IP Addressing in AVD

Introduction

AVD provides a powerful automatic IP addressing system that assigns IP addresses to fabric devices based on pools and node identifiers. This eliminates manual IP planning and ensures consistent, predictable addressing across your network.

This guide covers IP pool types, allocation mechanisms, and customization options available in AVD.

Key Concepts

Before diving into specifics, understand these core concepts:

IP Pool: A range of IP addresses from which AVD allocates individual addresses

Node ID: A unique numeric identifier for each device, used to calculate IP offsets

Offset: A value added to the base pool address to derive a specific IP

Prefix Length: The subnet mask size for allocated addresses (e.g., /32, /31)

IP Pool Types

AVD uses several IP pools for different purposes:

Pool Variable	Purpose	Default Interface
loopback_ipv4_pool	Router ID and BGP peering	Loopback0
vtep_loopback_ipv4_pool	VXLAN tunnel endpoints	Loopback1
uplink_ipv4_pool	P2P links between devices	Ethernet uplinks
mlag_peer_ipv4_pool	MLAG peer-link SVI	VLAN 4094
mlag_peer_l3_ipv4_pool	MLAG L3 iBGP peering	VLAN 4093
router_id_pool	BGP Router ID only (IPv6 underlay)	None (ID only)

Pool Hierarchy

Pools can be defined at multiple levels with the following precedence (highest to lowest):

1. Node type defaults - Applied to all nodes of a type
2. Node group level - Shared by devices in a group (e.g., MLAG pair)
3. Node level - Specific to a single device

Pool Hierarchy Example

---
l3leaf:
  defaults:
    platform: vEOS-lab
    loopback_ipv4_pool: 10.255.0.0/27
    loopback_ipv4_offset: 2
    vtep_loopback_ipv4_pool: 10.255.3.0/27
    uplink_switches: ['htipa-spine1', 'htipa-spine2']
    uplink_ipv4_pool: 10.255.255.0/26
    mlag_peer_ipv4_pool: 10.255.3.64/27
    mlag_peer_l3_ipv4_pool: 10.255.3.96/27
    virtual_router_mac_address: 00:1c:73:00:00:99
    spanning_tree_priority: 4096
    spanning_tree_mode: mstp

  node_groups:
    - group: HTIPA_L3_LEAFS
      bgp_as: 65101
      loopback_ipv4_pool: 10.255.1.0/24
      nodes:
        - name: htipa-leaf1
          id: 1
          mgmt_ip: 172.16.2.101/24
          uplink_switch_interfaces: [Ethernet1, Ethernet1]
          loopback_ipv4_pool: 10.255.2.0/24
        - name: htipa-leaf2
          id: 2
          uplink_switch_interfaces: [Ethernet2, Ethernet2]
          mgmt_ip: 172.16.2.102/24

Loopback IP Allocation

Basic Formula

For Loopback0 (Router ID):

IP = pool_base + node_id + loopback_ipv4_offset

Note

When spines and leafs share the same pool, use loopback_ipv4_offset to prevent IP conflicts

Example

Spine assignment:

• spine1 will be assigned 10.255.0.1/32 - calculation (node type default pool + id[1])
• spine2 will get 10.255.0.2/32 - calculation (node type default pool + id[2])

Loopback0 for spines

---
spine:
  defaults:
    platform: vEOS-lab
    loopback_ipv4_pool: 10.255.0.0/27
    bgp_as: 65100
  nodes:
    - name: htipa-spine1
      id: 1
      mgmt_ip: 172.16.2.11/24
    - name: htipa-spine2
      id: 2
      mgmt_ip: 172.16.2.12/24

spine1 configuration

interface Loopback0
   description ROUTER_ID
   no shutdown
   ip address 10.255.0.1/32
!

spine2 configuration

interface Loopback0
   description ROUTER_ID
   no shutdown
   ip address 10.255.0.2/32
!

Leaf assignment:

• leaf1 will get 10.255.2.3/32 (node specific pool + id[1] + offset[2])
• leaf2 will get 10.255.1.4/32 (group specific pool + id[2] + offset[(2])

Loopback0 for leafs

---
l3leaf:
  defaults:
    platform: vEOS-lab
    loopback_ipv4_pool: 10.255.0.0/27
    loopback_ipv4_offset: 2
    vtep_loopback_ipv4_pool: 10.255.3.0/27
    uplink_switches: ['htipa-spine1', 'htipa-spine2']
    uplink_ipv4_pool: 10.255.255.0/26
    mlag_peer_ipv4_pool: 10.255.3.64/27
    mlag_peer_l3_ipv4_pool: 10.255.3.96/27
    virtual_router_mac_address: 00:1c:73:00:00:99
    spanning_tree_priority: 4096
    spanning_tree_mode: mstp

  node_groups:
    - group: HTIPA_L3_LEAFS
      bgp_as: 65101
      loopback_ipv4_pool: 10.255.1.0/24
      nodes:
        - name: htipa-leaf1
          id: 1
          mgmt_ip: 172.16.2.101/24
          uplink_switch_interfaces: [Ethernet1, Ethernet1]
          loopback_ipv4_pool: 10.255.2.0/24
        - name: htipa-leaf2
          id: 2
          uplink_switch_interfaces: [Ethernet2, Ethernet2]
          mgmt_ip: 172.16.2.102/24

leaf1 configuration

interface Loopback0
   description ROUTER_ID
   no shutdown
   ip address 10.255.2.3/32
!

leaf2 configuration

interface Loopback0
   description ROUTER_ID
   no shutdown
   ip address 10.255.1.4/32
!

VTEP Loopback Allocation

VTEP loopbacks (Loopback1) use vtep_loopback_ipv4_pool:

VTEP Loopback for leafs

---
l3leaf:
  defaults:
    platform: vEOS-lab
    loopback_ipv4_pool: 10.255.0.0/27
    loopback_ipv4_offset: 2
    vtep_loopback_ipv4_pool: 10.255.3.0/27
    uplink_switches: ['htipa-spine1', 'htipa-spine2']
    uplink_ipv4_pool: 10.255.255.0/26
    mlag_peer_ipv4_pool: 10.255.3.64/27
    mlag_peer_l3_ipv4_pool: 10.255.3.96/27
    virtual_router_mac_address: 00:1c:73:00:00:99
    spanning_tree_priority: 4096
    spanning_tree_mode: mstp

  node_groups:
    - group: HTIPA_L3_LEAFS
      bgp_as: 65101
      loopback_ipv4_pool: 10.255.1.0/24
      nodes:
        - name: htipa-leaf1
          id: 1
          mgmt_ip: 172.16.2.101/24
          uplink_switch_interfaces: [Ethernet1, Ethernet1]
          loopback_ipv4_pool: 10.255.2.0/24
        - name: htipa-leaf2
          id: 2
          uplink_switch_interfaces: [Ethernet2, Ethernet2]
          mgmt_ip: 172.16.2.102/24

MLAG VTEP Sharing

MLAG pairs share the same VTEP IP. AVD automatically uses the MLAG primary ID in the group for both peers. This translates to IP = pool + mlag_primary_id + loopback_ipv4_offset. Non-MLAG nodes use their own ID.

P2P Uplink Allocation

Uplink IP addresses are calculated using a more complex formula to ensure unique /31 subnets for each link.

Formula

subnet_offset = ([node_id - 1] * max_uplink_switches * max_parallel_uplinks) + uplink_switch_index

Where:

• node_id: The leaf’s ID
• max_uplink_switches: Maximum number of uplink switches (default: length of uplink_switches)
• max_parallel_uplinks: Maximum parallel uplinks per switch (default: 1)
• uplink_switch_index: Index of the uplink switch (0-based)

Example

P2P Uplink for leafs

---
l3leaf:
  defaults:
    platform: vEOS-lab
    loopback_ipv4_pool: 10.255.0.0/27
    loopback_ipv4_offset: 2
    vtep_loopback_ipv4_pool: 10.255.3.0/27
    uplink_switches: ['htipa-spine1', 'htipa-spine2']
    uplink_ipv4_pool: 10.255.255.0/26
    mlag_peer_ipv4_pool: 10.255.3.64/27
    mlag_peer_l3_ipv4_pool: 10.255.3.96/27
    virtual_router_mac_address: 00:1c:73:00:00:99
    spanning_tree_priority: 4096
    spanning_tree_mode: mstp

  node_groups:
    - group: HTIPA_L3_LEAFS
      bgp_as: 65101
      loopback_ipv4_pool: 10.255.1.0/24
      nodes:
        - name: htipa-leaf1
          id: 1
          mgmt_ip: 172.16.2.101/24
          uplink_switch_interfaces: [Ethernet1, Ethernet1]
          loopback_ipv4_pool: 10.255.2.0/24
        - name: htipa-leaf2
          id: 2
          uplink_switch_interfaces: [Ethernet2, Ethernet2]
          mgmt_ip: 172.16.2.102/24

Resulting allocations for leaf1 (id=1):

Uplink	Subnet Offset	Leaf IP	Spine IP
To spine1	0	10.255.255.1/31	10.255.255.0/31
To spine2	1	10.255.255.3/31	10.255.255.2/31

leaf1 P2P to spines

interface Ethernet1
   description P2P_htipa-spine1_Ethernet1
   no shutdown
   mtu 1500
   no switchport
   ip address 10.255.255.1/31
!
interface Ethernet2
   description P2P_htipa-spine2_Ethernet1
   no shutdown
   mtu 1500
   no switchport
   ip address 10.255.255.3/31
!

Downlink Pools

While uplink_ipv4_pool is defined on the downstream device (leaf), downlink_pools provides an alternative approach where IP pools are defined on the upstream device (spine). This is useful when you want centralized IP management on parent switches.

Mutually Exclusive

downlink_pools on a parent switch cannot be combined with uplink_ipv4_pool on the child switch. Use one approach or the other.

Configuration

Define downlink_pools on the parent switch (spine) with a list of pools mapped to specific downlink interfaces:

Downlink Pools on Spine

spine:
  defaults:
    loopback_ipv4_pool: 192.168.0.0/24
    bgp_as: 65000
  nodes:
    - name: spine1
      id: 1
      downlink_pools:
        - ipv4_pool: 10.0.1.0/24
          downlink_interfaces: [Ethernet3-6]  # (1)!
        - ipv4_pool: 10.0.3.0/24
          downlink_interfaces: [Ethernet7-14]  # (2)!

1. First pool for interfaces Ethernet3 through Ethernet6
2. Second pool for interfaces Ethernet7 through Ethernet14

IP Allocation

The IP address is derived based on the interface’s index position in the downlink_interfaces list:

Interface	Index	Subnet from Pool
Ethernet3	0	10.0.1.0/31
Ethernet4	1	10.0.1.2/31
Ethernet5	2	10.0.1.4/31
Ethernet6	3	10.0.1.6/31

The spine gets the even IP (.0, .2, .4) and the leaf gets the odd IP (.1, .3, .5) in each /31 subnet.

When to Use Downlink Pools

Use Case	Recommended Approach
Leaf-centric IP management	uplink_ipv4_pool on leafs
Spine-centric IP management	downlink_pools on spines
Different pools per spine port range	downlink_pools with multiple entries
Simple uniform allocation	uplink_ipv4_pool on leafs

MLAG IP Allocation

MLAG requires two pools for peer connectivity:

Pool	Purpose	Default VLAN
mlag_peer_ipv4_pool	L2 peer-link SVI	4094
mlag_peer_l3_ipv4_pool	L3 iBGP peering	4093

MLAG Allocation Algorithms

AVD supports three MLAG IP allocation algorithms configured via fabric_ip_addressing.mlag.algorithm:

first_id (default)odd_idsame_subnet

Uses the first node’s ID in the MLAG group:

fabric_ip_addressing:
  mlag:
    algorithm: first_id  # Default

l3leaf:
  defaults:
    mlag_peer_ipv4_pool: 10.255.3.64/27
  node_groups:
    - group: HOW_TO_L3_LEAFS
      nodes:
        - name: htipa-leaf1
          id: 1  # Primary: 10.255.3.64/31
        - name: htipa-leaf2
          id: 2  # Secondary: 10.255.3.65/31

Formula: offset = (mlag_primary_id - 1)

Requires one node with an odd ID and one with an even ID:

fabric_ip_addressing:
  mlag:
    algorithm: odd_id

l3leaf:
  node_groups:
    - group: HOW_TO_L3_LEAFS
      nodes:
        - name: htipa-leaf1
          id: 1  # Odd - determines subnet
        - name: htipa-leaf2
          id: 2  # Even - must pair with odd

Formula: offset = (odd_id - 1)

All MLAG pairs share the same subnet (first in pool):

fabric_ip_addressing:
  mlag:
    algorithm: same_subnet

# All MLAG pairs get 10.255.3.64/31

Formula: offset = 0 (always)

Node ID Assignment

Node IDs are critical for IP allocation. AVD supports two assignment methods:

Static Assignment

Manually assign IDs to each node (default):

spine:
  nodes:
    - name: htipa-spine1
      id: 1
    - name: htipa-spine2
      id: 2

Dynamic Assignment

Automatically assign IDs based on fabric topology:

fabric_numbering:
  node_id:
    algorithm: pool_manager
    pools_file: intended/data/fabric-ids.yml

IDs are assigned based on: fabric_name, dc_name, pod_name, type, and rack.

Pool Formats

AVD supports flexible pool formats:

Single Subnet

loopback_ipv4_pool: 10.255.0.0/24

Multiple Subnets

Comma-separated list of subnets:

loopback_ipv4_pool: 10.255.0.0/25, 10.255.1.0/25

IP Ranges

loopback_ipv4_pool: 10.255.0.1-10.255.0.50

Static IP Overrides

Override any pool-calculated address with a static value:

l3leaf:
  nodes:
    - name: htipa-leaf1
      id: 1
      loopback_ipv4_address: 10.100.100.1  # Override loopback pool
      vtep_loopback_ipv4_address: 10.100.101.1  # Override VTEP pool

Available override variables:

Override Variable	Overrides Pool
loopback_ipv4_address	loopback_ipv4_pool
vtep_loopback_ipv4_address	vtep_loopback_ipv4_pool

Global IP Addressing Settings

Configure fabric-wide IP addressing behavior:

fabric_ip_addressing:
  mlag:
    algorithm: first_id  # first_id, odd_id, same_subnet
    ipv4_prefix_length: 31
  p2p_uplinks:
    ipv4_prefix_length: 31

Custom IP Addressing

For complex requirements, create a custom Python module:

First, create the custom node type keys and set your custom Python module and class name:

Custom Node Type with Custom IP Addressing

---
custom_node_type_keys:
  - key: custom_l3leaf
    type: custom_l3leaf
    connected_endpoints: true
    default_evpn_role: client
    mlag_support: true
    network_services:
      l2: true
      l3: true
    vtep: true
    default_ptp_priority1: 30
    cv_tags_topology_type: leaf
    ip_addressing:
      python_module: custom_modules.avd_overrides
      python_class_name: MyCustomIpAddressing

# Optional: pass custom variables to your module
custom_ip_offset: 25

Second, create your custom python module and class:

Custom IP Addressing Module

# Copyright (c) 2026 Arista Networks, Inc.
# Use of this source code is governed by the Apache License 2.0
# that can be found in the LICENSE file.
# custom_modules/avd_overrides.py
from functools import cached_property

from pyavd.api.ip_addressing import AvdIpAddressing


class MyCustomIpAddressing(AvdIpAddressing):
    """Custom IP addressing that adds an offset to all loopback IPs."""

    @cached_property
    def _custom_offset(self) -> int:
        """Read a custom offset from hostvars."""
        return int(self._hostvars.get("custom_ip_offset", 0))

    def router_id(self) -> str:
        """Override router_id to add custom offset."""
        offset = self._id + self._loopback_ipv4_offset + self._custom_offset
        return self._ip(self._loopback_ipv4_pool, 32, offset, 0)

    def vtep_ip(self) -> str:
        """Override VTEP IP with custom offset."""
        offset = self._id + self._loopback_ipv4_offset + self._custom_offset
        return self._ip(self._vtep_loopback_ipv4_pool, 32, offset, 0)

Note

The custom python module must be in the PYTHONPATH for Ansible to find it. export PYTHONPATH="${PYTHONPATH}:$(pwd)/custom_modules"

For a list of available methods, refer to the pyavd documentation.

Best Practices

1. Plan your ID scheme: Use consistent ID numbering across the fabric
2. Use offsets wisely: When sharing pools between node types, use loopback_ipv4_offset
3. Size pools appropriately: Ensure pools have enough addresses for growth
4. Document your allocation scheme: Keep records of pool assignments
5. Use node groups for MLAG: Define MLAG pairs as node groups for automatic VTEP sharing
6. Consider the algorithm: Choose the MLAG algorithm that fits your operational model

Troubleshooting

Common Issues

Issue	Cause	Solution
IP conflict between spine and leaf	Shared pool without offset	Add loopback_ipv4_offset to leaf defaults
Missing MLAG peer IP	mlag_peer_ipv4_pool not defined	Define pool at node_group or defaults level
Unexpected VTEP IP	MLAG pair not in same node_group	Move MLAG peers to same node_group
Pool exhausted	Too many nodes for pool size	Use larger pool or multiple subnets
Loopback IP not updating	Static IP override in use or more specific pools	Remove override or check pool definitions
Duplicate IP addresses on uplinks	Leafs have different numbers of uplinks	Set max_uplink_switches and/or max_parallel_uplinks to reserve IP space.

Reference

• Node Type VTEP and Loopback Configuration
• Fabric IP Addressing
• MLAG Configuration
• Custom IP Addressing