Platform Caveats¶
Arista EOS¶
Routed VLANs cannot be used in EVPN MPLS VLAN bundles
Arista EOS uses an invalid value for the suboption 150 of the DHCP option 82 when doing inter-VRF DHCPv4 relaying.
The DHCP client on Arista EOS is finicky. Management interface might lose its IPv4 address when the DHCP state changes on one of the data-plane Ethernet interfaces.
The following features do not work on Arista cEOS Ethernet interfaces:
Anycast gateways
DHCP and DHCPv6 clients
cEOS MPLS data plane was introduced in release 4.32.1F.
Aruba AOS-CX¶
Ansible automation of Aruba AOS-CX requires the installation of the ArubaNetworks Ansible Collection with
ansible-galaxy collection install arubanetworks.aoscx
.The limitations of the Aruba AOS-CX Simulator can be found here, selecting CX Simulator as platform.
It seems that the Aruba AOS-CX Simulator cannot generate ICMP Fragmentation-Needed messages.
VRF and L3VPN Caveats¶
OSPF processes can be only 1-63. VRF indexes usually are > 100, so a device tweak will map every vrfidx to a different OSPF process ID. That means you cannot have more than 62 VRF using OSPF.
On the Aruba AOS-CX Virtual version 10.11.0001, the MPLS L3VPN forwarding plane seems broken (while the control plane works fine).
VXLAN and EVPN Caveats¶
ECMP is not supported.
The VXLAN data plane (at least, on the virtual version) does not support VNI greater than 65535. If you set a higher value, an overflow will occur, and you may have overlapping VNIs. The workaround for this is to set
defaults.vxlan.start_vni: 20000
anddefaults.evpn.start_transit_vni: 10000
(especially on multi-vendor topologies).EVPN Symmetric IRB is supported only from the Aruba AOS-CX Virtual version 10.13.
CPU-generated traffic is not encapsulated in Symmetric IRB on the AOS-CX Simulator.
Active-Gateway MAC Addresses shall be the same across all VTEPs in AOS-CX Simulator.
Manually Setting Virtual-MAC for EVPN RT-5 is not working (but an autogenerated one is fine).
AOS-CX doing centralized VXLAN-to-VXLAN routing does not work with Linux-based edge switches bridging from VXLAN to edge VLANs. It works flawlessly with Arista cEOS and Cisco Nexus OS as edge devices.
BIRD Internet Routing Daemon¶
You must build the BIRD container image with the netlab clab build bird command.
BIRD is implemented as a pure control-plane daemon running on a Linux VM or as a container with a single external interface. You can set the node role to router to turn a BIRD instance into a more traditional networking device with a loopback interface.
netlab installs BIRD software in a container image or a VM running Ubuntu 22.04. The current version of BIRD shipping with Ubuntu 22.04 is 2.0.8.
BIRD supports a single router ID used for BGP and OSPF.
The VM or container running BIRD starts with static routes pointing to one of the adjacent routers (see host routes on Linux). After establishing routing adjacencies, BIRD copies BGP and OSPF into the kernel IP routing table.
OSPF Caveats¶
BIRD OSPF implementation has no reference bandwidth. The default OSPF cost is 10.
BGP Caveats¶
You must run OSPF on the BIRD daemon for the IBGP sessions to work.
BIRD will not advertise (reflect) an IBGP route if it has an equivalent OSPF route.
You cannot configure BGP community propagation on BIRD. All BGP communities are always propagated to all neighbors.
IPv6 Caveats¶
OSPFv3 does not advertise the prefix configured on the loopback interface even when the loopback interface is part of the OSPFv3 process.
If the BGP next hop of a reflected IBGP route is reachable as an OSPF route, BIRD advertises a link-local address as one of the next hops of the IBGP IPv6 prefix, potentially resulting in broken IPv6 connectivity.
Cisco ASAv Caveats¶
Some ASAv versions use older SSH protocols. For more details, see the Cisco IOSv SSH caveats.
Cisco Catalyst 8000v¶
Apart from the VLAN configuration, Catalyst 8000v implementation uses the same configuration templates as CSR 1000v.
Catalyst 8000v accepts CSR 1000v-based VXLAN configuration, but the validation tests fail. You cannot configure VXLAN on Catalyst 8000v with the current netlab release.
See also CSR 1000v and Cisco IOSv caveats.
Cisco CSR 1000v¶
Cisco CSR 1000v does not support interface MTU lower than 1500 bytes or IP MTU higher than 1500 bytes.
VLAN subinterfaces can be configured on Cisco CSR 1000v but do not work. CSR 1000v cannot be used as a router-on-a-VLAN-trunk device.
The minimum VXLAN VNI accepted by Cisco CSR 1000v is 4096. Using lower VNI values triggers a configuration error that is not caught by Ansible, resulting in a weird failure of the netlab initial command.
See also Cisco IOSv SSH, OSPF, RIPng, and BGP caveats.
Cisco IOSv¶
Cisco IOSv release 15.x does not support unnumbered interfaces. Use Cisco CSR 1000v.
BGP configuration is optimized for reasonable convergence times under lab conditions. Do not use the same settings in a production network.
Multiple OSPFv2 processes on Cisco IOS cannot have the same OSPF router ID. By default, netlab generates the same router ID for global and VRF OSPF processes, resulting in non-fatal configuration errors that Ansible silently ignores.
It’s impossible to configure RIPv2 on individual subnets on Cisco IOS. RIPv2 might be running on more interfaces than intended. netlab configures those interfaces to be passive.
Cisco IOS does not support passive interfaces in RIPng.
SSH Access to Cisco IOSv¶
Cisco IOSv SSH implementation uses RSA keys and older encryption algorithms that might not be allowed on newer Linux distributions. Most users running recent Ansible versions on Ubuntu won’t notice that as Ansible uses the ansible-pylibssh
package (installed together with Ansible) as its interface to libssh
.
If you’re still experiencing SSH key exchange errors, try these workarounds:
Add the following configuration to
~/.ssh/config
file[1]:
Host 192.168.121.*
KexAlgorithms +diffie-hellman-group1-sha1,diffie-hellman-group14-sha1
PubkeyAcceptedKeyTypes +ssh-rsa
HostbasedAcceptedAlgorithms +ssh-rsa
HostKeyAlgorithms +ssh-rsa
It seems that some
libssh
implementations cannot handle the+
specification in the~/.ssh/config
file. Try removing the+
sign in the crypto algorithm specifications or listing all standard algorithms and adding the legacy ones.Additionaly, you might have to execute
sudo update-crypto-policies --set LEGACY
on AlmaLinux/RHEL.
Cisco IOS XRv¶
netlab was tested with IOS XR release 7.4. Earlier releases might use a different management interface name. In that case, you must set defaults.devices.iosxr.mgmt_if parameter to the name of the management interface
Copying Vagrant public insecure SSH key into IOS XR during the box building process is cumbersome. The vagrant configuration file uses a fixed SSH password.
Maximum interface bandwidth on IOS XRv is 1 Gbps (1000000).
It seems IOS XR starts an SSH server before it parses the device configuration[2], and newer versions of Vagrant don’t like that and will ask you for the password for user vagrant. Ignore that prompt and the subsequent error messages[3]; you might get a running lab in a few minutes[4].
The IOS XRd container seems to be a resource hog. If you experience errors during the initial device configuration, reduce the number of parallel configuration processes – set the ANSIBLE_FORKS environment variable to one with
export ANSIBLE_FORKS=1
.
Cisco Nexus OS¶
Nexus OS release 9.3 requires 6 GB of RAM (netlab system default).
Nexus OS release 10.1 requires 8 GB of RAM and will fail with a cryptic message claiming it’s running on unsupported hardware when it doesn’t have enough memory.
Nexus OS release 10.2 requires at least 10 GB of RAM and crashes when you try to run it as an 8 GB VM.
To change the default amount of memory used by a nxos device, set the defaults.devices.nxos.memory parameter (in MB)[5]
Cumulus Linux¶
The Cumulus VX 4.4.0 Vagrant box for VirtualBox is broken. netlab is using Cumulus VX 4.3.0 with virtualbox virtualization provider.
netlab uses the VLAN-aware bridge paradigm to configure VLANs on Cumulus Linux. That decision results in the following restrictions:
The netlab-generated Cumulus Linux VLAN configuration cannot use routed subinterfaces; ifupdown2 version shipping with Cumulus Linux 4.4.0 refuses to create VLAN subinterfaces in combination with a VLAN-aware bridge.
The netlab-generated Cumulus Linux VLAN configuration cannot use routed native VLAN; ifupdown2 enslaves physical ports to the bridge and cannot configure IP addresses on physical ports.
FRRouting version bundled with Cumulus Linux 4.4 cannot run OSPFv3 in VRFs and fails to advertise local IPv6 prefixes in other areas.
See also other FRRouting caveats.
Running Cumulus Linux in Containerlab¶
containerlab could run Cumulus Linux as a container or as a micro-VM with firecracker. The default used by netlab is to run Cumulus Linux as a container. Add the clab.runtime parameter to node data to change that.
Cumulus Linux running as a container might report errors related to the DHCP client during initial configuration. In this case, you might have to disable apparmor for the DHCP client. The hammer-of-Thor command to fix this problem is
sudo systemctl disable apparmor
followed by a reboot; your sysadmin friends probably have a better suggestion.netlab uses Cumulus VX containers created by Michael Kashin and downloaded from his Docker Hub account. If Nvidia decides to release an official container image, change the container name with defaults.devices.cumulus.clab.image[5] parameter (or by editing the
topology-defaults.yml
file included with netlab).
Cumulus 5.0 with NVUE¶
You could configure Cumulus Linux 5.0 with configuration templates developed for Cumulus Linux 4.0 (use device type cumulus and specify desired device image), or with NVUE.
NVUE has several shortcomings that prevent netlab from configuring basic designs like IBGP on top of IGP. Don’t be surprised if the labs that work with cumulus device don’t work with cumulus_nvue device, and please create a GitHub issue whenever you find a glitch. We’d love to know (at least) what doesn’t work as expected.
To run Cumulus Linux 5.x with cumulus device type, add the following lines to your lab topology[5]:
defaults.devices.cumulus.libvirt.image: CumulusCommunity/cumulus-vx:5.2.0
defaults.devices.cumulus.libvirt.memory: 2048
Alternatively, you could add the following lines to your ~/.topology-defaults.yml
file:
devices.cumulus.libvirt.image: CumulusCommunity/cumulus-vx:5.2.0
devices.cumulus.libvirt.memory: 2048
Dell OS10¶
Dell OS10 uses a so-called Virtual Network interface to try to handle VLANs and VXLANs transparently in the same way. However, it seems that right now, it is NOT possible to activate OSPF on a Virtual Network (VLAN) SVI interface.
Sadly, it’s also NOT possible to use VRRP on a Virtual Network interface (but anycast gateway is supported).
At the same time, the anycast gateway is not supported on plain ethernet interfaces, so you need to use VRRP there.
Dell OS10 only allows configuring of the EVPN RD in the form
X.X.X.X:N.
By default, netlab usesN:M
for L3VNI, so on this platform the L3VNI RD is derived from the Router-ID and the VRF ID asrouter-id:vrf-id
(and the one generated by netlab is not used).
dnsmasq DHCP server¶
You have to build the dnsmasq container image with the netlab clab build dnsmasq command.
Fortinet FortiOS¶
FortiOS VM images by default have a 15 day evaluation license. The VM has limited capabilities without a license file. It will work for 15 days from first boot, at which point you must install a license file or recreate the vagrant box completely from scratch.
Ansible automation of FortiOS requires the installation of the FortiOS Ansible Collection 2.1.3 or greater and a FortiOS version > 6.0.
OSPF Caveats¶
Fortinet implementation of OSPF configuration module does not implement per-interface OSPF areas. All interfaces belong to the OSPF area defined in the node data.
Fortinet configuration templates set OSPF network type based on number of neighbors, not based on ospf.network_type link/interface parameter.
FRRouting¶
Many FRR configuration templates are not idempotent – you cannot run netlab initial multiple times. Non-idempotent templates include VLAN and VRF configurations.
VM version of FRR is an Ubuntu VM. The FRR package is downloaded and installed during the initial configuration process.
You can change the FRR default profile with the netlab_frr_defaults node parameter (
traditional
ordatacenter
, default isdatacenter
).netlab collect downloads FRR configuration but not Linux interface configuration.
FRR containers need host kernel modules (drivers) to implement the data-plane functionality of vrf, mpls, and vxlan netlab modules. The kernel modules are automatically loaded (when available) during the netlab up processing.
VRF and VXLAN kernel modules are usually bundled with a Linux distribution. If your Ubuntu distribution does not include the MPLS drivers, try installing them with
sudo apt install linux-generic
.You cannot load kernel modules in GitHub Codespaces and thus cannot use vrf, mpls, or vxlan modules on FRRouting nodes in that environment.
FRR containers have a management VRF. Use
ip vrf exec mgmt <command>
to run a CLI command that needs access to the outside world through the management interface. To disable the management VRF, set the netlab_mgmt_vrf node parameter to False.FRR initial container configuration might fail if your Ubuntu distribution does not include the VRF kernel module. Install the VRF kernel module with the
sudo apt install linux-generic
and reboot the server.FRR 9.0 and later creates malformed IS-IS LSPs; the bug has been fixed in release 10.0.1 (details). You cannot build an IS-IS network using Arista EOS and FRR if you’re running an affected version of FRR.
FRR configures BFD as part of OSPFv2/OSPFv3 configuration.
Common Junos caveats¶
Junos cannot have more than one loopback interface per routing instance. Using loopback links on Junos devices will result in configuration errors.
Junos configuration template configures BFD timers within routing protocol configuration, not on individual interfaces
Juniper vPTX¶
netlab release 1.7.0 supports only vJunosEvolved releases that do not require external PFE- and RPIO links. The first vJunosEvolved release implementing internal PFE- and RPIO links is the release 23.2R1-S1.8.
The rest of this section lists information you might find helpful if you’re a long-time Junos user:
vJunos Evolved (vJunos EVO, Juniper vPTX) uses Linux instead of BSD as the underlying OS. There are some basic differences from a “default” JunOS instance, including the management interface name, which is
re0:mgmt-0
.After the VM boots up, you need to wait for the virtual FPC to become Online before being able to forward packets. You can verify this with
show chassis fpc
. NOTE: You can see the network interfaces only after the FPC is online.It seems that the DHCP Client of the management interface does not install a default route, even if received by the DHCP server.
The VM will complain about missing licenses. You can ignore that.
See also Common Junos caveats.
Juniper vSRX in Containerlab¶
You can run Juniper vSRX as a container packaged by vrnetlab. See containerlab documentation for further details.
The Juniper vSRX image in vrnetlab uses the network 10.0.0.0/24
for its own internal network, which conflicts with the default network used by netlab for the loopback addressing. See Using vrnetlab Containers for details.
vSRX container built with vrnetlab uses flow based forwarding. You have two ways to use it:
Configure security zones, and attach interfaces and rules to them;
Change the mode to packet based forwarding.
See also Common Junos caveats.
vJunos-Switch in Containerlab¶
You can run Juniper vJunos-switch as a container packaged by vrnetlab. See containerlab documentation for further details.
The vrnetlab containers use the IP subnet 10.0.0.0/24
for the internal network, which conflicts with the default network used by netlab for the loopback addressing. See Using vrnetlab Containers for details.
See also Common Junos caveats.
Mikrotik RouterOS 6¶
Runs with the CHR image.
The CHR free license offers full features with a 1Mbps upload limit per interface, upgradeable to an unrestricted 60-day trial by registering a free MikroTik account and using the
/system license renew
command.LLDP on Mikrotik CHR RouterOS is enabled on all the interfaces.
A BGP VRF instance cannot have the same Router ID as the default one. The current configuration template uses the IP address of the last interface in the VRF as the VRF instance Router ID.
Mikrotik RouterOS 7¶
Runs with the CHR image.
LLDP on Mikrotik CHR RouterOS is enabled on all the interfaces.
The CHR free license offers full features with a 1Mbps upload limit per interface, upgradeable to an unrestricted 60-day trial by registering a free MikroTik account and using the
/system license renew
command.At the time of the build, testing is being performed with releases 7.14 (claimed as stable). With that release:
MPLS dataplane seems to have issues when using virtio networking, while the LDP and VPNv4 control plane work fine. With e1000 everything works fine.
BGP-to-OSPF route leaking is working on the control plane, but not on the dataplane.
There’s not an easy way to control the BGP community propagation.
Even if you configure the BGP RR cluster-id, this is not announced.
Route Reflection of inactive routes does not work.
There are still problems with VRFs and IPv6.
Nokia SR Linux¶
Only supported on top of Containerlab
Supports container image release 23.3.1 or later (due to YANG model changes)
Requires
nokia.srlinux
Ansible Galaxy collection (minimum version 0.5.0). Use ansible-galaxy collection install nokia.srlinux command to install it.MPLS and LDP are only supported on 7250 IXR (clab.type in [‘ixr6’,‘ixr6e’,‘ixr10’,‘ixr10e’])
Nokia SR Linux needs an EVPN control plane to enable VXLAN functionality. VXLAN ingress replication lists are built from EVPN Route Type 3 updates.
Inter-VRF route leaking is supported only in combination with BGP EVPN
SR Linux does not support multi-topology IS-IS.
SR Linux does not support configurable propagation of extended BGP communities.
Nokia SR OS¶
Only supported on top of Containerlab, using VRNetlab (VM running inside container)
Requires the latest
nokia.grpc
Ansible Galaxy collection and its dependencies to be installed from the git repo. You can also use the netlab install grpc command to install them
ansible-galaxy collection install git+https://github.com/nokia/ansible-networking-collections.git#/grpc/
python3 -m pip install grpcio protobuf==3.20.1
As of May 2024, the
nokia.grpc
collection crashes Ansible versions between 4.10.0 and 9.5.1. We recommend upgrading to Ansible release 9.5.1 (also included as part of netlab install grpc script):
sudo pip3 install --upgrade 'ansible>=9.5.1'
Sonic¶
Sonic implementation was tested with Azure sonic-vs VM image (release 2023-11) with FRR running in a container. Other Sonic distributions might use different approaches that would require significant modifications to the configuration deployment process.
BGP is the only routing protocol running on Azure Sonic. The choice is hardcoded in FRR compilation flags.
You cannot use IBGP as there’s no IGP protocol to resolve IBGP next hops, unless you believe in running IBGP over EBGP.
The Azure Sonic VM image has to be started with a preconfigured BGP AS number (specified in config_db.json); otherwise, it does not start the FRR container. That BGP process is removed during the initial BGP configuration and replaced with the actual BGP AS number specified in the lab topology.
netlab configures BGP on Sonic through vtysh, not through config_db.
VyOS¶
**netlab ** uses VyOS 1.5, which is currently a rolling release with daily builds. However, all the configurations should also work on the 1.4 LTS release (since it was tested just before it became the new LTS).
The use of a rolling release means potentially any build is broken or with regressions, even if the VyOS team is smart enough to perform some automated smoke tests and load arbitrary configurations to ensure there are no errors during config migration and system bootup.
Using the latest build published on Vagrant Hub should allow us to easily track and react to any configuration syntax change (which, anyway, is a very rare event). In any case, if you find a misalignment between the VyOS config and the netlab templates, feel free to Open an Issue or Submit a PR.
It looks like the official VyOS container is not updated as part of the daily builds; netlab uses a third-party container (ghcr.io/sysoleg/vyos-container
) to run VyOS with containerlab.
Other VyOS caveats:
VyOS configuration template configures BFD timers only at the global level
Multi-topology IS-IS (assumed by the IS-IS configuration module) cannot be configured with VyOS IS-IS CLI (bug report).
VyOS containers need host kernel modules (drivers) to implement the data-plane functionality of vrf, mpls, and vxlan netlab modules. The kernel modules are automatically loaded (when available) during the netlab up processing.
VRF and VXLAN kernel modules are usually bundled with a Linux distribution. If your Ubuntu distribution does not include the MPLS drivers, try installing them with
sudo apt install linux-generic
.You cannot load kernel modules in GitHub Codespaces and thus cannot use vrf, mpls, or vxlan modules on VyOS nodes in that environment.
While VyOS itself supports IPv6 transport for VXLAN, using static flooding with the vxlan module, this seems not to work with EVPN, where an IPv4 VTEP is always announced by frr.
VyOS does not have a simple way to handle a management VRF on containerlab, so it will always have a default IPv4 route (
0.0.0.0/0
) on the default routing table. This can cause some problems if you want to originate a default only if received by other routers.