Signed-off-by: bingwang <bingwang@microsoft.com>
Why I did it
This PR is to add two extra lossless queues for bounced back traffic.
HLD sonic-net/SONiC#950
SKUs include
Arista-7050CX3-32S-C32
Arista-7050CX3-32S-D48C8
Arista-7260CX3-D108C8
Arista-7260CX3-C64
Arista-7260CX3-Q64
How I did it
Update the buffers.json.j2 template and buffers_config.j2 template to generate new BUFFER_QUEUE table.
For T1 devices, queue 2 and queue 6 are set as lossless queues on T0 facing ports.
For T0 devices, queue 2 and queue 6 are set as lossless queues on T1 facing ports.
Queue 7 is added as a new lossy queue as DSCP 48 is mapped to TC 7, and then mapped into Queue 7
How to verify it
Verified by UT
Verified by coping the new template and generate buffer config with sonic-cfggen
Signed-off-by: Yong Zhao yozhao@microsoft.com
Why I did it
This PR aims to fix an issue (#10088) by enhancing the script memory_checker.
Specifically, if container is not created successfully during device is booted/rebooted, then memory_checker do not need check its memory usage.
How I did it
In the script memory_checker, a function is added to get names of running containers. If the specified container name is not in current running container list, then this script will exit without checking its memory usage.
How to verify it
I tested on a lab device by following the steps:
Stops telemetry container with command sudo systemctl stop telemetry.service
Removes telemetry container with command docker rm telemetry
Checks whether the script memory_checker ran by Monit will generate the syslog message saying it will exit without checking memory usage of telemetry.
- Why I did it
Recent change to delay PMON service in case of fast/warm reboot introduce an issue when restarting only SWSS service after fast/warm reboot for Nvidia platform.
Since the timer is triggered only when the system boot, in a scenario when the system is after a fast/warm reboot and the user restart SWSS service, as part of syncd.sh script, PMON service will stop but the timer will not start again.
- How I did it
On syncd.sh script, in case of fast/warm indication, check if pmon.timer is running.
If it is running it means we are at the first boot and continue normally.
If it is not running, meaning the service was restarted, start the timer to keep the system behavior consistent.
- How to verify it
Run fast/warm reboot.
service swss restart.
Observe PMON service starting.
Signed-off-by: Shlomi Bitton <shlomibi@nvidia.com>
* When reloading config after crashes, VTEP interfaces are sometimes not created since the tunnel still exists in the STATE_DB.
* Adding VXLAN_TUNNEL_TABLE to the list of tables to be cleaned in swss.sh fixes the problem.
- Why I did it
Yang Model about password hardening feature, the sonic CLI of this feature was autogenerated from this Yang model
- How I did it
Create new Yang model in src/sonic-yang-models/yang-models/sonic-passwh.yang.
- How to verify it
There are unitests(yang test) in this P.R covering all the passwords policies with good and bad values cases.
Or is possible manually using the config/show password commands that were autogenerated from this Yang model. (this CLI code added in sonic-utilities)
This reverts commit 15cf9b0d70.
Why I did it
Revert the PR #10775, for it has impact on onie installation.
It is caused by the symbol links not supported in some of the onie unzip.
We will enable after fixing the issue, see #10914
What/Why I did:
Issue1: By setting up of ipvlan interface in interface-config.sh we are not tolerant to failures. Reason being interface-config.service is one-shot and do not have restart capability.
Scenario: For example if let's say database service goes in fail state then interface-services also gets failed because of dependency check but later database service gets restart but interface service will remain in stuck state and the ipvlan interface nevers get created.
Solution: Moved all the logic in database service from interface-config service which looks more align logically also since the namespace is created here and all the network setting (sysctl) are happening here.With this if database starts we recreate the interface.
Issue 2: Use of IPVLAN vs MACVLAN
Currently we are using ipvlan mode. However above failure scenario is not handle correctly by ipvlan mode. Once the ipvlan interface is created and ip address assign to it and if we restart interface-config or database (new PR) service Linux Kernel gives error "Error: Address already assigned to an ipvlan device." based on this:https://github.com/torvalds/linux/blob/master/drivers/net/ipvlan/ipvlan_main.c#L978Reason being if we do not do cleanup of ip address assignment (need to be unique for IPVLAN) it remains in Kernel Database and never goes to free pool even though namespace is deleted.
Solution: Considering this hard dependency of unique ip macvlan mode is better for us and since everything is managed by Linux Kernel and no dependency for on user configured IP address.
Issue3: Namespace database Service do not check reachability to Supervisor Redis Chassis Server.
Currently there is no explicit check as we never do Redis PING from namespace to Supervisor Redis Chassis Server. With this check it's possible we will start database and all other docker even though there is no connectivity and will hit the error/failure late in cycle
Solution: Added explicit PING from namespace that will check this reachability.
Issue 4:flushdb give exception when trying to accces Chassis Server DB over Unix Sokcet.
Solution: Handle gracefully via try..except and log the message.
Why I did it
At present, there is no mechanism in an event driven model to know that the system is up with all the essential sonic services and also, all the docker apps are ready along with port ready status to start the network traffic. With the asynchronous architecture of SONiC, we will not be able to verify if the config has been applied all the way down to the HW. But we can get the closest up status of each app and arrive at the system readiness.
How I did it
A new python based system monitor tool is introduced under system-health framework to monitor all the essential system host services including docker wrapper services on an event based model and declare the system is ready. This framework gives provision for docker apps to notify its closest up status. CLIs are provided to fetch the current system status and also service running status and its app ready status along with failure reason if any.
How to verify it
"show system-health sysready-status" click CLI
Syslogs for system ready
Why I did it
To upgrade SSD firmware in initramfs while rebooting from SONiC to SONiC and during NOS to SONiC migration.
How I did it
New option 'ssd-upgrader-part’ is introduced in grub command line, to indicate the partition and its filesystem type in which the SSD firmware updater is present. ‘ssd-upgrader-part’ syntax is ssd-upgrader-part=<partition>,<filesystem type>. Example: ssd-upgrader-part=/dev/sda8,ext4
A new initramfs script ‘ssd-upgrade’ is included in init-premount and it invokes the SSD firmware updater (ssd-fw-upgrade) present in the partition indicated by the boot option 'ssd-upgrader-part'
How to verify it
In SONiC, the SSD firmware updater is copied to “/host/” directory.
Fast-reboot is to be initiated with the ‘-u’ option ([scripts/fast-reboot] Add option to include ssd-upgrader-part boot option with SONiC partition sonic-utilities#2150)
After reboot, while booting into SONiC the SSD firmware updater will be executed in initramfs.
Why I did it
The image size is too large, when there are multiple lazy packages and multiple platforms. It is not necessary to keep the lazy installation packages in multiple copies.
For cisco image, the image size will reduce from 3.5G to 1.7G.
How I did it
Use symbol links to only keep one package for each of the lazy package.
Make a new folder fsroot/platform/common
Copy the lazy packages into the folder.
When using a package in each of the platform, such as x86_64-grub, x86_64-8800_rp-r0, x86_64-8201_on-r0, etc, only make a symbol link to the package in the common folder.
Why I did it
The PR is aimed to fix a bug that mgmt port eth0 may loss IP even if user configured static IP of eth0. This is not a always reproduceable issue, the reproducing flow is like:
Systemd starts networking service, which runs a dhcp based configuration and assigned an ip from dhcp.
Systemd starts interface-config service who depends on networking service
Interface-config service runs command “ifdown –force eth0”, check line. but networking service is still running so that this line failed with error: “error: Another instance of this program is already running.”. This error is printed by ifupdown2 lib who is the main process of networking service. So, ifdown actually does not work here, the ip of eth0 is not down.
Interface-config service updates /etc/networking/interface to static configuration.
Interface-config service runs command “systemctl restart networking”. This command kills the previous networking related processes (log: networking.service: Main process exited, code=killed, status=15/TERM), and try to reconfigure the ip address with static configuration. But it detects that the configured IP and the existing IP are the same, and it does not really configure the ip to kernel. Hence, the ip is still getting from dhcp. (this could be a bug of ifupdown2: previous ip is from dhcp, new ip is a static ip, it treats them as same instead of re-configuring the IP)
When the lease of the ip expires, the ip of eth0 is removed by kernel and the issue reproduces.
The issue is not always reproduceable because networking service usually runs fast so that it won't hit step#3.
How I did it
Check networking service state before running "ifdown –force eth0", wait for it done if it is activating.
How to verify it
Manual test.
- Why I did it
Profiling the system state on init after fast-reboot during create_switch function execution, it is possible to see few python scripts running at the same time.
This parallel execution consume CPU time and the duration of create_switch is longer than it should be.
Following this finding, and the motivation to ensure these services will not interfere in the future, PMON is delayed in 90 seconds until the system finish the init flow after fastboot.
- How I did it
Add a timer for PMON service.
Exclude for MLNX platform the start trigger of PMON when SYNCD starts in case of fastboot.
Copy the timer file to the host bin image.
- How to verify it
Run fast-reboot on MLNX platform and observe faster create_switch execution time.
- Why I did it
Profiling the system state on init after fast-reboot during create_switch function execution, it is possible to see few python scripts running at the same time.
This parallel execution consume CPU time and the duration of create_switch is longer than it should be.
Following this finding, and the motivation to ensure these services will not interfere in the future, LLDP is delayed in 90 seconds until the system finish the init flow after fastboot.
- How I did it
Add a timer for LLDP service.
Copy the timer file to the host bin image.
- How to verify it
Run fast-reboot on MLNX platform and observe faster create_switch execution time.
This PR is dependent on PR: #10567
Why I did it
Can not start sonic-hostservice
How I did it
Install python3-dbus and systemd-python, and replace invalid path
How to verify it
Start the service with below commands:
sudo systemctl start sonic-hostservice
sudo systemctl status sonic-hostservice
Signed-off-by: Gang Lv ganglv@microsoft.com
* Remove SSH host keys after installing the custom version of sshd
Signed-off-by: Saikrishna Arcot <sarcot@microsoft.com>
* Use an override for for sshd instead of overwriting the service file
Don't overwrite upstream's .service file, and instead use an override
file for making sure the host key(s) are generated.
Signed-off-by: Saikrishna Arcot <sarcot@microsoft.com>
Signed-off-by: Yong Zhao <yozhao@microsoft.com>
Why I did it
This PR aims to fix the Monit issue which shows Monit can't reset its counter when monitoring memory usage of telemetry container.
Specifically the Monit configuration file related to monitoring memory usage of telemetry container is as following:
check program container_memory_telemetry with path "/usr/bin/memory_checker telemetry 419430400"
if status == 3 for 10 times within 20 cycles then exec "/usr/bin/restart_service telemetry"
If memory usage of telemetry container is larger than 400MB for 10 times within 20 cycles (minutes), then it will be restarted.
Recently we observed, after telemetry container was restarted, its memory usage continuously increased from 400MB to 11GB within 1 hour, but it was not restarted anymore during this 1 hour sliding window.
The reason is Monit can't reset its counter to count again and Monit can reset its counter if and only if the status of monitored service was changed from Status failed to Status ok. However, during this 1 hour sliding window, the status of monitored service was not changed from Status failed to Status ok.
Currently for each service monitored by Monit, there will be an entry showing the monitoring status, monitoring mode etc. For example, the following output from command sudo monit status shows the status of monitored service to monitor memory usage of telemetry:
Program 'container_memory_telemetry'
status Status ok
monitoring status Monitored
monitoring mode active
on reboot start
last exit value 0
last output -
data collected Sat, 19 Mar 2022 19:56:26
Every 1 minute, Monit will run the script to check the memory usage of telemetry and update the counter if memory usage is larger than 400MB. If Monit checked the counter and found memory usage of telemetry is larger than 400MB for 10 times
within 20 minutes, then telemetry container was restarted. Following is an example status of monitored service:
Program 'container_memory_telemetry'
status Status failed
monitoring status Monitored
monitoring mode active
on reboot start
last exit value 0
last output -
data collected Tue, 01 Feb 2022 22:52:55
After telemetry container was restarted. we found memory usage of telemetry increased rapidly from around 100MB to more than 400MB during 1 minute and status of monitored service did not have a chance to be changed from Status failed to Status ok.
How I did it
In order to provide a workaround for this issue, Monit recently introduced another syntax format repeat every <n> cycles related to exec. This new syntax format will enable Monit repeat executing the background script if the error persists for a given number of cycles.
How to verify it
I verified this change on lab device str-s6000-acs-12. Another pytest PR (Azure/sonic-mgmt#5492) is submitted in sonic-mgmt repo for review.
The interface renaming logic fails if one interface is missing.
Because of the `set -e` the whole initramfs hook would abort early on
error.
This change fixes the current behavior to make sure missing interfaces
are properly skipped and ensure existing interface are renamed.
Fix the issues #10501 and #9733
If having gearbox, we need:
* add gbsyncd as a peer since swss also has dependency on gbsyncd
* add service gbsyncd to FEATURE table if it is missing
Why I did it
To support address sanitizer for Mellanox syncd
How I did it
/var/log/asan is mapped for syncd container (the same as for swss)
container stop() has a timeout (60s) for syncd (the same as for swss)
This is so libasan has enough time to generate a report.
added ASAN's log path to Mellanox syncd supervisord.conf
added "asan: yes" to sonic_version.yml
How to verify it
Added artificial memory leaks
Compiled with ENABLE_ASAN=y
Installed the image on DUT
Rebooted the DUT
Verified that /var/log/asan/syncd-asan.log contains the leaks
Signed-off-by: Yakiv Huryk <yhuryk@nvidia.com>
If it is run during image install, it's not guaranteed that the
installation environment will have tune2fs available. Therefore, run it
during initramfs instead.
Signed-off-by: Saikrishna Arcot <sarcot@microsoft.com>
Why I did it
In the bringup of tomahawk4/trident4, we realized that such chips need a larger size of /dev/shm in syncd container, so we added the option --shm-size to the docker create for syncd. The default value for shm-size is 64m; after this change, people can add SYNCD_SHM_SIZE=128m to platform_env.conf to change it to 128m.
How to verify it
We verified that after this change, 1) on existing platforms without platform_env.conf, the size of /dev/shm in syncd container (df -h | grep shm) is still the default 64M; 2) after we add SYNCD_SHM_SIZE=128m to platform_env.conf, /dev/shm in syncd becomes 128M.
- Why I did it
Fixes#9628
During bootup, this error log is seen
Dec 22 04:26:29 sonic interfaces-config.sh[2546]: error: main exception: cannot find interfaces: eth0 (interface was probably never up ?)
This is of non-functional nature and doesn't affect the flow.
- How I did it
Dont take the ifdown if not needed
- How to verify it
Verified during reboot. Log did not appear and IP was acquired on eth0 as expected
Signed-off-by: Vivek Reddy Karri <vkarri@nvidia.com>
Why I did it
To reduce the processing time of rc.local, refactoring s6100 platform initialization.
Porting changes from 202012 branch [202012] Refactoring DELL platform init to reduce rc.local processing time #10171
Why I did it
[Build]: Fix armhf mirrors not existing issue
The mirror endpoint debian-archive.trafficmanager.net does not support armhf, change to use deb.debian.org and security.debian.org.
Why I did it
improvement of starting barefoot SDK
How I did it
restart of the interface for cleaning txquee through which communication takes place between Sonic and openBMC
How to verify it
run sonic autorestart tests
On some products the pci enumeration adds randomness into which nic gets
initialized first.
Because SONiC doesn't use deterministic interface naming but instead old
style interface naming, this leads to eth0 not always being the
management port.
To make sure eth0 is always the management port (SONiC expectation)
rename the interfaces in the initramfs for Arista products.
# Why I did it
Reduce the disk space taken up during bootup and runtime.
# How I did it
1. Remove python package cache from the base image and from the containers.
2. During bootup, if logs are to be stored in memory, then don't create the `var-log.ext4` file just to delete it later during bootup.
3. For the partition containing `/host`, don't reserve any blocks for just the root user. This just makes sure all disk space is available for all users, if needed during upgrades (for example).
* Remove pip2 and pip3 caches from some containers
Only containers which appeared to have a significant pip cache size are
included here.
Signed-off-by: Saikrishna Arcot <sarcot@microsoft.com>
* Don't create var-log.ext4 if we're storing logs in memory
Signed-off-by: Saikrishna Arcot <sarcot@microsoft.com>
* Run tune2fs on the device containing /host to not reserve any blocks for just the root user
Signed-off-by: Saikrishna Arcot <sarcot@microsoft.com>
This can save 6 sec for teamd LAG restoration - the time between:
```
Mar 9 13:51:10.467757 r-panther-13 WARNING teamd#teamd_PortChannel1[28]: Got SIGUSR1.
Mar 9 13:52:33.310707 r-panther-13 INFO teamd#teamd_PortChannel1[27]: carrier changed to UP
```
- Why I did it
Optimize warm boot. Specifically reduce the time needed for LAG restoration.
- How I did it
Kill teamd docker after graceful shutdown of teamd processes.
- How to verify it
Run warm reboot.
Signed-off-by: Stepan Blyschak <stepanb@nvidia.com>
Why I did it
The marvel-armhf build is hung, it does not exit after waiting for a long time.
It is caused by the process /etc/entropy.py which is started by the postinst script in target/debs/buster/sonic-platform-nokia-7215_1.0_armhf.deb
When mounting the partition that contains `/host` during initramfs, the
mount binary available there (coming from busybox) tries each filesystem
in `/proc/filesystems` and sees which one succeeds. During this time,
there may be some error messages logged into dmesg because some of the
incorrect filesystems failed to mount the partition.
Specify the filesystem type explicitly so that initramfs knows it's that
type, and we know what filesystem will always get used there.
Fixes#9998
Signed-off-by: Saikrishna Arcot <sarcot@microsoft.com>
- Why I did it
To implement blocking feature state change.
- How I did it
Record the actual feature state in STATE DB from hostcfg.
- How to verify it
UT + verification by running on the switch and checking STATE DB.
Signed-off-by: Stepan Blyschak <stepanb@nvidia.com>
Why I did it
In parallel of this change Arista added a custom logrotate configuration as part of its driver library.
Having 2 logrotate configuration for the same log file triggers an issue.
Fixesaristanetworks/sonic#38
How I did it
Arista merged a few changes in sonic-buildimage which added a logrotate configuration aristanetworks/sonic@e43c797
It is therefore the right path to remove the arista.log line from the logrotate.d/rsyslog configuration.
How to verify it
Logrotate works without any error message, arista log rotation happens and arista daemons still append logs once file was truncated.
