Explain how to use dedicated node groups with labels and taints

docs/_sidebar.md

@@ -64,6 +64,8 @@
   * [Ingress Nginx](/docs/ingress/ingress_nginx.md)
   * [Kube-vip](/docs/ingress/kube-vip.md)
   * [Metallb](/docs/ingress/metallb.md)
+* Manage Your Inventory
+  * [Dedicated Node Groups](/docs/manage_your_inventory/dedicated_node_groups.md)
 * Operating Systems
   * [Amazonlinux](/docs/operating_systems/amazonlinux.md)
   * [Bootstrap-os](/docs/operating_systems/bootstrap-os.md)

docs/manage_your_inventory/dedicated_node_groups.md

@@ -0,0 +1,20 @@
+# Dedicated node groups
+
+To reserved some nodes for particular workload (for instances, reserved GPU node for workload using them), you should
+define an Ansible group for theses nodes (statically or dynamically) and set taints and labels for these those in your inventory,
+using [Ansible group variable](https://docs.ansible.com/ansible/latest/inventory_guide/intro_inventory.html#assigning-a-variable-to-many-machines-group-variables).
+
+See below for concrete examples.
+
+## Nvidia GPU
+
+When using the Nvidia GPU support in kubespray, the nodes should have the following variables set:
+
+```yaml
+node_taints:
+- key: nvidia.com/gpu
+  effect: NoSchedule
+  ... (any other node taints)
+```
+
+TODO: add the same for labels once