accepting new instances. This is especially important in certain scenarios such as scaling down a LXD cluster. When you want to remove a node from the LXD cluster, the node must not have any instances. Hence, all running instances must be removed or disconnected and AMS must stop considering the node for new instances.

Use the following command to prevent the node from accepting new instances:

amc node set <node> unschedulable true

Configure GPU slots and GPU encoder slots

GPU slots are used to share GPUs amongst instances. See GPUs and instances and GPU slots for more information.

Each GPU-equipped cluster node is configured with a number of GPU slots and a number of GPU encoder slots. See Node-specific configuration for the default values that are used. Nodes without GPU are configured with 0 GPU slots and 0 GPU encoder slots.

Use the following commands to change the number of GPU slots and GPU encoder slots for a node:

amc node set <node> gpu-slots <number>
amc node set <node> gpu-encoder-slots <number>

Replace <node> with the node name (for example, lxd0), and <number> with the number of slots.

Use the following commands to set the number of GPU slots and GPU encoder slots for each GPU of a node:

amc node set <node>.gpus.<n>.slots <number>
amc node set <node>.gpus.<n>.encoder-slots <number>

Replace <node> with the node name (for example, lxd0), <n> with the GPU number and <number> with the number of slots.

Tags

A node can have a set of tags which can be used for different purposes. Use the following command to set the tags for a specific node:

amc node set <node> tags <comma-separated list of tags>