If you are using the current version of Cumulus Linux, the content on this page may not be up to date. The current version of the documentation is available here. If you are redirected to the main page of the user guide, then this page may have been renamed; please search for it there.

Cumulus Linux 5.4 Release Notes

Download 5.4 Release Notes xls    Download all 5.4 release notes as .xls

5.4.0 Release Notes

Open Issues in 5.4.0

Issue IDDescriptionAffectsFixed
When you try to upgrade a switch from Cumulus Linux 5.5 or earlier to 5.8.0 or later with package upgrade, you see errors for expired GPG keys that prevent you from upgrading. To work around this issue, install the new keys with the following commands, then upgrade the switch.
cumulus@switch:~$ wget https://download.nvidia.com/cumulus/apt.cumulusnetworks.com/repo/pool/cumulus/c/cumulus-archive-keyring/cumulus-archive-keyring_4-cl5.6.0u5_all.deb
cumulus@switch:~$ sudo apt install ./cumulus-archive-keyring_4-cl5.6.0u5_all.deb
cumulus@switch:~$ sudo apt update
cumulus@switch:~$ sudo apt upgrade
4.0.0-4.4.5, 5.0.0-5.9.1
When sending untagged frames to the CPU with an MTU higher than the SVD (single VXLAN device) MTU, the kernel might crash.5.4.0-
When monitoring system statistics and network traffic with sFlow, an aggressive link flap might produce a memory leak in the sFlow service hsflowd.5.1.0-
Traditional SNMP snippets do not take effect unless you first enable SNMP with the NVUE nv set service snmp-server enable on and nv set service snmp-server listening-address commands. Alternatively, you can use the equivalent REST API methods.5.4.0-
When the MAC address of a neighbor changes, the zebra IP routing manager might crash.5.2.1-
When you configure extended nexthop encoding for a peer group, the peers in the group do not inherit the configuration. To work around this issue, configure extended nexthop encoding on each individual peer in the group. NVIDIA recommends that you upgrade to Cumulus Linux 5.6 or later to avoid this issue.5.4.0-
DHCP lease information is not collected in the cl-support file.4.3.0-
Changing non-default BGP timers with NCLU or vtysh commands sets the hold time and keep alive interval to 0 seconds. To work around this issue, restart the FRR service with the sudo systemctl restart frr.service command.5.3.1-
If BGP remote-as is set to an integer and you try to configure the local-as for a BGP instance, you see the following error:
% AS specified for local as is the same as the remote as and this is not allowed
This configuration is not allowed; it is considered to be eBGP and local preference is not advertised.
5.0.0-, 5.6.0-5.9.1
With VM migration from one VTEP to another, traffic loss might occur during a MAC move as locally learned MAC addresses are frequently refreshed between switchd and the kernel.5.2.1-
In an EVPN symmetric routing configuration, running the NVUE nv set vrf vlan auto command to derive layer 3 VNIs automatically might result in duplicate VLAN entries in the system. This most often occurs at scale when many VRFs have similar names. To work around this issue, manually specify a unique VLAN for each VRF.5.3.0-
NVUE and ip link show traditional bridge VLAN subinterface counts incorrectly. The ingress (Rx) count increments correctly but the egress (Tx) count does not increment. This issues occurs because the hardware does not support transmit counters for a VLAN subinterface; therefore, no statistics from the hardware are updated. Statistics for software forwarded packets show correctly.5.0.0-
On the NVIDIA SN2410 switch with an Innodisk SSD, you might see the following message in syslog:
smartd[501]: Device: /dev/sda [SAT], CHECK POWER STATUS spins up disk (0x00 -> 0xff)
This is a cosmetic issue and does not affect how the switch operates. To prevent this message from occurring, run the hdparm -S 24 /dev/sda command to change the HD timeout.
After you apply a change to the router MAC address on an SVI with the ifreload -a command, the old router MAC address still remains in the FDB table. To work around this issue, remove the old router MAC address with the sudo bridge fdb del <old_mac> dev bridge vlan <vlan_id> command.5.3.1-
In an EVPN multihoming environment with VRRP, when the master VRRP router fails, the standby router takes around 30 seconds to become active.5.3.1-
ACLs do not process inbound DHCP packets and the packets do not contribute to ACL counters5.2.0-
When you configure a route distinguisher (RD) or a route target (RT) manually for layer 2 VNIs, type-1 routes are not properly updated, type-1 EVI routes with the old RD are not properly withdrawn, and type-1 ES routes do not have the corresponding layer 2 VNI route target updated.5.0.0-
CVE-2023-38408: The PKCS#11 feature in ssh-agent in OpenSSH before 9.3p2 has an insufficiently trustworthy search path, leading to remote code execution if an agent is forwarded to an attacker-controlled system. (Code in /usr/lib is not necessarily safe for loading into ssh-agent.) NOTE: this issue exists because of an incomplete fix for CVE-2016-10009
Mitigation: Do not use ssh-agent forwarding (the man page for ssh_config says that “agent forwarding should be enabled with caution”), or start the ssh-agent program with the -P option to allow only specific PKCS#11 libraries (or none with -P ‘')
For Cumulus Linux 4.3.2, the /usr/bin/ssh-agent program has all permissions turned off (chmod 0) to prevent its execution if a vulnerable version is detected.
4.0.0-4.3.1, 5.0.0-
Collecting a cl-support file in a high VNI and interface environment can result in an out-of-memory (OOM) event on the switch. An OOM event can cause critical services to restart and might impact traffic.5.1.0-
In an EVPN environment, VM migration (IP and MAC address migration) might not work because the new local VTEP to which the VM migrates does not install the entry in the kenel. To work around this issue, restart the switchd service.5.4.0-
When you update a prefix list associated with an RP, the pimd service might crash if the prefix list exists without any prefixes. To work around this issue, ensure that any prefix list associated with an RP includes at least one prefix at all times.5.3.1-
When PTP is not enabled on the switch, NVUE nv show ptp commands freeze. This might cause other NVUE commands to fail and the NVUE service to restart.5.3.1-
The NVUE nv show service ptp current command output shows an incorrect value. To work around this issue, run the nv show service ptp monitor timestamp-log command or the Linux pmc utility.5.3.1-
When BGP receives an EVPN type-5 route with a gateway IP overlay attribute, the gateway IP overlay attribute in the attr memory (which is already inserted in the attribute hash) might change. As a result, the modified attr memory might match with another attr in the attribute hash, which produces duplicate entries in the hash table. As a result, BGP might crash when deleting one of the duplicate attr structures.5.0.0-
When zebra receives route updates that include both a route with a recursive next hop and the route used to resolve that next hop, zebra might mark the route with the recursive next hop as inactive. To work around this issue, reprocess the route updates by running the appropriate clear command for the protocol in use. For example, for BGP, clear inbound routes from the relevant neighbor using the nv action clear vrf router bgp neighbor address-family in command.4.2.1-
When you power on the NVIDIA SN3420 switch with no connected cables, the QSFP ports LEDs light in amber.5.3.0-
The switchd service does not handle certain route and next hop updates, which causes a synchronization loop. For example, in a VRF route leaking configuration, where a next hop group spans across multiple VRFs, when one of the routes is withdrawn and the next hop is no longer used, switchd has problems synchronizing other next hops in the group
To work around this issue, disable next hop groups in zebra with the vtysh zebra nexthop proto only command, and then reboot the switch.
The SNMP MIB definition file /usr/share/snmp/mibs/Cumulus-BGPVRF-MIB.txt does not define the INDEX of the bgpPeerEntry correctly. This issue does not impact SNMP functionality for this MIB.4.3.1-
The json output for the vtysh -c ‘show bgp all json command is missing a string key in front of the list of routes under the l2vpnevpn address family.5.4.0-
On certain QSFP-DD and OSFP optical modules, the ethtool -m command, and the related NCLU and NVUE commands that display optical module information fail.5.4.0-
On the NVIDIA SN4700 switch, reversing the upper four lanes on a port does not work and might cause link degradation. If you swap the upper and lower four lanes on a port, the firmware gets stuck.5.3.0-
BGP aggregate routers are not advertised after learning the same route from another protocol. To work around this issue, restart the FRR service or, if possible, don’t learn the route from another protocol (use route maps instead).5.3.0-
In rare cases when there is high load, the clagd service might experience a buffer overflow and MLAG bonds stay in a proto-down state on the secondary switch. You see a “NetlinkThread: Netlink overflow” log message and the MLAG state indicates VLAN conflicts between peers. To work around this issue, restart the clagd service with the sudo systemctl restart clagd command on the switch that reports the overflow log message.5.2.0-
If you run the NVUE nv show vrf router bgp peer-group command when the BGP instance is not enabled, you see an NVUE traceback error.5.4.0-
If the NVUE startup.yaml configuration file is invalid, the nv config apply startup command times out without providing details on the error.5.4.0-
The vtysh show version command shows an incorrect version number; for example, instead of Cumulus Linux 5.4, the command output shows Cumulus Linux
FRR does not apply Type-0 ESI configuration for EVPN multihoming bonds consistently after an FRR service reload. This issue occurs because the system MAC address value (es-sys-mac) is only compatible with a 3-byte Ethernet segment ID (es-id) for Type-3 ESIs, but still renders even when the Ethernet segment ID is 10 bytes for Type-0 ESIs. To work around this issue, configure EVPN multihoming bonds with a Type-3 ESI (es-sys-mac plus a 3-byte es-id).5.0.0-
If the number of bonds and its bond members exceeds the total number of physical ports on the switch
 and LACP bypass is enabled, switchd might crash when frequent link flaps occur or switchd restarts. To work around this issue, disable LACP bypass.
When using WJH, if you export dropped packets to a file in PCAP format, the file contains custom WJH header data. As a result, certain tools, such as Wireshark, cannot decode the data. To work around this issue, use the –no_metadata option with the export command:
cumulus@switch:~$ what-just-happened poll –export –no_metadata
Changing the ebgp-multihop setting for a BGP peer always resets the peer, even if the configured TTL value matches the existing TTL value of the peer.5.3.1-
The wjh_dissector.lua WJH packet decoder script provided with Cumulus Linux might fail to decode all WJH packets.5.4.0-
When you remove the restriction from a TACACS+ mapped user to remove per command authorization, the tacplus-restrict -R command does not restore ownership of restored files correctly. As a result, some commands might fail due to permission errors in the files or directories under the home directory. To work around this issue, run the sudo chown command to correct the ownership of the affected files and directories.5.0.0-
On the Spectrum-2 and Spectrum-3 switch, multiple interfaces (in the same PLL quarter) might flap intermittently at the same time.4.2.1-
In certain cases, Cumulus Linux does not process next hop updates because the zebra IP routing manager thinks the state of the next hops is unchanged. As a result, route installation fails and remains in a rejected state.5.3.0-
sudo for TACACS+ users with privilege level 15 does not work when reaching the TACACS+ server through the default VRF. To work around this issue, specify the interface name that the default VRF uses in the vrf= setting of the /etc/tacplus_servers file or run the NVUE nv set system aaa tacacs vrf command. If you don’t run either command, a TACACS+ user with privilege level 15 can run vrf task exec default sudo … to execute the sudo command.5.0.0-5.9.1
The ADVA 5401 SFP module with hardware revision 5.01 does not come up at layer 1 when you use 10G QSA adaptors. To work around this issue, use 25G QSA adaptors.4.4.0-4.4.5, 5.0.0-5.9.1
The NVUE PIM timer command option names keep-alive and rp-keep-alive are inconsistent and need to change to keepalive and rp-keepalive.5.4.0-
On the Spectrum-2 and Spectrum-3 switch, if you use module SPQCELRCDFB when connected to a 3rd party switch, you might see no link or a very long link up time (around two minutes). To work around this issue, bring down the port, then bring it back up.
During upgrade, when one MLAG node is upgraded and the other MLAG node is not yet upgraded, permanent neighbors cannot synchronize between MLAG nodes. The clagctl dumppermanentneighs command only shows local neighbors.4.2.1-4.3.1, 4.4.0-, 5.5.0-5.9.1
To reach the TACACS+ server through the default VRF, you must specify the egress interface you use in the default VRF. Either run the NVUE nv set system aaa tacacs vrf command (for example, nv set system aaa tacacs vrf swp51) or set the vrf= option in the /etc/tacplus_servers file (for example, vrf=swp51). A similar issue might prevent TACACS+ users with privilege level 15 from using sudo if the TACACS+ server is reachable only on the default VRF. If this occurs, and you do not run the above configuration workaround, the TACACS+ user with privilege level 15 can use vrf task exec default sudo … to execute the sudo command using the TACACS+ server on the default VRF.5.0.0-
The NVUE command to set the frequency of LLDP updates nv set service lldp tx-interval and the NVUE command to set the amount of time to hold the information before discarding it nv set service lldp tx-hold-multiplier do not provide reasonable maximum and minimum values. Cumulus Linux 5.5.0 and later provides new values. For the nv set service lldp tx-interval command, you can now set a minimum value of 5 and a maximum value of 32768. For the nv set service lldp tx-hold command, you can set a minimum value of 1 and a maximum value of 8192.
NGNIX might be listening on port 80 through its default configuration in /etc/nginx/sites-enabled/default. To work around this issue, run the following commands:
 cumulus@switch:~$ sudo rm -f /etc/nginx/sites-enabled/defaultcumulus@switch:~$ systemctl is-active nginx && sudo invoke-rc.d nginx restart
For layer 3 interfaces configured on the switch, certain triggers, such as port flaps and subinterface flaps, or when configuring the ports to and from layer 2 and layer 3, cause the dummy internal VLAN to not free up, which can result in exhaustion of the dummy internal VLANs designated for the layer 3 interfaces. When this occurs, you see the following switchd log messages:
ERR dummy internal vlans exhaustedERR cannot allocate vlan for sub-interface
When an ECMP route is present in a non-default VRF, resilient hashing does not work as expected and flows might get remapped to a new next hop when the set of nexthops changes.
The NVUE nv show system forwarding profile-option command reports an incorrect Max ipv4 mcast routes value. To work around this issue, validate values with cl-resource-query.
If you restart FRR with the log file debugging level set to informational, BGP crashes.5.4.0-
On a VX, NVUE commands with an argument parameter that can be multiple types (such as IPv4 and IPv6) do not provide auto complete or additional options when you use a question mark.
The python-netaddr package is not preinstalled on the switch, which leads to an error similar to the following when SNMP accesses data from the CUMULUS-BGPVRF-MIB
CUMULUS-BGPVRF-MIB::bgpPeerFsmEstablishedTransitions = No Such Instance currently exists at this OID
To work around this issue, manually install the python-netaddr package with the sudo -E apt-get install python-netaddr command.
The neighmgrd service does not enable the snooper unless ARP suppression is enabled on at least one VXLAN interface. This can result in missing ARP and NDP entries if the host does not directly interact with the switch.5.3.1-
The NVUE nv unset command does not completely remove IPv6 DNS server configuration
Cumulus Linux does not let you add an interface to the bond interface when the bridge-allow-untagged no option is present.
TACACS+ packages in the local apt repository might be out of date; as a result, the upgrade does not install tacacs0 through tacacs15 users in the correct NVUE groups. When you run NVUE commands as a TACACS+ user, the commands fail and you see the error You do not have permission to execute that command
To obtain the correct packages, install the tacplus-client package and its dependencies from apt.cumulusnetworks.com.
apt source linux fails to download the Linux kernel source code. To work around this issue, run the sudo apt update && sudo apt install linux-source-5.10 command or download the desired version from https://apt.cumulusnetworks.com/repo/pool/cumulus/l/linux/ and install it with the sudo dpkg -i $filename command. The source code in a tar.xz file will then be located in the /usr/src/ directory.5.2.0-
When configuring PTP, NVUE does not use the PTP priority2 setting but uses the priority1 setting instead.
After you add or delete a static MAC entry on the bridge FDB, a core dump occurs if the interface is VXLAN and the MAC address is 00:00:00:00:00:
If you run the NVUE nv set service snmp-server readonly-community command to set an SNMP V2 trap community string that includes fewer than eight characters, the configuration fails. The SNMP V2 trap community string must include eight or more characters.
When daylight saving time changes, the MLAG initDelay timer resets and all MLAG bonds go down.4.4.4-4.4.5, 5.2.1-5.9.1
If you restore an NVUE startup.yaml file or run the nv config patch command after an upgrade that includes breakout ports with QoS configuration, the NVUE configuration fails to apply. Subsequent attempts to run nv config apply fail with a message similar to Invalid config [rev_id: 11] qos config is not supported on the following invalid interface: swp1s0. Supported on swp and bond interface types. To work around this issue, run nv unset on the configured QoS settings, then apply the breakout port configuration before you configure QoS. Alternatively, you can remove the QoS configuration from the yaml file and patch it separately after applying the breakout configuration.5.4.0-5.9.1
The NVUE PTP shaping commands are available in the NVUE command list; however, these commands are disabled and do not configure PTP shaping. PTP shaping is not supported in Cumulus Linux
Cumulus Linux 5.4 package upgrade (apt-upgrade) does not support warm restart to complete the upgrade; performing an unsupported upgrade can result in unexpected or undesirable behavior, such as a traffic outage.
NVUE deprecated the port split command options (2x10G, 2x25G, 2x40G, 2x50G, 2x100G, 2x200G, 4x10G, 4x25G, 4x50G, 4x100G, 8x50G) with no backwards compatibility.5.0.0-
If you use TACACS+ authentication, modifying the TACACS+ configuration with NVUE might result in a timeout error when you run the nv config apply command. To work around the issue, restart the nvued service with the sudo systemctl restart nvued.service command, then apply the configuration again.
If you uninstall dynamic NAT rules and switchd restarts before all the dynamic NAT flows age out and are deleted, you might see dynamic flow deletion errors in switchd.log. These errors do not affect new dynamic NAT flows from new NAT rules.5.4.0-5.9.1
On the Spectrum-2 and Spectrum-3 switch with ports operating at 1G speed, there is loss of frames that have an odd or random frame size. In the frame size range of 75 to 1000 bytes, there is frame loss of less than approximately one percent for all odd or random frame sizes in the range. In the frame size range greater than 1000 bytes, there is no loss observed.
The switch does not learn MAC addresses from DHCP packets. When a DHCP enabled host is plugged in for the first time, it tries to obtain an IP address through DHCP. The switch does not learn the MAC address of the host when it receives these DHCP packets; therefore, the host MAC address is not updated in the local forwarding database and it does not get advertised across EVPN. The switch learns the MAC address when it receives other packets, such as ARP or ND from the host. To work around this issue, either configure a temporary IP address on the host to initiate ARP/ND or enable IPv6, which sends ND after link local address creation.5.2.0-
In an MLAG configuration, when a link failure occurs on the peerlink or the peerlink shuts down, the switch in the secondary role attracts traffic to its local VTEP as it advertises the local VTEP IP address momentarily just before the VXLAN device is protodown. This traffic is dropped for a brief moment (between 5 and 10 seconds) because the MLAG bonds on the secondary switch are already protodown.5.1.0-
When connecting NVIDIA-to-NVIDIA in PAM4, you must enable auto-negotiation.5.4.0-5.9.1
The Spectrum-3 hardware configuration is not optimized for the best PTP accuracy when using 25GbE. You might see higher than expected PTP offsets on this platforms and interface speed.5.4.0-
If you use the NVUE REST API to configure a local user with a hashed password, the user cannot log in and the /etc/nvue.d/startup.yaml file shows the password as plain text.5.4.0-5.9.1
When you run the NVUE nv show interface command, you see an error similar to the following:
Error: GET /nvue_v1/interface/swp45?rev=operational responded with 500 INTERNAL SERVER ERROR
When you use the NVUE REST API with a TACACS+ user account, you see authentication failures. To work around this issue, replace the /etc/pam.d/nvueapi file with the following content:
@include common-auth@include common-account@include common-session-noninteractive
After you replace the content, run these two commands:
cumulus@switch:~$ sudo chown root:shadow /run/tacacs_client_map
cumulus@switch:~$ sudo chmod 0664 /run/tacacs_client_map
If you run the NVUE nv set interface description command without providing a description, the nv config apply command fails with the error Unable to restart services (ifreload-nvue.service).5.4.0-
When you run the sensors command, the output shows an erroneous fault on some front panel ports.5.2.0-
When using TACACS+, if the /etc/nsswitch.conf file specifies passwd: files tacplus (files is listed before tacplus), a user that is present in both the local /etc/passwd file and the TACACS+ server cannot log into the switch. NVIDIA recommends that when using TACACS+, you list tacplus before files in /etc/nsswitch.conf. When using NVUE, ensure that tacacs has priority over local.5.4.0-5.9.1
Ethtool HwIfInDot3FrameErrors (Rx FCS Errors) might lead to an incorrect and very large HwIfInErrors count. To work around this issue, stop the source of the FCS errors, then reset the interface counters. First, run the sudo mst status command to find the device, then run the sudo mlxlink -d -p <port_number> -pc command to reset the interface counters; for example, sudo mlxlink -d /dev/mst/mt53104_pciconf0 -p 39 -pc.5.3.1-
If you use su to change to a user specified through TACACS+, the user becomes the local tacacs0 thru tacacs15 user instead of the named user to run sudo commands. As a result, the named user password might not match the local tacacs0 thru tacacs15 user password.3.7.0-3.7.16, 4.0.0-4.4.5, 5.0.0-5.9.1
If you have a large number of MAC addresses, they do not age out at the MAC ageing timeout value configured on the switch. It might take up to 30 seconds more for the MAC addresses to age out and be deleted from the hardware. To work around this issue, wait for the ageing timeout value plus 30 seconds.5.4.0-5.9.1
If you run NVUE commands to break out a port into four interfaces, NVUE disables the subsequent port automatically. However, if you run NVUE commands to break out a port into eight interfaces, NVUE does not disable the subsequent port automatically; you have to run the NVUE command to disable the subsequent port.
In Cumulus Linux 5.4 and earlier, the command to enable Neighbor Discovery (ND) router advertisement is inverted and causes confusion; nv set interface ip neighbor-discovery router-advertisement enable off. In Cumulus Linux 5.5 and later, the command to enable router advertisement is updated to nv set interface ip neighbor-discovery router-advertisement enable on.5.3.0-
When you disable BGP globally with the nv set router bgp enable off command, applying the configuration with NVUE might fail due to an FRR reload failure.5.2.0-
When you change the BGP router ID that causes a change to an EVPN VNI RD, EVPN EAD-per-EVI routes are not updated properly.5.3.0-
If you use NVUE to configure multiple SNMP listener addresses at the same time, the SNMP service fails to start. To work around this issue, configure multiple SNMP listener addresses one at a time.5.3.0-
When you apply switch configuration for the first time on a freshly booted switch and you run the nv config apply command after setting the hostname with nv set system hostname, you might see the error message Failed to start Hostname Service. To work around this issue, run the nv config apply command a second time.5.3.0-5.9.1
With double tagged QinQ interfaces, if the bridge corresponding to the QinQ interface flaps, you might see invalid learning notifications and errors from similar to the following:
Can’t set non-static MAC address for non-vPort 0x0001006B when VID is VFID. 
The NVUE nv unset interface link lanes command does not restore the port lane setting to the default value. To work around this issue, run the nv set interface link lanes command.5.4.0-
The l1-show eth0 command does not show port information and is not supported in this release.5.3.0-5.9.1
Cumulus Linux VX images might include an incorrect entry at the end of /etc/apt/sources.list, which produces warnings when you run apt update. Remove this entry to avoid these warnings.5.2.0-
At high scale with 160 VRFs and 10 VLANs per VRF (a total of 1600 VLANs), you see traffic loss during primary switch reboot. To work around this issue, reduce the scale to 40 VRFs with no more than 400 VLANs in the configuration, and use a common MAC address.5.1.0-
The cl-support generation script causes TC filter collection to run as a background process for each interface, which can lead to memory exhaustion on a high scale configuration and on a switch with a small memory footprint.5.1.0-
On rare occasions, when you query the system hostname through the hostnamctl application, you see a timeout. NVUE uses the hostnamctl application to determine the system hostname, which can result in an nv config apply command failure.5.2.0-
When you connect the NVIDIA SN4600C switch to a Spectrum 1 or Spectrum-3 switch with a 40GbE passive copper cable (Part Number: MC2210126-005) on edge ports 1-4 and 61-64, there is an Effective BER of 1E-12 in PHY.5.2.0-5.9.1
You cannot use NVUE to configure an SNMP view to include a subtree beginning with a period. For example:
cumulus@switch:~$ nv set service snmp-server viewname cumulusOnly included . GET /nvue_v1/service/snmp-server/viewname/cumulusOnly/included?pointers=%5B%22%2Fparameters%22%2C+%22%2Fpatch%2FrequestBody%2Fcontent%2Fapplication~1json%2Fschema%22%2C+%22%2Fpatch%2Fparameters%22%2C+%22%2Fpatch%2Fresponses%2F200%2Flinks%22%5D responded with 404 NOT FOUND
To work around this issue, reference the OID without the preceding period ( . ) in the command.
On a Spectrum-3 switch, the PTP offset in 10GbE changes between +-27. The average offset is around
On the NVIDIA Spectrum-1 switch, the nv show system forwarding command shows GTP hashing output, which is not supported on this switch.5.2.0-
The /var/lib/snmp/snmpd.conf file contains multiple Warning: Unknown token: ifXTable messages. To avoid these warnings, add the -noTokenWarnings option to the SNMPDOPTS variable in the /etc/defaults/snmpd file, then restart the snmpd service.5.2.0-
The BGP4-MIB.txt file is missing from Net-SNMP agent.5.0.0-
A slow memory leak (~5KB over 24 hour period at a 60 second polling interval) might occur in SNMP when you walk the following system MIB objects ( –> Entity MIB –> Entity Sensor MIB –> rip2 –> interface/interfaces –> ifMIB –> IP –> hostResource
PAM4 split cables (such as 2x100G, 4x100G, and 4x50G) do not work with a forced speed setting (when auto-negotiation is off) as the default speed enabled is for NRZ mode (such as 100G_4X). To work around this issue, set the appropriate lanes for forced speed (with auto-negotation off) with the ethtool -s swpX speed <port_speed> autoneg off lanes <no_of_lanes> command. For example:
cumulus@switch:~$ sudo ethtool -s swp1 speed 100000 autoneg off lanes 2
On the NVIDIA SN4700 switch, inserting and removing the PSU might cause loss of frames.5.2.0-5.9.1
When you configure two VNIs in the same VLAN, ifupdown2 shows a vlan added to two or more VXLANS warning, which is only issued after the VNI is already added to the bridge. This leaves the new VNI in the PVID even if there is already an existing VNI configured in that PVID.5.1.0-5.9.1
On the NVIDIA SN4700 switch, inserting and removing the PSU might cause loss of frames.5.2.0-5.9.1
After you disable traffic shaping in the /etc/cumulus/datapath/qos/qos_features.conf file, the default QOS traffic shaping configuration does not restore. To work around this issue, restart switchd.4.4.3, 5.0.0-
Under a high load, you might see ingress drop counters increase. The drops are classified as HwIfInDiscards in ethtool and shown as ingress_general in hardware.4.3.0-4.4.5, 5.0.0-5.9.1
On rare occasions, after you reboot or restart switchd on a Spectrum 1 switch, any 25G connections with Direct Attach Copper (DAC) cables that connect from the switch to a non-NVIDIA device might flap continuously. To work around this issue, bring the affected link administratively down for a few seconds on the non-NVIDIA device, then bring the link back up.4.4.4-4.4.5, 5.1.0-5.9.1
When you run the systemctl reload switchd command, there is momentary traffic loss after a port configured with lossless buffers goes down. This is only temporary and the traffic stabilizes after the initial drops.5.1.0-
When the CPU load is high during a warm boot, bonds with a slow LACP rate fail to forward layer 2 traffic for up to 60 seconds (depending on the duration of the CPU load) and static bonds fail to forward layer 2 traffic for up to 5 seconds.5.1.0-5.9.1
After you run Linux commands to enable a custom ECMP or LAG hash parameter, if you set the hash_config.enable or lag_hash_config.enable parameter to false, the custom parameters do not restore their default values. To work around this issue, change the custom ECMP or LAG hash parameters to their default values manually.5.1.0-
When the CPU load is high during a warm boot, bonds with a slow LACP rate fail to forward layer 2 traffic for up to 60 seconds (depending on the duration of the CPU load) and static bonds fail to forward layer 2 traffic for up to 5 seconds.5.1.0-5.9.1
If GTP Hashing is set to true, after more than two warm boots, switchd fails and a cl-support file is generated.5.1.0-
In an MLAG EVPN deployment when either of the MLAG peers reboots, FRR incorrectly programs the local host entries in the ARP table as remote. To work around this issue, either restart FRR or use BGP policies to mark and drop routes within an MLAG pair. Both MLAG peers must have an outbound policy that add a community representing the unique MLAG pair to Type-2 EVPN routes and an inbound policy to match and drop that community.4.4.4-
With RADIUS enabled for user shell authentication, there might be a delay in local user authentication for non cumulus user accounts.5.0.0-5.9.1
When a VNI flaps, an incorrect list of layer 2 VNIs are associated with a layer 3 VNI. The NCLU net show evpn vni detail command output shows duplicate layer 2 VNIs under a layer 3 VNI.3.7.15, 4.4.2-4.4.5, 5.0.0-
The net show time ntp servers command does not show any output with the management VRF.3.7.15-3.7.16, 4.1.1-4.4.5, 5.0.0-5.9.1
When you run the ethtool -m or the l1-show command, the 400G interface optical values do not show.4.4.0-4.4.5, 5.0.0-5.9.1
CVE-2021-39925: Buffer overflow in the Bluetooth SDP dissector in Wireshark 3.4.0 to 3.4.9 and 3.2.0 to 3.2.17 allows denial of service via packet injection or crafted capture file.
Vulnerable: <= 2.6.20-0+deb10u1
Fixed: 2.6.20-0+deb10u2
4.0.0-4.4.1, 5.0.0-
CVE-2021-42771: relative path traversal in Babel, a set of tools for internationalising Python applications, could result in the execution of arbitrary code
Vulnerable: 2.6.0+dfsg.1-1Fixed: 2.6.0+dfsg.1-1+deb10u1
4.0.0-4.4.1, 5.0.0-
When connecting the NVIDIA SN4600 switch to another NVIDIA Spectrum switch, you must use auto-negotiation mode (not force mode); otherwise the switch might use the wrong Tx configuration.5.0.0-5.9.1
When you use NCLU to remove the configuration for a peer that is a member of a group but also has other peer-specific configuration, you must remove the peer-specific configuration before you delete the peer in a separate NCLU commit.5.0.0-5.9.1
Cumuls Linux does not support a bond with more than 64 ports. Any configuration with more than 64 ports in a bond changes all ports to down when you apply the configuration.5.0.0-5.9.1
When you change the VRRP advertisement interval on the master, the master advertisement interval field in the show vrrp command output does not show the updated value.4.4.0-4.4.5, 5.0.0-5.9.1
When configured with NVUE, SVIs do not inherit the pinned MAC address of the bridge.4.3.0, 5.0.0-
The NVUE nv show vrf default router bgp peer command produces a 404 not found error.4.4.0-4.4.5, 5.0.0-5.9.1
When you enable a service in the management VRF, systemctl issues a warning similar to the following:
Warning: The unit file, source configuration file or drop-ins of ntp@mgmt.service changed on disk. Run ‘systemctl daemon-reload’ to reload unit
You can safely ignore this warning.
4.0.0-4.4.5, 5.0.0-5.9.1

Fixed Issues in 5.4.0

Issue IDDescriptionAffects
In rare circumstances, attempting to install a Cumulus Linux 5.3 image can fail during installation. The device stops at the (initramfs) prompt. To resume installation, enter the exit command at the (initramfs) prompt.5.3.0-5.3.1
Currently, the default core dump size limit on Cumulus Linux is 256M but the SDK generates core dumps around 800M. To avoid incomplete core files, you can increase the core dump size limit.4.2.1-4.3.1, 4.4.0-5.3.1
Switch fans run at very high speed but the temperature is normal.5.2.0-5.3.1
When the switch boots up, you might see logs similar to the following in the nvued log files because switchd is not up and running. This does not impact switch functionality
2023-01-29T06:05:18.683152+00:00 cumulus nvued:  INFO: apply_config.py:2177 Apply Issues: (b’),(update-ports returned with error (code 254): ports validation node file is not accessibleswitchd validate_node is absent),(ports configuration(ports.conf/ports_width.conf) is invalid),(')
The ethtool -m command does not show Digital Optical Monitoring (DOM) for SFP transceivers. To work around this issue, run the l1-show or mlxlink command instead.5.2.0-5.3.1
When a switch is operating as a PTP Grand Master, the phc2sys service might exit shortly after starting as the initial offset to correct is the delta from epoch, which is too large to correct.
When using TACACS+, a TACACS+ server name that returns more than one IP address, such as an IPv6 and IPv4 address, is counted many times against the limit of seven TACACS+ servers, which might cause some of the later listed servers to be ignored as over the limit. To work around this issue, you can set the prefer_ip_version configuration option (the default value is 4) to choose between an IPv4 or IPv6 address if both are present.3.7.0-5.3.1
The SNMP monitor might fail to send the expected traps.5.3.0-5.3.1
The traffic control rules that the EVPN multihoming configuration adds to an interface are deleted when the hsflowd service restarts. The hsflowd service deletes the EVPN multihoming traffic control filters after you stop hsflowd, then adds back the match-all filters with the psample action; however, hsflowd does not add back the EVPN multihoming traffic control rules.4.4.0-5.3.1
The ptmd service causes memory leaks.5.3.0-5.3.1
On the NVIDIA SN2410 switch, ports with optical transceivers show FAULT errors in the sensor command output.4.2.1-5.3.1
DHCP packets do not forward over VXLAN interfaces in multicast replication environments. This issue does not affect VXLAN environments using head end replication (HER).5.2.0-5.3.1
The memory consumption in ptmd can grow when the socket being used for a BFD session needs to be recreated. This is often seen when the route being used to forward the BFD packets is removed; for example, if the connected route is removed when an interface goes down, over which a single hop BFD session is formed.5.3.0-5.3.1
Some EVPN multihoming show commands might cause BGP to crash if you use the json flag and attempt to reference the default VRF by name. For example, show bgp l2vpn evpn es-vrf json.5.0.0-5.3.1
When upgrading from Cumulus Linux 5.0.0 thru 5.2.1 to Cumulus Linux 5.3.0 or 5.3.1, the babeltrace and python3-babeltrace packages are not added automatically even though they are in the default image in Cumulus Linux 5.3.0 and later. You may need these packages to decode LTTNG traces with /usr/lib/frr/frr_babeltrace.py.. If you need to use this script, run the sudo apt update && sudo apt install babeltrace python3-babeltrace command to install the packages.5.3.0-5.3.1
NVUE gracefully detects and handles upgrades that include valid flexible snippets. For any invalid (incompatible) flexible snippets, you must delete the snippets before you apt upgrade Cumulus Linux; otherwise, the NVUE nv config apply command and the equivalent REST API, do not run.5.3.0-5.3.1
When you clear interface counters with the ethtool -S clear command, the command fails with the following message:
switch:~$ ethtool -S swp1 clearethtool (-S): unknown parameter ‘clear’
When you add the /etc/frr/frr.conf file to the ignore list for NVUE, any configuration change causes FRR to restart because a check is done to see if any running configuration has changed since the previously applied configuration in the vtysh shell.5.3.0-5.3.1
NVUE requires the SNMPv2 community string to be a minimum of eight characters.5.3.0-5.3.1
When the switch needs to forward a frame that has a source MAC address of 00:00:00:00:00:00, the dmesg log might report the message bridge: RTM_NEWNEIGH with invalid ether address in a loop every 30 seconds. The log message is harmless and frames with that MAC forward correctly.5.3.0-5.3.1
The memory consumption in ptmd can grow when the socket being used for a BFD session needs to be recreated. This is often seen when the route being used to forward BFD packets is removed; for example, if the connected route is removed when an interface goes down, over which a single hop BFD session is formed.5.2.0-5.3.1
After you restart the FRR service, show commands incorrectly reflect the VLAN associated with layer 3 VNIs as 0:
# net show evpn vni 123VNI: 123Type: L3Tenant VRF: BLUEVlan: 0
On Spectrum 1 switches when configuring ACLs in non-atomic mode, if there are too many IPv6 matches due to rules with both input-interface and output-interface matches on SVIs, the ACL install fails and switchd crashes.5.2.0-5.3.1
Due to a race at the initial configuration, the SDK RDQ test may test RDQ configured for WJH and fail the test resulting in a fatal health event.5.2.0-5.3.1
When an FRR routing service (such as bgpd) becomes unresponsive, watchfrr might fail to stop and restart service. To work around this issue, restart FRR with the systemctl restart frr command.4.4.0-5.3.1
The Linux utility that sends ARP packets is constrained to 512 interfaces on the system. In large scale deployments, the warm boot process fails repeatedly as it sends gratuitous ARP requests for each local address. This issue does not impact the functionality and can be ignored.5.2.0-5.3.1
ACL configurations fail when the TCAM memory is exhausted because the CTCAM profile is configured with duplicate entries.5.2.0-5.3.1
When you delete a route under the following conditions, switchd might crash:- The minimum number of routes is set to a non-zero value
- KVD utilization is higher than sixty percent
- The number of routes currently configured is less than the minimum reserved value, and multiple KVD linear resources have just been freed and are waiting in the Garbage Collector queue.
When you configure or unconfigure a BGP peer and interface towards a host, memory corruption can cause BGP to crash.4.3.0-4.3.1
When using TACACS+, if you configure per-command authorization with the tacplus-restrict command, NVUE configuration commands fail for any user with a privilege level lower than 15. This occurs because NVUE is not able to create a .local user directory.5.2.0-5.3.1
The NVUE nv show system forwarding –output json command does not provide any output. To work around this issue, run the nv show system forwarding command.5.2.0-5.3.1
You can not apply NVUE configurations when TACACS is enabled for user authentication. To work around this issue, add the nvue account to the exclude_users line in /etc/tacplus_nss.conf:
The NVUE nv show interface link state command shows an empty table instead of showing the port link state.5.0.0-5.3.1
The NVUE nv show interface command shows the operational state of the tunnel as down even though the tunnel is up, and encapsulation and decapsulation occurs correctly.5.1.0-5.3.1
FRR restarts even when the NVUE configuration overwrite mode is set.5.0.0-5.3.1