diff --git a/content/cumulus-linux-516/Whats-New/rn.md b/content/cumulus-linux-516/Whats-New/rn.md index 1e380fa460..ac32834ae0 100644 --- a/content/cumulus-linux-516/Whats-New/rn.md +++ b/content/cumulus-linux-516/Whats-New/rn.md @@ -14,11 +14,11 @@ pdfhidden: True | Issue ID | Description | Affects | Fixed | |--- |--- |--- |--- | -| 5225576 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | +| 5225576, 5236486 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | | 5224270 | On switches with TACACS+ servers configured by hostname (FQDN) instead of IP address, the switch might generate an excessive rate of DNS queries for the configured TACACS+ server names, including for local user or UID lookups that never actually need to contact a TACACS+ server. | 5.16.1-5.18.0 | | | 5221592 | Adaptive routing ECMP updates during route deletion sometimes result in ECMP database corruption followed by ECMP operation failures. | 5.16.5-5.18.0 | | | 5221589 | In rare cases, you cannot cancel bulk counter session(MOCS). This issue results in missing GNMI metrics during export. | 5.16.5-5.18.0 | | -| 5217147, 5225676 | After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is
The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it
The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected
Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE
| 5.16.6-5.18.0 | | +| 5217147, 5225676 | When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. | 5.16.6-5.18.0 | | | 5199525 | When EVPN prefixes learned from a BGP neighbor are withdrawn, the gNMI EVPN installed-prefix count does not decrease. | 5.16.1-5.18.0 | | | 5183442 | If the running version of Cumulus Linux is not the newest, installing packages in cumulus-local-apt-archive (RADIUS or TACACS+ packages) might bring in a newer version from remote locations that does not match the running version of Cumulus Linux. To avoid this problem add an /etc/apt/preferences.d/10_prefer_cumulus_local_apt_archive file with the following content (if not already present) before doing a package update:
Package: *
Pin: release a=cumulus-local-apt-archive
Pin-Priority: 992
| 5.16.3-5.18.0 | | | 5172157 | The nv config apply command fails to apply configuration changes because NVUE fails to handle stale sessions and does not prompt you to clear them. To work around this issue, run the nv config detach command to clear the stale session, then run the nv config replace command if there is a stale pending revision. | 5.16.4-5.17.0 | 5.18.0| @@ -250,7 +250,7 @@ pdfhidden: True | 5217107, 5183514, 5221591 | The NVUE set command returns a traceback in the output when the per user repo_incarnation file (/home/$USER/.local/share/nvue/repo_incarnation) is non-writable. | 5.16.1-5.16.6 | | | 5217106, 4922221 | If the core generator service enters a busy loop, the switch cannot auto generate a cl-support file and core files might accumulate, which consume disk space. To work around this issue, restart the core generator service with the sudo systemctl restart cumulus-core.path and sudo systemctl restart cumulus-core.service commands so that it can start processing newly generated cores. | 5.16.0-5.16.6 | | | 5217101 | AR ECMP updates during route deletion might result in ECMP database corruption and ECMP operation failures resulting in the following log messages:
switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:789 ERR ECMP: AR cmd CREATE failed: Entry Already Exists, nexthops 18
switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:934 ERR ECMP: failed to CREATE adaptive ecmp in hw
switchd[23072]: hal_mlx_ecmp.c:2426 ERR ECMP: error allocating AR ecmp
switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:707 ERR ECMP: Failed to set adaptive container attributes for ecmp
| 5.16.5-5.16.6 | | -| 5217097, 5179671, 5221593, 5221604 | After optimized image upgrade, the /run/tmpfs which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | | +| 5217097, 5179671, 5221593, 5221604 | After optimized image upgrade, /run/tmpfs, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | | | 5217095, 5172073 | The LLDP Chassis ID might be non-unique when front-panel ports share MAC addresses. | 5.16.5-5.16.6, 5.17.0 | | | 5217094, 5167708 | When an ACL policy is reinstalled while ACL rule counter statistics are being collected, a kernel panic can occur and the switch might reboot unexpectedly. The panic occurs because the ACL counter buffer is released while the hardware counter read for that buffer is still in progress. | 5.16.3-5.16.6 | | | 5217093, 5175518, 5221594 | The switchd watchdog crashes due to a netlink buffer overflow triggered by continuous next hop group churn. | 5.16.1-5.16.6, 5.18.0 | | @@ -262,12 +262,12 @@ pdfhidden: True | Issue ID | Description | Affects | Fixed | |--- |--- |--- |--- | -| 5225576 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | +| 5225576, 5236486 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | | 5224270 | On switches with TACACS+ servers configured by hostname (FQDN) instead of IP address, the switch might generate an excessive rate of DNS queries for the configured TACACS+ server names, including for local user or UID lookups that never actually need to contact a TACACS+ server. | 5.16.1-5.18.0 | | | 5221592 | Adaptive routing ECMP updates during route deletion sometimes result in ECMP database corruption followed by ECMP operation failures. | 5.16.5-5.18.0 | | | 5221589 | In rare cases, you cannot cancel bulk counter session(MOCS). This issue results in missing GNMI metrics during export. | 5.16.5-5.18.0 | | | 5221130 | When multiple gNMI bulk counter sessions are requested frequently, gNMI metric export might fail. | 5.16.5-5.16.6 | 5.16.7-5.18.0| -| 5217147, 5225676 | After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is
The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it
The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected
Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE
| 5.16.6-5.18.0 | | +| 5217147, 5225676 | When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. | 5.16.6-5.18.0 | | | 5217101 | AR ECMP updates during route deletion might result in ECMP database corruption and ECMP operation failures resulting in the following log messages:
switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:789 ERR ECMP: AR cmd CREATE failed: Entry Already Exists, nexthops 18
switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:934 ERR ECMP: failed to CREATE adaptive ecmp in hw
switchd[23072]: hal_mlx_ecmp.c:2426 ERR ECMP: error allocating AR ecmp
switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:707 ERR ECMP: Failed to set adaptive container attributes for ecmp
| 5.16.5-5.16.6 | 5.16.7-5.18.0| | 5215201, 5221379, 5221588 | In rare scenarios, a link fault recovery sequence might trigger a switchd watchdog timeout on Spectrum-4 switches. | 5.16.3-5.16.6 | 5.16.7-5.18.0| | 5200160, 5215402, 5221596 | During an optimized image upgrade, the configuration restore step copies the files in /etc/hw-management-thermal/ from the previous release onto the new partition. These files do not contain the settings that Cumulus Linux requires to select the current thermal control algorithm, so the switch runs the previous algorithm instead. That algorithm does not apply the transceiver temperature setpoint or the module temperature cap, so the fan curve follows the transceivers' own factory temperature limits instead of the configured values. | 5.16.6 | 5.16.7-5.18.0| @@ -275,7 +275,7 @@ pdfhidden: True | 5183514, 5217107, 5221591 | The NVUE set command returns a traceback in the output when the per user repo_incarnation file (/home/$USER/.local/share/nvue/repo_incarnation) is non-writable. | 5.16.1-5.16.6 | 5.16.7-5.18.0| | 5183442 | If the running version of Cumulus Linux is not the newest, installing packages in cumulus-local-apt-archive (RADIUS or TACACS+ packages) might bring in a newer version from remote locations that does not match the running version of Cumulus Linux. To avoid this problem add an /etc/apt/preferences.d/10_prefer_cumulus_local_apt_archive file with the following content (if not already present) before doing a package update:
Package: *
Pin: release a=cumulus-local-apt-archive
Pin-Priority: 992
| 5.16.3-5.18.0 | | | 5182789, 5217108, 5221590, 5216045, 5215790 | BGP peer sessions might repeatedly flap and re-establish. During outbound TCP connect, the BGP service might clear the local nexthop information for the peer and fail nexthop setup when the remote endpoint is not yet available, which resets the session. Logs might show nexthop_set failed with remote: (null) and Transport endpoint is not.... | 5.16.1-5.16.6 | 5.16.7-5.18.0| -| 5179671, 5217097, 5221593, 5221604 | After optimized image upgrade, the /run/tmpfs which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | 5.16.7-5.18.0| +| 5179671, 5217097, 5221593, 5221604 | After optimized image upgrade, /run/tmpfs, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | 5.16.7-5.18.0| | 5175518, 5217093, 5221594 | The switchd watchdog crashes due to a netlink buffer overflow triggered by continuous next hop group churn. | 5.16.1-5.16.6, 5.18.0 | 5.16.7| | 5172157 | The nv config apply command fails to apply configuration changes because NVUE fails to handle stale sessions and does not prompt you to clear them. To work around this issue, run the nv config detach command to clear the stale session, then run the nv config replace command if there is a stale pending revision. | 5.16.4-5.17.0 | 5.18.0| | 5172073, 5217095 | The LLDP Chassis ID might be non-unique when front-panel ports share MAC addresses. | 5.16.5-5.16.6, 5.17.0 | 5.16.7, 5.18.0| @@ -536,7 +536,7 @@ pdfhidden: True | 5183514, 5217107, 5221591 | The NVUE set command returns a traceback in the output when the per user repo_incarnation file (/home/$USER/.local/share/nvue/repo_incarnation) is non-writable. | 5.16.1-5.16.6 | 5.16.7-5.18.0| | 5183442 | If the running version of Cumulus Linux is not the newest, installing packages in cumulus-local-apt-archive (RADIUS or TACACS+ packages) might bring in a newer version from remote locations that does not match the running version of Cumulus Linux. To avoid this problem add an /etc/apt/preferences.d/10_prefer_cumulus_local_apt_archive file with the following content (if not already present) before doing a package update:
Package: *
Pin: release a=cumulus-local-apt-archive
Pin-Priority: 992
| 5.16.3-5.18.0 | | | 5182789, 5217108, 5221590, 5216045, 5215790 | BGP peer sessions might repeatedly flap and re-establish. During outbound TCP connect, the BGP service might clear the local nexthop information for the peer and fail nexthop setup when the remote endpoint is not yet available, which resets the session. Logs might show nexthop_set failed with remote: (null) and Transport endpoint is not.... | 5.16.1-5.16.6 | 5.16.7-5.18.0| -| 5179671, 5217097, 5221593, 5221604 | After optimized image upgrade, the /run/tmpfs which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | 5.16.7-5.18.0| +| 5179671, 5217097, 5221593, 5221604 | After optimized image upgrade, /run/tmpfs, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | 5.16.7-5.18.0| | 5175518, 5217093, 5221594 | The switchd watchdog crashes due to a netlink buffer overflow triggered by continuous next hop group churn. | 5.16.1-5.16.6, 5.18.0 | 5.16.7| | 5172157 | The nv config apply command fails to apply configuration changes because NVUE fails to handle stale sessions and does not prompt you to clear them. To work around this issue, run the nv config detach command to clear the stale session, then run the nv config replace command if there is a stale pending revision. | 5.16.4-5.17.0 | 5.18.0| | 5172073, 5217095 | The LLDP Chassis ID might be non-unique when front-panel ports share MAC addresses. | 5.16.5-5.16.6, 5.17.0 | 5.16.7, 5.18.0| diff --git a/content/cumulus-linux-516/rn.xml b/content/cumulus-linux-516/rn.xml index 7e2be04fea..fb0a97219b 100644 --- a/content/cumulus-linux-516/rn.xml +++ b/content/cumulus-linux-516/rn.xml @@ -7,7 +7,7 @@ Fixed -5225576 +5225576, 5236486 After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the {{/var/images}} directory before the upgrade. In this state, the {{nv show system image files}} command might not list any images. To work around this issue, fetch the image with the {{nv action fetch system image <remote-url>}} command before installing it. Alternatively, correct the directory permissions with {{sudo chmod 0755 /var/images}}. 5.16.6-5.18.0 @@ -32,14 +32,7 @@ 5217147, 5225676 -After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is. - -The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it. - -The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected. - -Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE. - +When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. 5.16.6-5.18.0 @@ -1480,7 +1473,7 @@ switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:707 ERR ECMP: Failed to set adaptive 5217097, 5179671, 5221593, 5221604 -After optimized image upgrade, the {{/run/tmpfs}} which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. +After optimized image upgrade, {{/run/tmpfs}}, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. 5.16.5-5.16.6 @@ -1517,7 +1510,7 @@ switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:707 ERR ECMP: Failed to set adaptive Fixed -5225576 +5225576, 5236486 After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the {{/var/images}} directory before the upgrade. In this state, the {{nv show system image files}} command might not list any images. To work around this issue, fetch the image with the {{nv action fetch system image <remote-url>}} command before installing it. Alternatively, correct the directory permissions with {{sudo chmod 0755 /var/images}}. 5.16.6-5.18.0 @@ -1548,14 +1541,7 @@ switchd[23072]: hal_mlx_sdk_nexthop_wrap.c:707 ERR ECMP: Failed to set adaptive 5217147, 5225676 -After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is. - -The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it. - -The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected. - -Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE. - +When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. 5.16.6-5.18.0 @@ -1612,7 +1598,7 @@ Pin-Priority: 992 5179671, 5217097, 5221593, 5221604 -After optimized image upgrade, the {{/run/tmpfs}} which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. +After optimized image upgrade, {{/run/tmpfs}}, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. 5.16.5-5.16.6 5.16.7-5.18.0 @@ -3191,7 +3177,7 @@ Pin-Priority: 992 5179671, 5217097, 5221593, 5221604 -After optimized image upgrade, the {{/run/tmpfs}} which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. +After optimized image upgrade, {{/run/tmpfs}}, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. 5.16.5-5.16.6 5.16.7-5.18.0 diff --git a/content/cumulus-linux-517/Whats-New/rn.md b/content/cumulus-linux-517/Whats-New/rn.md index 6a9e4e7def..cd2ecf4503 100644 --- a/content/cumulus-linux-517/Whats-New/rn.md +++ b/content/cumulus-linux-517/Whats-New/rn.md @@ -14,11 +14,11 @@ pdfhidden: True | Issue ID | Description | Affects | Fixed | |--- |--- |--- |--- | -| 5225576 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | +| 5225576, 5236486 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | | 5224270 | On switches with TACACS+ servers configured by hostname (FQDN) instead of IP address, the switch might generate an excessive rate of DNS queries for the configured TACACS+ server names, including for local user or UID lookups that never actually need to contact a TACACS+ server. | 5.16.1-5.18.0 | | | 5221592 | Adaptive routing ECMP updates during route deletion sometimes result in ECMP database corruption followed by ECMP operation failures. | 5.16.5-5.18.0 | | | 5221589 | In rare cases, you cannot cancel bulk counter session(MOCS). This issue results in missing GNMI metrics during export. | 5.16.5-5.18.0 | | -| 5217147, 5225676 | After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is
The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it
The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected
Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE
| 5.16.6-5.18.0 | | +| 5217147, 5225676 | When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. | 5.16.6-5.18.0 | | | 5199525 | When EVPN prefixes learned from a BGP neighbor are withdrawn, the gNMI EVPN installed-prefix count does not decrease. | 5.16.1-5.18.0 | | | 5183442 | If the running version of Cumulus Linux is not the newest, installing packages in cumulus-local-apt-archive (RADIUS or TACACS+ packages) might bring in a newer version from remote locations that does not match the running version of Cumulus Linux. To avoid this problem add an /etc/apt/preferences.d/10_prefer_cumulus_local_apt_archive file with the following content (if not already present) before doing a package update:
Package: *
Pin: release a=cumulus-local-apt-archive
Pin-Priority: 992
| 5.16.3-5.18.0 | | | 5172157 | The nv config apply command fails to apply configuration changes because NVUE fails to handle stale sessions and does not prompt you to clear them. To work around this issue, run the nv config detach command to clear the stale session, then run the nv config replace command if there is a stale pending revision. | 5.16.4-5.17.0 | 5.18.0| diff --git a/content/cumulus-linux-517/rn.xml b/content/cumulus-linux-517/rn.xml index ee336ee514..31427dabcb 100644 --- a/content/cumulus-linux-517/rn.xml +++ b/content/cumulus-linux-517/rn.xml @@ -7,7 +7,7 @@ Fixed -5225576 +5225576, 5236486 After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the {{/var/images}} directory before the upgrade. In this state, the {{nv show system image files}} command might not list any images. To work around this issue, fetch the image with the {{nv action fetch system image <remote-url>}} command before installing it. Alternatively, correct the directory permissions with {{sudo chmod 0755 /var/images}}. 5.16.6-5.18.0 @@ -32,14 +32,7 @@ 5217147, 5225676 -After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is. - -The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it. - -The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected. - -Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE. - +When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. 5.16.6-5.18.0 diff --git a/content/cumulus-linux-518/Whats-New/rn.md b/content/cumulus-linux-518/Whats-New/rn.md index f0d9852629..a868e19658 100644 --- a/content/cumulus-linux-518/Whats-New/rn.md +++ b/content/cumulus-linux-518/Whats-New/rn.md @@ -22,7 +22,7 @@ pdfhidden: True | 5197214 | If the cumulus-platform-common package is updated, login notification might fail on each login and show errors. | 5.18.0-5.18.1 | | | 5190122 | On a Spectrum-6 switch (such as the SN6600-LD), gNMI reports an incorrect shared-buffer cell size of 192 bytes at /qos/shared-buffer/state/cell-size instead of the correct 256 bytes for Spectrum-6. Because this cell size is used to convert internal buffer cell counts into byte values, several shared-buffer telemetry values derived from it (such as per-queue or traffic-class buffer usage and maximum occupancy) report incorrectly over gNMI. The values are inconsistent with the values that NVUE reports and the hardware buffer counters. | 5.18.0-5.18.1 | | | 5184005 | In certain SRv6 configuration rollback scenarios, Zebra might fail to restore all explicit SRv6 SIDs. This can occur when multiple SRv6 locators share the same parent SID block and one of the locators is replayed during rollback. To work around this issue, restart FRR to clear the stale in-memory SRv6 SID allocation state, then reapply the intended configuration if needed. | 5.18.0-5.18.1 | | -| 5183654 | When you upgrade gNMI with package upgrade, the nv-gnmi.service fails and the switch collects a tech support bundle after the first step of the process (nv action upgrade system packages to latest. The package upgrade process triggers a switch reboot straight afterwards as a second and final step. There is no functional impact after switch reboot completes. | 5.18.0-5.18.1 | | +| 5183654, 5234851 | When you upgrade gNMI with package upgrade, the nv-gnmi.service fails and the switch collects a tech support bundle after the first step of the process (nv action upgrade system packages to latest. The package upgrade process triggers a switch reboot straight afterwards as a second and final step. There is no functional impact after switch reboot completes. | 5.18.0-5.18.1 | | | 5183442 | If the running version of Cumulus Linux is not the newest, installing packages in cumulus-local-apt-archive (RADIUS or TACACS+ packages) might bring in a newer version from remote locations that does not match the running version of Cumulus Linux. To avoid this problem add an /etc/apt/preferences.d/10_prefer_cumulus_local_apt_archive file with the following content (if not already present) before doing a package update:
Package: *
Pin: release a=cumulus-local-apt-archive
Pin-Priority: 992
| 5.16.3-5.18.1 | | | 5180032, 5180016 | Running a transceiver query with the ethtool -m or nv show interface transceiver command might intermittently discard the reply, causing the command to hang for approximately 30 seconds and fail with netlink error: Connection timed out before recovering. | 5.18.0-5.18.1 | | | 5174873 | On an EVPN MH topology with route leaking, if we enable Layer 3 VXLAN Device Mode from existing Single VXLAN Device for All Layer 3 VNIs model, FDB/mac entry is not installed in hardware for some remote VTEP router macs. This affects L3 VxLAN traffic through the VTEPs whose RMACs are not installed on the DUT. | 5.18.0-5.18.1 | | @@ -181,20 +181,20 @@ pdfhidden: True | Issue ID | Description | Affects | Fixed | |--- |--- |--- |--- | -| 5225576 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | +| 5225576, 5236486 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.18.0 | | | 5224270 | On switches with TACACS+ servers configured by hostname (FQDN) instead of IP address, the switch might generate an excessive rate of DNS queries for the configured TACACS+ server names, including for local user or UID lookups that never actually need to contact a TACACS+ server. | 5.16.1-5.18.0 | | | 5221595 | In certain cases, the switch does not forward traffic on some interfaces in an ECMP group. To work around this issue, flap the affected interface. | 5.18.0 | | | 5221592 | Adaptive routing ECMP updates during route deletion sometimes result in ECMP database corruption followed by ECMP operation failures. | 5.16.5-5.18.0 | | | 5221589 | In rare cases, you cannot cancel bulk counter session(MOCS). This issue results in missing GNMI metrics during export. | 5.16.5-5.18.0 | | | 5220910 | NVUE accepts invalid telemetry snapshot file paths; however, the corresponding show commands fail with an error in the NVUE logs. | 5.18.0 | | -| 5217147, 5225676 | After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is
The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it
The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected
Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE
| 5.16.6-5.18.0 | | +| 5217147, 5225676 | When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. | 5.16.6-5.18.0 | | | 5217091 | In some scenarios after configuration bring-up or an interface carrier transition, traffic might not be forwarded on some interfaces in an ECMP group. To work around this issue, flap the affected interface. | 5.18.0 | 5.16.7| | 5215333 | If you provide a value that contains a non-ASCII character in NVUE, you see an unhandled Python exception instead of a normal CLI validation error. | 5.18.0 | | | 5199525 | When EVPN prefixes learned from a BGP neighbor are withdrawn, the gNMI EVPN installed-prefix count does not decrease. | 5.16.1-5.18.0 | | | 5197214 | If the cumulus-platform-common package is updated, login notification might fail on each login and show errors. | 5.18.0 | | | 5190122 | On a Spectrum-6 switch (such as the SN6600-LD), gNMI reports an incorrect shared-buffer cell size of 192 bytes at /qos/shared-buffer/state/cell-size instead of the correct 256 bytes for Spectrum-6. Because this cell size is used to convert internal buffer cell counts into byte values, several shared-buffer telemetry values derived from it (such as per-queue or traffic-class buffer usage and maximum occupancy) report incorrectly over gNMI. The values are inconsistent with the values that NVUE reports and the hardware buffer counters. | 5.18.0 | | | 5184005 | In certain SRv6 configuration rollback scenarios, Zebra might fail to restore all explicit SRv6 SIDs. This can occur when multiple SRv6 locators share the same parent SID block and one of the locators is replayed during rollback. To work around this issue, restart FRR to clear the stale in-memory SRv6 SID allocation state, then reapply the intended configuration if needed. | 5.18.0 | | -| 5183654 | When you upgrade gNMI with package upgrade, the nv-gnmi.service fails and the switch collects a tech support bundle after the first step of the process (nv action upgrade system packages to latest. The package upgrade process triggers a switch reboot straight afterwards as a second and final step. There is no functional impact after switch reboot completes. | 5.18.0 | | +| 5183654, 5234851 | When you upgrade gNMI with package upgrade, the nv-gnmi.service fails and the switch collects a tech support bundle after the first step of the process (nv action upgrade system packages to latest. The package upgrade process triggers a switch reboot straight afterwards as a second and final step. There is no functional impact after switch reboot completes. | 5.18.0 | | | 5183442 | If the running version of Cumulus Linux is not the newest, installing packages in cumulus-local-apt-archive (RADIUS or TACACS+ packages) might bring in a newer version from remote locations that does not match the running version of Cumulus Linux. To avoid this problem add an /etc/apt/preferences.d/10_prefer_cumulus_local_apt_archive file with the following content (if not already present) before doing a package update:
Package: *
Pin: release a=cumulus-local-apt-archive
Pin-Priority: 992
| 5.16.3-5.18.0 | | | 5180032, 5180016 | Running a transceiver query with the ethtool -m or nv show interface transceiver command might intermittently discard the reply, causing the command to hang for approximately 30 seconds and fail with netlink error: Connection timed out before recovering. | 5.18.0 | | | 5175518, 5217093, 5221594 | The switchd watchdog crashes due to a netlink buffer overflow triggered by continuous next hop group churn. | 5.16.1-5.16.6, 5.18.0 | 5.16.7| diff --git a/content/cumulus-linux-518/rn.xml b/content/cumulus-linux-518/rn.xml index 99d8c2fc83..0250bc3128 100644 --- a/content/cumulus-linux-518/rn.xml +++ b/content/cumulus-linux-518/rn.xml @@ -55,7 +55,7 @@ -5183654 +5183654, 5234851 When you upgrade gNMI with package upgrade, the {{nv-gnmi.service}} fails and the switch collects a tech support bundle after the first step of the process ({{nv action upgrade system packages to latest}}. The package upgrade process triggers a switch reboot straight afterwards as a second and final step. There is no functional impact after switch reboot completes. 5.18.0-5.18.1 @@ -1005,7 +1005,7 @@ cumulus@switch:~$ sudo ethtool -s swp1 speed 100000 autoneg off lanes 2 Fixed -5225576 +5225576, 5236486 After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the {{/var/images}} directory before the upgrade. In this state, the {{nv show system image files}} command might not list any images. To work around this issue, fetch the image with the {{nv action fetch system image <remote-url>}} command before installing it. Alternatively, correct the directory permissions with {{sudo chmod 0755 /var/images}}. 5.16.6-5.18.0 @@ -1042,14 +1042,7 @@ cumulus@switch:~$ sudo ethtool -s swp1 speed 100000 autoneg off lanes 2 5217147, 5225676 -After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is. - -The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it. - -The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected. - -Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE. - +When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. 5.16.6-5.18.0 @@ -1090,7 +1083,7 @@ Because the switch appears healthy in the interval between the factory reset and -5183654 +5183654, 5234851 When you upgrade gNMI with package upgrade, the {{nv-gnmi.service}} fails and the switch collects a tech support bundle after the first step of the process ({{nv action upgrade system packages to latest}}. The package upgrade process triggers a switch reboot straight afterwards as a second and final step. There is no functional impact after switch reboot completes. 5.18.0 diff --git a/content/cumulus-linux-519/Whats-New/rn.md b/content/cumulus-linux-519/Whats-New/rn.md index 38d3dde211..d22ec04167 100644 --- a/content/cumulus-linux-519/Whats-New/rn.md +++ b/content/cumulus-linux-519/Whats-New/rn.md @@ -14,6 +14,7 @@ pdfhidden: True | Issue ID | Description | Affects | Fixed | |--- |--- |--- |--- | +| 5236486 | After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the /var/images directory before the upgrade. In this state, the nv show system image files command might not list any images. To work around this issue, fetch the image with the nv action fetch system image command before installing it. Alternatively, correct the directory permissions with sudo chmod 0755 /var/images. | 5.16.6-5.19.0 | | | 5217091 | In some scenarios after configuration bring-up or an interface carrier transition, traffic might not be forwarded on some interfaces in an ECMP group. To work around this issue, flap the affected interface. | 5.18.0-5.19.0 | 5.16.7| | 5183442 | If the running version of Cumulus Linux is not the newest, installing packages in cumulus-local-apt-archive (RADIUS or TACACS+ packages) might bring in a newer version from remote locations that does not match the running version of Cumulus Linux. To avoid this problem add an /etc/apt/preferences.d/10_prefer_cumulus_local_apt_archive file with the following content (if not already present) before doing a package update:
Package: *
Pin: release a=cumulus-local-apt-archive
Pin-Priority: 992
| 5.16.3-5.19.0 | | | 5159852, 4667526 | When you configure multiple parallel IPv6 numbered eBGP sessions between the same pair of switches, some sessions can remain in an Idle state during simultaneous link or session bringup. Affected sessions repeatedly report Cease/Connection Collision Resolution. To work around this issue, configure update-source separately for every numbered IPv6 BGP neighbor on both switches, using the local IPv6 address assigned to that neighbor’s link.
Do not configure a single shared source on the peer group when each parallel link has a different local address. After applying the configuration, clear the affected BGP sessions to remove any existing incorrectly associated connections. | 5.18.0-5.19.0 | | @@ -141,7 +142,7 @@ pdfhidden: True |--- |--- |--- | | 5224270 | On switches with TACACS+ servers configured by hostname (FQDN) instead of IP address, the switch might generate an excessive rate of DNS queries for the configured TACACS+ server names, including for local user or UID lookups that never actually need to contact a TACACS+ server. | 5.16.1-5.18.0 | | | 5220910 | NVUE accepts invalid telemetry snapshot file paths; however, the corresponding show commands fail with an error in the NVUE logs. | 5.18.0 | | -| 5217147, 5225676 | After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is
The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it
The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected
Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE
| 5.16.6-5.18.0 | | +| 5217147, 5225676 | When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. | 5.16.6-5.18.0 | | | 5215333 | If you provide a value that contains a non-ASCII character in NVUE, you see an unhandled Python exception instead of a normal CLI validation error. | 5.18.0 | | | 5200160, 5215402, 5221596 | During an optimized image upgrade, the configuration restore step copies the files in /etc/hw-management-thermal/ from the previous release onto the new partition. These files do not contain the settings that Cumulus Linux requires to select the current thermal control algorithm, so the switch runs the previous algorithm instead. That algorithm does not apply the transceiver temperature setpoint or the module temperature cap, so the fan curve follows the transceivers' own factory temperature limits instead of the configured values. | 5.16.6 | | | 5199525 | When EVPN prefixes learned from a BGP neighbor are withdrawn, the gNMI EVPN installed-prefix count does not decrease. | 5.16.1-5.18.0 | | @@ -151,8 +152,9 @@ pdfhidden: True | 5183514, 5217107, 5221591 | The NVUE set command returns a traceback in the output when the per user repo_incarnation file (/home/$USER/.local/share/nvue/repo_incarnation) is non-writable. | 5.16.1-5.16.6 | | | 5182789, 5217108, 5221590, 5216045, 5215790 | BGP peer sessions might repeatedly flap and re-establish. During outbound TCP connect, the BGP service might clear the local nexthop information for the peer and fail nexthop setup when the remote endpoint is not yet available, which resets the session. Logs might show nexthop_set failed with remote: (null) and Transport endpoint is not.... | 5.16.1-5.16.6 | | | 5180032, 5180016 | Running a transceiver query with the ethtool -m or nv show interface transceiver command might intermittently discard the reply, causing the command to hang for approximately 30 seconds and fail with netlink error: Connection timed out before recovering. | 5.18.0 | | -| 5179671, 5217097, 5221593, 5221604 | After optimized image upgrade, the /run/tmpfs which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | | +| 5179671, 5217097, 5221593, 5221604 | After optimized image upgrade, /run/tmpfs, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. | 5.16.5-5.16.6 | | | 5175518, 5217093, 5221594 | The switchd watchdog crashes due to a netlink buffer overflow triggered by continuous next hop group churn. | 5.16.1-5.16.6, 5.18.0 | | +| 5174595, 5233487 | On an EVPN MH topology with route leaking, if we enable Layer 3 VXLAN Device Mode from existing Single VXLAN Device for All Layer 3 VNIs model, cl-route-check errors are observed. Neighbor entry is not resolved for the next hop and hence traffic on the routes shown in cl-route-check errors will be impacted. | 5.18.0 | | | 5170159 | NAS IP addresses are missing from RADIUS accounting packets. | 5.18.0 | | | 5167882 | In rare cases switchd crashes and the cl-support file triggers twice, which interrupts legacy mlxlink collection and causes multiple per port mlxlink support files to generate. | 5.16.1-5.18.0 | | | 5165432, 5154620 | With EVPN L3VXI mode enabled, routes leaked between tenant vrfs (using import vrfs) are installed without the encap info leading to blackholing of the traffic. | 5.18.0 | | diff --git a/content/cumulus-linux-519/rn.xml b/content/cumulus-linux-519/rn.xml index 4a15fcf627..eb8fb6a7d0 100644 --- a/content/cumulus-linux-519/rn.xml +++ b/content/cumulus-linux-519/rn.xml @@ -7,6 +7,12 @@ Fixed +5236486 +After an optimized image upgrade from Cumulus Linux 5.13.1 or earlier to a later release, the install step fails with a permission error if the image being installed is already present in the {{/var/images}} directory before the upgrade. In this state, the {{nv show system image files}} command might not list any images. To work around this issue, fetch the image with the {{nv action fetch system image <remote-url>}} command before installing it. Alternatively, correct the directory permissions with {{sudo chmod 0755 /var/images}}. +5.16.6-5.19.0 + + + 5217091 In some scenarios after configuration bring-up or an interface carrier transition, traffic might not be forwarded on some interfaces in an ECMP group. To work around this issue, flap the affected interface. 5.18.0-5.19.0 @@ -803,14 +809,7 @@ cumulus@switch:~$ sudo ethtool -s swp1 speed 100000 autoneg off lanes 2 5217147, 5225676 -After a two-partition (A/B) optimized upgrade, performing a factory reset can cause the switch to lose the record of what model it is. - -The A/B upgrade takes its reference configuration backup before the boot configuration has been written to the newly staged partition, so that backup does not contain /etc/default/grub. A factory reset restores /etc from that backup and deletes anything not present in it, which removes the live /etc/default/grub. That file carries the cl_platform= kernel argument — the only record of the switch model — and no package ships it, so nothing on the switch can recreate it. - -The switch continues to operate normally after the reset, because it is still running from a boot configuration generated while the file existed. The failure appears later, the first time anything regenerates that boot configuration — an update-grub, a kernel or package upgrade, or a subsequent image upgrade. From then on the switch boots without cl_platform, platform detection fails, update-ports.service fails, switchd does not start, and no ASIC interfaces are present. Management access over eth0 is unaffected. - -Because the switch appears healthy in the interval between the factory reset and that regeneration, an affected switch can run for an extended period before the fault becomes visible. Once it does, reboots do not clear it — nothing regenerates the boot configuration at boot time — and recovery requires reinstalling via ONIE. - +When you perform a factory reset after an optimized image upgrade, the switch loses record of its model information. 5.16.6-5.18.0 @@ -860,7 +859,7 @@ Because the switch appears healthy in the interval between the factory reset and 5179671, 5217097, 5221593, 5221604 -After optimized image upgrade, the {{/run/tmpfs}} which holds the upgraded partition runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. +After optimized image upgrade, {{/run/tmpfs}}, which holds the upgraded partition, runs out of space. As a result, the switch does not migrate all the certificates to the new partition. This issue prevents NVUE from installing the user defined certificates defaulting to a self signed certificate. 5.16.5-5.16.6 @@ -869,6 +868,11 @@ Because the switch appears healthy in the interval between the factory reset and 5.16.1-5.16.6, 5.18.0 +5174595, 5233487 +On an EVPN MH topology with route leaking, if we enable Layer 3 VXLAN Device Mode from existing Single VXLAN Device for All Layer 3 VNIs model, cl-route-check errors are observed. Neighbor entry is not resolved for the next hop and hence traffic on the routes shown in cl-route-check errors will be impacted. +5.18.0 + + 5170159 NAS IP addresses are missing from RADIUS accounting packets. 5.18.0