Damien/arista-evpn-vxlan-clab

Fork 0

Files

Damien Arnodo e1fb8e27f5 Document all fixes applied in fix-bgp-and-mlag branch

2025-11-28 09:18:32 +00:00

6.0 KiB

Raw Blame History

Fixes Applied in fix-bgp-and-mlag Branch

This branch contains critical fixes discovered during lab testing to make the EVPN-VXLAN fabric functional.

🔧 Fixes Applied

1. Spine Switches - Enable IP Routing

Problem: BGP was disabled on spine switches with error "BGP is disabled for VRF default" and "IP routing not enabled"

Fix: Added ip routing command to both spine configurations

configs/spine1.cfg - Added line: ip routing (before service routing protocols model multi-agent)
configs/spine2.cfg - Added line: ip routing (before service routing protocols model multi-agent)

Impact: This enables BGP to function properly on spines, allowing:

Underlay BGP IPv4 Unicast sessions to establish
EVPN BGP sessions to establish
Route exchange between spines and leafs

2. Leaf Switches - MLAG Port-Channel Mode

Problem: LACP bonding (mode active) doesn't work properly in Alpine Linux containers due to lack of kernel module support

Fix: Changed from LACP to static LAG

Changed channel-group 1 mode active to channel-group 1 mode on in all leaf configs
This creates a static LAG that works in containerized environments

Status: ✅ Already applied in main branch (pushed by user)

3. Leaf Switches - Port-Channel Switchport Mode

Problem: Port-Channel configured as trunk, but Alpine containers send untagged traffic

Fix Needed: Change Port-Channel1 from trunk to access mode on all leafs:

interface Port-Channel1
   switchport mode access
   switchport access vlan 40  # or appropriate VLAN for each VTEP

Status: ⚠️ NOT YET APPLIED - Needs manual configuration or config file update

4. Host Configuration - Simplified Bonding

Problem: Alpine Linux containers cannot properly configure 802.3ad LACP bonding

Fix in topology: Remove bonding complexity, use single interface:

host1:
  exec:
    - ip addr add 10.40.40.101/24 dev eth1
    - ip link set eth1 up

Status: ⚠️ NOT YET APPLIED - topology file not updated in this branch

📋 Summary of Issues Found

Issue #1: Missing `ip routing` on Spines

Symptoms:
- show ip bgp summary returned "BGP is disabled for VRF default"
- Attempting to configure BGP showed "! IP routing not enabled"
Root Cause: Arista EOS requires explicit ip routing command to enable L3 functionality
Status: ✅ FIXED

Issue #2: LACP Bonding in Containers

Symptoms:
- Port-Channel showing "waiting for LACP response"
- Host bond interface in DOWN state
Root Cause: Alpine containers don't have bonding kernel modules
Status: ✅ FIXED (by changing to static LAG)

Issue #3: Trunk vs Access Mode

Symptoms:
- No MAC learning on switch
- Port-Channel counters showed traffic but no unicast packets
Root Cause: Hosts send untagged traffic, switch expects tagged (trunk mode)
Status: ⚠️ NEEDS MANUAL FIX

🚀 Deployment Instructions

Option 1: Deploy with Manual Post-Configuration

Deploy the lab:

cd ~/arista-evpn-vxlan-clab
git checkout fix-bgp-and-mlag
sudo containerlab deploy -t evpn-lab.clab.yml

Fix Port-Channel mode on all leafs (manual):

for leaf in leaf1 leaf2 leaf3 leaf4 leaf5 leaf6 leaf7 leaf8; do
  ssh admin@clab-arista-evpn-fabric-$leaf << 'EOF'
enable
configure terminal
interface Port-Channel1
   switchport mode access
   switchport access vlan 40
write memory
EOF
done

Configure hosts (manual):

# Host1 (VLAN 40 - L2 VXLAN)
docker exec clab-arista-evpn-fabric-host1 sh -c '
ip link set bond0 down 2>/dev/null
ip link del bond0 2>/dev/null
ip addr flush dev eth1
ip addr add 10.40.40.101/24 dev eth1
ip link set eth1 up
'

# Host3 (VLAN 40 - L2 VXLAN)
docker exec clab-arista-evpn-fabric-host3 sh -c '
ip link set bond0 down 2>/dev/null
ip link del bond0 2>/dev/null
ip addr flush dev eth1
ip addr add 10.40.40.103/24 dev eth1
ip link set eth1 up
'

# Host2 (VRF gold - L3 VXLAN)
docker exec clab-arista-evpn-fabric-host2 sh -c '
ip link set bond0 down 2>/dev/null
ip link del bond0 2>/dev/null
ip addr flush dev eth1
ip addr add 10.34.34.102/24 dev eth1
ip link set eth1 up
ip route add default via 10.34.34.1
'

# Host4 (VRF gold - L3 VXLAN)
docker exec clab-arista-evpn-fabric-host4 sh -c '
ip link set bond0 down 2>/dev/null
ip link del bond0 2>/dev/null
ip addr flush dev eth1
ip addr add 10.78.78.104/24 dev eth1
ip link set eth1 up
ip route add default via 10.78.78.1
'

Verify:

# Check BGP
ssh admin@clab-arista-evpn-fabric-leaf1 "show bgp evpn summary"

# Check VXLAN
ssh admin@clab-arista-evpn-fabric-leaf1 "show vxlan vtep"

# Test connectivity
docker exec -it clab-arista-evpn-fabric-host1 ping -c 4 10.40.40.103
docker exec -it clab-arista-evpn-fabric-host2 ping -c 4 10.78.78.104

Option 2: Wait for Complete Fix

A complete fix will require:

✅ Spine configs updated (DONE)
⏳ All leaf Port-Channel configs updated to access mode
⏳ Topology file updated to simplify host networking
⏳ README updated with correct testing procedures

🧪 Testing Results

After applying fixes manually:

✅ BGP underlay sessions establish (eBGP between spine-leaf, iBGP between MLAG pairs)
✅ BGP EVPN overlay sessions establish
✅ MLAG pairs form correctly (active-full, up/up)
✅ MAC addresses learned locally on leaf switches
✅ EVPN Type-2 routes advertised (pending overlay establishment)
⏳ End-to-end connectivity (requires all fixes applied)

📝 Notes

The ip routing fix is critical and must be in the startup-config for clean deployments
Static LAG (mode on) is more reliable than LACP in containerized environments
Access mode port-channels work better with simple Alpine containers
For production environments with proper bonding support, LACP can be re-enabled

Spine BGP not starting: Missing ip routing command
MLAG port-channels not forming: LACP bonding incompatibility
No MAC learning: Trunk vs access mode mismatch

6.0 KiB Raw Blame History