# Groups of Lab Devices (Nodes)

Network devices (nodes) used in a virtual lab are automatically grouped into platform-specific groups like *iosv* or *nxos*. You can use these groups or custom groups to:

* Set node attributes or Ansible group variables
* Set device type or [configuration modules](module-reference.md) for a set of devices
* Attach additional configuration templates to a set of devices
* Limit the scope of **[netlab](netlab/cli.md)** commands that invoke Ansible playbooks. 

For example, you could use `netlab config mpls.j2 --limit iosv` to deploy configuration specified in Jinja2 template `mpls.j2` only on Cisco IOSv devices.

```eval_rst
.. contents:: Table of Contents
   :depth: 2
   :local:
   :backlinks: none
```

## Custom Node Groups

Virtual lab topology can specify additional (custom) device groups in the **groups** top-level parameter. The **groups** parameter must be a dictionary with group names as dictionary keys. Dictionary values could be either a list of member nodes or a further dictionary specifying **members**:


```
---
defaults.device: iosv
  
nodes: [ a,b,c,d,e,f ]

groups:
  g1: [ a,b,c ]
  g2:
    members: [ d,e,f ]
```

The custom node groups are used to create additional groups in the Ansible inventory file. You can use custom node groups in any **netlab** command that invokes an Ansible playbook.

For example, `netlab config mpls.j2 --limit g1` would deploy configuration template `mpls.j2` only on lab devices A, B, and C.

(group-special-names)=
### Special Group Names

The following groups have special meaning in *netlab*-generated Ansible inventory:

* `unprovisioned`: **netlab up** and **netlab initial** will skip devices in this group while deploying device configurations.

(default-groups)=
### Default Groups

You can specify system-wide or [project-wide](defaults-locations) groups in user- or system defaults file(s).

As expected, the default group settings are merged with the lab topology groups, and you can even use them to create new groups not defined in the lab topology.

The only peculiarity of the default groups is the handling of group members:

* Default groups can have a list of **members**
* A non-existent node in a group **members** list usually generates an error message. For default groups, such a member is silently removed from the **members** list[^DGM] (but see also [](groups-auto-create))
* The members of a default group are copied into the lab topology group only when the corresponding lab topology group has no members.

[^DGM]: That makes it possible to have default group members not present in the lab topology.

(group-node-data)=
## Setting Node Data in Groups

Sometimes, you'd like to set a node attribute for all group members. For example, in a BGP anycast scenario, we should set **[bgp.advertise_loopback](module/bgp.md#advertised-bgp-prefixes)** to *false* on all anycast server -- they should advertise only the anycast prefix, not individual loopback prefixes. 

While it's perfectly OK to set the desired attribute(s) on individual nodes, it's much more convenient to set them in a group definition -- you can use any valid node attribute (including configuration module node attributes[^MNA]) in group definitions[^NDT].

[^MNA]: If a group definition contains a **module** attribute, you cannot set attributes for modules not listed in the **module** attribute in the group definition.
 
[^NDT]: Node attributes were stored in **node_data** group attribute prior to *netlab* release 1.4. While you can still use the **node_data** dictionary, it has a few undesired side effects that we'll not fix. It is deprecated in release 1.8; using it will generate a warning during the data transformation phase.

The node group attribute will be set on all members of the group. The data is [deep-merged](defaults.md#deep-merging) with the existing node data -- for example, you could set **bgp.advertise_loopback** attribute in group definition without affecting **bgp.as** node attribute[^NDGP].

[^NDGP]: In a topology with hierarchical groups, attributes from the innermost groups take precedence. Node attributes from groups with static members have precedence over node attributes from BGP-generated groups.

Using this functionality, a BGP anycast topology file becomes much more concise than it would have been otherwise:

```
defaults:
  device: iosv

module: [ bgp, ospf ]

bgp.as: 65000

groups:
  anycast:
    members: [ a1, a2, a3 ]
    bgp.as: 65001
    bgp.advertise_loopback: false

nodes: [ l1, l2, l3, s1, a1, a2, a3 ]
```

### Using Group Node Data with VRFs and VLANs

VRFs and VLANs mentioned in **vrfs** or **vlans** group attributes will be defined as global (topology-wide) VRFs/VLANs. The VLAN ID/VNI or VRF RT/RD values will be copied from **vlans**/**vrfs** into global **vlans**/**vrfs**. Every VLAN  needs a unique ID/VNI (likewise for VRF RT/RD), so you cannot define different ID/VNI or RT/RD values for the same VLAN/VRF in different groups.

Example:

```
module: [ vlan,ospf ]

groups:
  g1:
    members: [ r1, r2 ]
    vlans:
      red:
        ospf.cost: 10
      blue:
        ospf.cost: 20

nodes: [r1, r2]

links:
- r1:
  r2:
  vlan.trunk: [ red, blue ]
```

The above topology will:

* Create topology-wide *red* and *blue* VLANs.
* Auto-assign VLAN ID and VNI to those VLANs.
* Copy group **vlans** into R1 and R2 (setting OSPF cost for VLAN interfaces)
* Merge the global **vlans** definitions into **nodes.r1.vlans** and **nodes.r2.vlans**, ensuring the VLANs on R1 and R2 have the correct VLAN ID/VNI.

```{tip}
As the group VLANs/VRFs are copied into all nodes in a group, you'll get all VLANs/VRFs (and VLAN interfaces) mentioned in the group definition defined on all group members regardless of whether they use those VLANs/VRFs.
```

## Setting Device Type or List of Modules in Groups

You can set node device type (**device** attribute) or the list of configuration modules (**module** attribute) in group definitions, but only on groups with static members.

The device type is copied from groups to nodes with no **device** attribute. Modules listed in a group are added to the node **module** attribute. The merging of node- and group modules takes precedence over the global (topology-level) list of modules.

The following example uses this functionality to:

* Use Cumulus VX on routers advertising an anycast IP address,
* Specify BGP as the only configuration module on those devices.

```
defaults.device: iosv

module: [ bgp, ospf ]
bgp.as: 65000

groups:
  anycast:
    members: [ a1, a2, a3 ]
    module: [ bgp ]
    device: cumulus
    bgp.as: 65001
    bgp.advertise_loopback: false

nodes: [ l1, l2, l3, s1, a1, a2, a3 ]
```

**Notes:**
* The BGP module specified in the **anycast** group is added to the list of modules specified in the group members. No group members have an explicit module definition, resulting in `module: [ bgp ]` being set on A1, A2, and A3.
* Device type specified in the **anycast** group is copied into A1, A2, and A3.
* Default device type specified in **defaults.device** is copied into nodes that still have no device type (L1, L2, L3, S1)
* Default list of modules (`module: [ bgp, ospf ]`) is copied into nodes that still have no **module** attribute (L1, L2, L3, S1).

(group-ansible-vars)
## Ansible Group Variables

Group definition could include Ansible group inventory variables in the **vars** element, a dictionary of name/value pairs. The following example creates two groups (g1 and g2) and sets Ansible group variables on g2.

```
---
defaults.device: iosv
  
nodes: [ a,b,c,d,e,f ]

groups:
  g1: [ a,b,c ]
  g2:
    members: [ d,e,f ]
    vars:
      x1: 1
      x2: 2
```

Group variables are stored in **group_vars** directory when **[netlab create](netlab/create.md)** creates an Ansible inventory from the topology file. The Ansible inventory group variables can then be used in Ansible playbooks and related Jinja2 templates, for example, in the [custom configuration templates](custom-config-templates.md) deployed with the **[netlab initial](netlab/initial.md)** or **[netlab config](netlab/config.md)** commands.

```{warning}
* Use this functionality only when you need custom attributes in Jinja2 templates but don't want to specify them as [valid node attributes](extend-attributes.md).
* You cannot use Ansible group variables to overwrite node data specified in a custom group. The [group node data](group-node-data) is copied into the node data and stored in Ansible host variables.
* Ansible inventory is not used in the **[‌netlab connect](netlab/connect.md)** command. To overwrite connection-specific variables (connection method, username, password), specify `ansible_something` or `netlab_something` variables in node data. See [Node Attributes](node-ansible-data) for more details.
```

### Changing Group Variables for Predefined Groups

If you want to set one or more Ansible facts for all devices in your lab, use the **vars** element in the **all** group:

```
defaults.device: cumulus

nodes: [ a,b,c ]

groups:
  all:
    vars:
      http_server: true
```

To make the definition a bit more concise, use the *netlab* hierarchical attribute names:

```
defaults.device: cumulus

nodes: [ a,b,c ]

groups.all.vars.http_server: true
```

You could also change group variables for auto-created device-specific groups. For example, to change the default username used to connect to Cumulus VX nodes, use:

```
defaults.device: cumulus

nodes: [ a,b,c ]

groups.cumulus.vars.ansible_user: other
```

## Automatic BGP Groups

The BGP module creates a group named *asnnn* where *nnn* is the AS number for every BGP AS present in the lab topology. The members of the group are all nodes in that autonomous system.

You can set inventory variables (with **vars** attribute), deployment templates (with **config** attribute), or node data on an automatic BGP group, but you cannot specify static group members.

Here is a BGP anycast topology file that depends on setting node data within an automatic BGP group (the topology file uses [BGP as-list](module/bgp.md#global-bgp-configuration-parameters) functionality to specify AS membership):

```
defaults:
  device: iosv

module: [ bgp, ospf ]

bgp:
  as_list:
    65000:
      members: [ l1, l2, l3, s1 ]
      rr: [ s1 ]
    65001:
      members: [ a1, a2, a3 ]

groups:
  as65001:
    bgp.advertise_loopback: false

nodes: [ l1, l2, l3, s1, a1, a2, a3 ]
```

## Specifying Groups in Nodes

You could specify a group or a list of groups a node belongs to in the **group** attribute of a node instead of specifying group members in the **groups** topology attribute. You can also combine the two ways of defining groups, for example:

```
defaults:
  device: cumulus
  
nodes:
  a:
    group: [ g2 ]
  b:
  c:
  d:
  e:
    group: [ g2, g3 ]
  f:
    group: g1 

groups:
  g1: [ a,b,c ]
  g2:
    vars:
      x1: 1
      x2: 2
```

The above topology file:

* Creates two groups (g1 and g2) from the **groups** element.
* Adds nodes A and E to group g2. The **members** list is automatically created when the first node is added to the group.
* Adds node E to group g3. The group is automatically created the first time a node **group** list mentions it.
* Adds F to group g1. Please note that the **group** node attribute could be a string or a list of strings.

The final value of the **groups** element is thus:

```
groups:
  g1:
    members:
    - a
    - b
    - c
    - f
  g2:
    members:
    - a
    - e
    vars:
      x1: 1
      x2: 2
  g3:
    members:
    - e
```

## Hierarchical Groups

*netlab* supports _hierarchical groups_ -- groups could be members of other groups, for example:

```
nodes: [ a,b,c,d,e,f ]

groups:
  g1:
    members: [ a,b ]
  g2:
    members: [ d,g1,g3 ]
  g3:
    members: [ e ]
```

The hierarchical groups specified in a lab topology file are directly translated into Ansible inventory groups. The above example will generate the following data structure for group **g2** in the Ansible inventory file:

```
g2:
  children:
    g1: {}
    g3: {}
  hosts:
    d: {}
```

### Node Data in Hierarchical Groups

When faced with a group hierarchy, processing of node data takes great care to use the node values specified in the most specific group (see also [](custom-config-groups))

Continuing the previous example, now with node data in groups[^FINV]:

[^FINV]: Please note that **foo** is not a valid node attribute.

```
groups:
  g1: 
    members: [ a,b ]
  g2:
    members: [ d,g1,g3 ]
    foo: bar
  g3:
    members: [ e ]
    foo: baz
```

* Nodes **a**, **b** and **d** (direct and indirect members of group **g2**) will have the node attribute **foo** set to **bar**.
* Node **e** (member of group **g3**) will have the node attribute **foo** set to **baz** -- **g3** overwrites the **foo** value set by the parent group **g2**.

Group node data processing performs [deep dictionary merge](defaults.md#deep-merging) when an attribute specified in the group ‌and the current value of the node attribute are both dictionaries, allowing you to specify various parts of the same dictionary in different groups, for example:

```
nodes:
  a:
    bgp:
      rr: true
  b:
  c:
  d:
  e:
    config: [ e ]
  f:

groups:
  g1: [ a,b ]
  g2:
    members: [ d,g1,g3 ]
    bgp:
      as: 65000
  g3:
    members: [ e ]
    bgp:
      as: 65001
```

* Node **a** has **bgp.rr** set to *True* (direct node data) and **bgp.as** set to 65000 (attribute merged from **g2** node data).
* Nodes **b** and **d** have **bgp.as** set to 65000.
* Node **e** has **bgp.as** set to 65001 (deep merge results in value from **g2** being overwritten by value from **g3**).
* Nodes **c** and **f** do not have any BGP-related attributes

(groups-auto-create)=
## Create Nodes From Group Members

The group **members** attribute must contain valid node names specified in the **nodes** dictionary to prevent typos and duplicate names. However, if a group contains **\_auto\_create**  attribute set to *True*, _netlab_ creates missing nodes from group members. You can set the **\_auto\_create** attribute:

* In individual groups. You can specify the **\_auto\_create** attribute in individual default groups to create nodes in all labs using those defaults.
* In **groups** or **defaults.groups** dictionary. The global **\_auto\_create**  attribute does not apply to default groups.

For example, the following topology creates nodes from the members of all topology groups (due to global **\_auto\_create** attribute) and from the members of the **g2** default group:

```
defaults:
  device: cumulus
  
defaults.groups:
  g1:
    members: [ A, B ]
  g2:
    _auto_create: True
    members: [ C, D ]
    device: eos

groups:
  _auto_create: True
  g3: [ E, F ]
```

The lab will contain nodes **C** and **D** running Arista EOS and nodes **E** and **F** running Cumulus Linux.