kdump initial config not translated to the USE_KDUMP kernel flag

[arista-nwolfe]

the config kdump [enable|disable] command will set the KDUMP|config in CONFIG_DB.

> sonic-db-cli CONFIG_DB hgetall "KDUMP|config"
{'enabled': 'true', 'memory': '0M-2G:256M,2G-4G:320M,4G-8G:384M,8G-:448M', 'num_dumps': '3'}

Changes to this config are handled by KdumpCfg in hostcfgd:
https://github.com/sonic-net/sonic-host-services/blob/202405/scripts/hostcfgd#L1943

Changing the enabled flag in KDUMP|config will result in the sonic-kdump-config [--enable|--disable] script to be called:
https://github.com/sonic-net/sonic-host-services/blob/202405/scripts/hostcfgd#L1152-L1155

That script will in turn set the USE_KDUMP=[0|1] in /etc/default/kdump-tools depending on whether or not it's enabled:
https://github.com/sonic-net/sonic-utilities/blob/202405/scripts/sonic-kdump-config#L320

Changing this config requires a system reboot.

This behavior works as expected in a running system.
However, on boot-up the entry in KDUMP|config won't be propagated to the USE_KDUMP kernel flag.

Problematic scenario:

1. config_db.json has 'enabled': 'false'
2. /etc/default/kdump-tools has USE_KDUMP=1
3. Reboot

Steps to Reproduce:

1. config_db.json has 'enabled': 'false'

root@nfc406-7:~# config kdump disable
KDUMP configuration changes may require a reboot to take effect.
Save SONiC configuration using 'config save' before issuing the reboot command.

root@nfc406-7:~# sonic-db-cli CONFIG_DB hgetall "KDUMP|config"
{'enabled': 'false', 'memory': '0M-2G:256M,2G-4G:320M,4G-8G:384M,8G-:448M', 'num_dumps': '3'}

root@nfc406-7:~# config save -y
Running command: /usr/local/bin/sonic-cfggen -d --print-data > /etc/sonic/config_db.json

root@nfc406-7:~# grep "USE_KDUMP" /etc/default/kdump-tools
# USE_KDUMP - controls kdump will be configured
USE_KDUMP=0

root@nfc406-7:~# show kdump config
Kdump administrative mode: Disabled
Kdump operational mode: Not Ready
Kdump memory reservation: 0M-2G:256M,2G-4G:320M,4G-8G:384M,8G-:448M
Maximum number of Kdump files: 3

2. Set USE_KDUMP=1 (set but don't save)

root@nfc406-7:~# config kdump enable
KDUMP configuration changes may require a reboot to take effect.
Save SONiC configuration using 'config save' before issuing the reboot command.

root@nfc406-7:~# grep "USE_KDUMP" /etc/default/kdump-tools
# USE_KDUMP - controls kdump will be configured
USE_KDUMP=1

root@nfc406-7:~# show kdump config
Kdump administrative mode: Enabled
Kdump operational mode: Ready after reboot
Kdump memory reservation: 0M-2G:256M,2G-4G:320M,4G-8G:384M,8G-:448M
Maximum number of Kdump files: 3

3. Reboot

root@nfc406-7:~# reboot

Result: KDUMP|config says it's disabled but the USE_KDUMP=1 kernel flag says it's enabled

root@nfc406-7:~# sonic-db-cli CONFIG_DB hgetall "KDUMP|config"
{'enabled': 'false', 'memory': '0M-2G:256M,2G-4G:320M,4G-8G:384M,8G-:448M', 'num_dumps': '3'}

root@nfc406-7:~# grep "USE_KDUMP" /etc/default/kdump-tools
# USE_KDUMP - controls kdump will be configured
USE_KDUMP=1

root@nfc406-7:~# show kdump config
Kdump administrative mode: Disabled
Kdump operational mode: Ready
Kdump memory reservation: 0M-2G:256M,2G-4G:320M,4G-8G:384M,8G-:448M
Maximum number of Kdump files: 3

Fallout: The startup config (config_db.json) requested the KDUMP feature be disabled but it's still enabled and rebooting the switch won't help, you'll need to re-run the config kdump disable on a running switch to re-sync the config_db and kernel flags

Note: The same thing can occur with the opposite config (CONFIG_DB: enabled=True, USE_KDUMP=0)

Analysis:
Looking at the load functions of the classes in hostcfgd it looks like it's typically used to call the handler function for the initial CONFIG_DB processing.
I'm not sure why in the case of KdumpCfg we only setup the KDUMP|config in CONFIG_DB if it's not already present:

    def load(self, kdump_table):
        """
        Set the KDUMP table in CFG DB to kdump_defaults if not set by the user
        """
        syslog.syslog(syslog.LOG_INFO, "KdumpCfg init ...")
        kdump_conf = kdump_table.get("config", {})
        for row in self.kdump_defaults:
            value = self.kdump_defaults.get(row)
            if not kdump_conf.get(row):
                self.config_db.mod_entry("KDUMP", "config", {row : value})

https://github.com/sonic-net/sonic-host-services/blob/202405/scripts/hostcfgd#L1130-L1139

[arista-nwolfe]

Actually it looks like in the original PR which added KdumpCfg the load function did call the kdump_update function:
sonic-net/sonic-host-services@1c458da#diff-f46602bac3380a0d28ddfcbe698d95d87c7bc57e504fac08dd6906bdb31ec388R243-R254

But a large commit came later and removed it:
sonic-net/sonic-host-services@58a6e49#diff-f46602bac3380a0d28ddfcbe698d95d87c7bc57e504fac08dd6906bdb31ec388L735-R785

[bingwang-ms]

@judyjoseph Can you help take a look?

[judyjoseph]

@arista-nwolfe could you create a PR to fix this, since you have already done the background work? Talked to @dgsudharsan from nvidia team found that the PR in which this is removed is not related to kdump.

So I feel we can go ahead and revert that part of code to call self.kdump_update back