PMON Voltage/Current Sensor Monitoring Enhancement

[bmridul]

HLD for Voltage and Current sensor monitoring.

Related PRs:

PR	Description	Status
sonic-net/sonic-utilities#2941	Sensormon CLIs	Merged
sonic-net/sonic-platform-common#393	Sensors Base Classes	Merged
sonic-net/sonic-platform-daemons#393	Sensormon Daemon	Merged
sonic-net/sonic-buildimage#16089	Add Sensormon in PMON	Merged
sonic-net/sonic-buildimage#16646	System Health support	Open
sonic-net/sonic-snmpagent#295	SNMP support	Open
sonic-net/sonic-mgmt#24036	SONiC Mgmt test cases	Merged
sonic-net/sonic-platform-common#426	File system bases sensors	Merged

[prgeor]

@bmridul why not use the existing lm-sensors in linux which provides much more detailed output ? We have sensord running in PMON which can monitor voltage/current. We also syslog alerts for high temperature and current.

root@3POs-8101-H11-18:~# docker exec pmon service sensord status
sensord is running.
root@3POs-8101-H11-18:~#

[bmridul]

@bmridul why not use the existing lm-sensors in linux which provides much more detailed output ? We have sensord running in PMON which can monitor voltage/current. We also syslog alerts for high temperature and current.

root@3POs-8101-H11-18:~# docker exec pmon service sensord status
sensord is running.
root@3POs-8101-H11-18:~#

Hi @prgeor,

Please note that the solution proposed is not mutually exclusive to Sensord/Lmsensors. The solution proposed here can be used alongside or without Sensord.

There are 2 main reasons we are proposing this.

1. Sensord provides monitoring of sensors which are supported by the linux hwmon infrastructure. The list of devices supported are mentioned here : https://hwmon.wiki.kernel.org/device_support_status

However, the issue is that there are a number of devices which report voltage/current measurements that are not supported by hwmon.

An example is device LTC2497 which can report voltage measurements.

The open source driver available for this device supports a different linux infrastructure – Industrial I/O
https://www.kernel.org/doc/html/v4.12/driver-api/iio/intro.html

The driver code is here : https://github.com/torvalds/linux/blob/master/drivers/iio/adc/ltc2497.c

This device will not be monitored by hwmon unless we patch its standard kernel driver which is not something we would want to do.

Sensormon can cover this gap and provide one CLI/rpcoess to view/monitor all the voltage and current sensors.

2. Even when hwmon does support the devices, we observed that the fault reporting only works when the device indicates a fault. A user supplied threshold would not generate a minor or major alarm. Sensormon can provide this support and alert if there is a failure even when the device may not have the capability to generate an alarm.

[prgeor]

@bmridul why not use the existing lm-sensors in linux which provides much more detailed output ? We have sensord running in PMON which can monitor voltage/current. We also syslog alerts for high temperature and current.

root@3POs-8101-H11-18:~# docker exec pmon service sensord status
sensord is running.
root@3POs-8101-H11-18:~#

Hi @prgeor,

Please note that the solution proposed is not mutually exclusive to Sensord/Lmsensors. The solution proposed here can be used alongside or without Sensord.

There are 2 main reasons we are proposing this.

1. Sensord provides monitoring of sensors which are supported by the linux hwmon infrastructure. The list of devices supported are mentioned here : https://hwmon.wiki.kernel.org/device_support_status

However, the issue is that there are a number of devices which report voltage/current measurements that are not supported by hwmon.

An example is device LTC2497 which can report voltage measurements.

The open source driver available for this device supports a different linux infrastructure – Industrial I/O https://www.kernel.org/doc/html/v4.12/driver-api/iio/intro.html

The driver code is here : https://github.com/torvalds/linux/blob/master/drivers/iio/adc/ltc2497.c

This device will not be monitored by hwmon unless we patch its standard kernel driver which is not something we would want to do.

Sensormon can cover this gap and provide one CLI/rpcoess to view/monitor all the voltage and current sensors.

2. Even when hwmon does support the devices, we observed that the fault reporting only works when the device indicates a fault. A user supplied threshold would not generate a minor or major alarm. Sensormon can provide this support and alert if there is a failure even when the device may not have the capability to generate an alarm.

@bmridul I am still not quite convinced the use of sensormon, since devices using IIO is mostly used in consumer products like laptop, mobile phones. The operating range of the chip is best known to the chip manufacturer and hence its better to depend upon the chip based alarm rather than platform integrator arriving at the customized threshold to raise alarms. As you rightly pointed out, hwmon already includes almost all sensors barring few exceptions, so I would prefer if you make effort to include these few exception devices included in hwmon support.

Nevertheless, please capture the reasons in the HLD for sensormon.

[bmridul]

@bmridul why not use the existing lm-sensors in linux which provides much more detailed output ? We have sensord running in PMON which can monitor voltage/current. We also syslog alerts for high temperature and current.

root@3POs-8101-H11-18:~# docker exec pmon service sensord status
sensord is running.
root@3POs-8101-H11-18:~#

Hi @prgeor,
Please note that the solution proposed is not mutually exclusive to Sensord/Lmsensors. The solution proposed here can be used alongside or without Sensord.
There are 2 main reasons we are proposing this.

1. Sensord provides monitoring of sensors which are supported by the linux hwmon infrastructure. The list of devices supported are mentioned here : https://hwmon.wiki.kernel.org/device_support_status

However, the issue is that there are a number of devices which report voltage/current measurements that are not supported by hwmon.
An example is device LTC2497 which can report voltage measurements.
The open source driver available for this device supports a different linux infrastructure – Industrial I/O https://www.kernel.org/doc/html/v4.12/driver-api/iio/intro.html
The driver code is here : https://github.com/torvalds/linux/blob/master/drivers/iio/adc/ltc2497.c
This device will not be monitored by hwmon unless we patch its standard kernel driver which is not something we would want to do.
Sensormon can cover this gap and provide one CLI/rpcoess to view/monitor all the voltage and current sensors.

2. Even when hwmon does support the devices, we observed that the fault reporting only works when the device indicates a fault. A user supplied threshold would not generate a minor or major alarm. Sensormon can provide this support and alert if there is a failure even when the device may not have the capability to generate an alarm.

@bmridul I am still not quite convinced the use of sensormon, since devices using IIO is mostly used in consumer products like laptop, mobile phones. The operating range of the chip is best known to the chip manufacturer and hence its better to depend upon the chip based alarm rather than platform integrator arriving at the customized threshold to raise alarms. As you rightly pointed out, hwmon already includes almost all sensors barring few exceptions, so I would prefer if you make effort to include these few exception devices included in hwmon support.

Nevertheless, please capture the reasons in the HLD for sensormon.

@prgeor , Thanks for your comments. I will update the HLD with more details. Few responses

1. The component vendors might push a driver based on initial application of the device. However the components can be used in a variety of applications once generally available.
2. Many devices do not provide an alarm function. If the device has hardware logic for monitoring the voltage/current, it might provide an alarm, but many devices do not.
3. Even though the device might be in its max/min limits, it might have an adverse affect on the components it is delivering the voltage to. So the sensor thresholds should be adjusted from a board design perspective.

The uber point is that the sensord may not provide complete coverage. If Linux Sensord is good enough for a platform, it can choose not to use this infra. We can discuss further in community meeting.

[jeffyin-dell]

As discussed during the community call today -- please explore integration with SONiC PDDF (https://github.com/sonic-net/SONiC/blob/master/doc/platform/brcm_pdk_pddf.md) so that platform vendors can easily integrate support for voltage/current sensors using their PDDF JSON files.

[Junchao-Mellanox]

A few general comments:

1. How is the daemon connect to system health service?
2. How is the daemon connect to snmp?
3. How to build the sensor hierarchy relation? For example, should PSU voltage sensor object under PSU object or under chassis object?
4. There is already logic in PSUD to handle PSU voltage threshold, do you plan to remove it?
5. How to handle hot swappable sensors? For example, PSU sensor, module sensor?

[bmridul]

As discussed during the community call today -- please explore integration with SONiC PDDF (https://github.com/sonic-net/SONiC/blob/master/doc/platform/brcm_pdk_pddf.md) so that platform vendors can easily integrate support for voltage/current sensors using their PDDF JSON files.

I looked briefly into the PDDF documentation and some code. It seems the PDDF support for voltage and current sensors can be added in a similar way as as other platform components are supported. E,g, The code below is implementing the platform support for thermal sensors.

https://github.com/sonic-net/sonic-buildimage/blob/204579a0ccfe80c1f2afea2bda0574528faecb91/platform/pddf/platform-api-pddf-base/sonic_platform_pddf_base/pddf_thermal.py

Voltage and current sensors support can be similarly added.

[Junchao-Mellanox]

A few general comments:

1. How is the daemon connect to system health service?
2. How is the daemon connect to snmp?
3. How to build the sensor hierarchy relation? For example, should PSU voltage sensor object under PSU object or under chassis object?
4. There is already logic in PSUD to handle PSU voltage threshold, do you plan to remove it?
5. How to handle hot swappable sensors? For example, PSU sensor, module sensor?

Hi @bmridul @abdosi , how about these comments?

[bmridul]

A few general comments:

1. How is the daemon connect to system health service?

It would be good to show the out of out of range sensor readings in system health o/p. Currently out of range temperature sensors are also not shown in system health. We should add those as well.

I have added PR for system health integration.

[bmridul]

A few general comments:
2. How is the daemon connect to snmp?

SNMP support will need to be added to represent the voltage/current sensors in Entity MIB. We will need to address this with changes in rfc2737 implmentation.

[zhangyanzhao]

HLD is merged but code PRs are still open, move to backlog