Skip to content

Latest commit

 

History

History
789 lines (523 loc) · 40.3 KB

G022 - K3s cluster setup 05 ~ Connecting the VM to the NUT server.md

File metadata and controls

789 lines (523 loc) · 40.3 KB

G022 - K3s cluster setup 05 ~ Connecting the VM to the NUT server

In the G004 guide you set up a standalone NUT server so your PVE node could monitor the UPS unit it is plugged in. Now that you started creating VMs, it'll be very convenient to make them aware of your UPS unit so, when there's a power cut and the UPS kicks in, you can automate a proper shutdown sequence for your whole server setup. Think that your VMs, in this context, will be like real computers plugged in to the same UPS unit as your Proxmox VE host, so they'll be also directly affected to whatever happens to your PVE host's power supply.

Reconfiguring the NUT master server on your Proxmox VE host

First, you need to change the configuration of the NUT master server running in your Proxmox VE host, so it can also serve the NUT slave clients in any of your VMs.

Changing the nut.conf file

  1. Open a shell as mgrsys on your Proxmox VE host, then cd to /etc/nut and make a backup of the nut.conf file.

    $ cd /etc/nut ; sudo cp nut.conf nut.conf.bkp
  2. Edit the nut.conf file and change the value of the MODE parameter to netserver.

    MODE=netserver
  3. Save the changes on the nut.conf file.

Reconfiguring the UPS driver with the ups.conf file

  1. Remain at /etc/nut, and make a backup of the ups.conf file.

    $ sudo cp ups.conf ups.conf.bkp
  2. Edit the ups.conf file. At its end, you should have just the configuration block for your UPS unit you configured previously in the G004 guide.

    [apc]
        driver = usbhid-ups
        port = auto
        desc = "APC Back-UPS ES 700"

    Let's add to this block a few more parameters, so it looks as the following.

    [apc]
        driver = usbhid-ups
        port = auto
        desc = "APC Back-UPS ES 700"
        offdelay = 60
        ondelay = 70
        lowbatt = 80

    Here's a brief explanation of the three new parameters added to the [upsunit] configuration.

    • offdelay : Value in seconds, default is 20 seconds. Time that passes between the upsdrvctl shutdown command and the moment the UPS shuts itself down.

    • ondelay : Value in seconds, default is 30 seconds. Time that must pass between the upsdrvctl shutdown command and the moment when the UPS will react to the return of wall power and turn on the power to the system.

      BEWARE!
      This ondelay value must be greater than the offdelay number.

    • lowbatt : Value in percentage. Percentage of battery charge remaining in the UPS unit that should be considered as "low charge".

    To set up those parameters properly, you must be fully aware of the real capacity of your UPS unit to hold your system running in case of a power cut. In my case, my UPS unit can only hold for just very few minutes before discharging completely, so I'm forced to set very conservative values (in particular in the lowbatt parameter). Of course, how long your UPS unit can run, depends a lot on how much (or how fast) energy will be drained from its battery when in use. Also, bear in mind the battery degradation over time (even if it's never used), and how long does your server take to shutdown gracefully (including your VMs and services running within them).

  3. Save the changes on the ups.conf file.

Adding access control rules in the upsd.conf file

  1. Still at /etc/nut, make a backup of the upsd.conf file.

    $ sudo cp upsd.conf upsd.conf.bkp
  2. Edit the upsd.conf file, jump to the LISTEN lines and modify them as follows.

    # LISTEN <address> [<port>]
    #LISTEN 127.0.0.1 3493
    LISTEN 0.0.0.0 3493
    # LISTEN ::1 3493

    By setting 0.0.0.0 as listening address, you're making the upsd service listen for traffic from all IP v4 sources available in your Proxmox VE host.

  3. Save the changes on the upsd.conf file.

Adding a NUT slave client user in the upsd.users file

  1. Still at /etc/nut, make a backup of the upsd.users file.

    $ sudo cp upsd.users upsd.users.bkp
  2. Edit the upsd.users file by appending the following block.

    [upsmonclient]
        password = s3c4R3_Kly3Nt_p4sSw0rD!
        upsmon slave

    Notice the upsmon slave line, which indicates that the upsmonclient will be the user of slave client systems.

  3. Save the changes on the upsd.users file.

After adding this user, your NUT system will have three different NUT users: one administrator user, one monitor user to be used only locally in the master NUT server, and other monitor user only to be used remotely from slave/client NUT systems such as the VMs you'll create later on your PVE node.

Declaring executable actions related to concrete NUT events

  1. Yet remaining at /etc/nut, now make a backup of the current upsmon.conf file.

    $ sudo cp upsmon.conf upsmon.conf.bkp
  2. Edit the upsmon.conf file, adding below the customized SHUTDOWNCMD parameter a new NOTIFYCMD line.

    SHUTDOWNCMD "logger -t upsmon.conf \"SHUTDOWNCMD calling /sbin/shutdown to shut down system\" ; /sbin/shutdown -h +0"
    NOTIFYCMD /usr/sbin/upssched

    In NOTIFYCMD, the configured upssched command is a NUT program that provides a rich set of actions to execute in response to events detected by upsmon.

  3. Also, modify the customized NOTIFYFLAG lines for the ONLINE, ONBATT and LOWBATT events by adding a +EXEC flag.

    NOTIFYFLAG ONLINE SYSLOG+EXEC
    NOTIFYFLAG ONBATT SYSLOG+EXEC
    NOTIFYFLAG LOWBATT SYSLOG+EXEC

    The +EXEC flag makes upsmon call the program configured in the NOTIFYCMD parameter.

  4. Save the changes on the upsmon.conf file.

  5. Rename the upssched.conf file to upssched.conf.orig, and create a new empty upssched.conf file.

    $ sudo mv upssched.conf upssched.conf.orig
    $ sudo touch upssched.conf ; sudo chmod 640 upssched.conf
  6. Edit the upssched.conf file and put in it the following content.

    # upssched.conf
    CMDSCRIPT /usr/sbin/upssched-cmd
    PIPEFN /var/run/nut/upssched.pipe
    LOCKFN /var/run/nut/upssched.lock
    
    AT ONLINE apc@localhost EXECUTE online
    AT ONBATT apc@localhost EXECUTE onbatt
    AT LOWBATT apc@localhost EXECUTE lowbatt

    The parameters above mean the following.

    • CMDSCRIPT : points to a user script that will be executed by upssched in response to the UPS events notified by the upsmon NUT monitor service.

      BEWARE!
      This script is provided by the system administrator, you in this case. Don't confuse it with the /usr/sbin/upssched command, which is the NUT program that will call your script. On the other hand, the name upssched-cmd is the assumed standard in the NUT community, so don't change it.

    • PIPEFN : socket file used for communication between upsmon and upssched.

      BEWARE!
      The directory containing this file should be accessible only by the NUT software and nothing else.

    • LOCKFN : file required by the upsmon NUT daemon to avoid race conditions.

      BEWARE!
      The directory should be the same as for the PIPEFN file.

    • AT lines: declarations for actions to EXECUTE with the CMDSCRIPT only AT the events defined in this configuration file. These declarations follow the pattern below:

      AT notifytype UPSunit-name command
      • AT is the keyword to start the declaration.

      • notifytype is the code identifying a NUT-monitored event.

      • UPSunit-name is the name of the monitored UPS unit. This argument admits the use of the wildcard * to refer to all available UPS units, although it's not recommended; you'll usually prefer to have particular rules for each of those UPS units.

      • command specifies what to do when the event happens. The EXECUTE command is just one of several available.

  7. Create an empty /usr/sbin/upssched-cmd file with the right permissions.

    $ sudo touch /usr/sbin/upssched-cmd ; sudo chmod 755 /usr/sbin/upssched-cmd
  8. Edit the new upssched-cmd so it has the following shell script code.

    #!/bin/bash -u
    # upssched-cmd
    logger -i -t upssched-cmd Calling upssched-cmd $1
    
    UPS="apc"
    STATUS=$( upsc $UPS ups.status )
    CHARGE=$( upsc $UPS battery.charge )
    CHMSG="[$STATUS]:$CHARGE%"
    
    case $1 in
        online) MSG="$UPS, $CHMSG - INFO - Power supply is back online." ;;
        onbatt) MSG="$UPS, $CHMSG - WARNING - Power failure. Now running on battery!" ;;
        lowbatt) MSG="$UPS, $CHMSG - WARNING - Low battery. Shutdown now!" ;;
        *) logger -i -t upssched-cmd "ERROR - Bad argument: \"$1\", $CHMSG" 
            exit 1 ;;
    esac
    
    logger -i -t upssched-cmd $MSG
    notify-send-all "$MSG"

    BEWARE!
    This script is just an example of what you can do which, in this case, is just some logging and sending messages regarding a particular set of UPS events. Notice how it has one input argument ($1), and the options appearing in the case block are used in the AT declarations defined in the upssched.conf file.

  9. Save the upssched-cmd file and restart the NUT related services already running in your Proxmox VE host.

    $ sudo systemctl restart nut-server.service nut-monitor.service

Opening the upsd port on the Proxmox VE node

You need to make the upsd port (the NUT standard 3493 in this guide) accessible in your PVE host, so the NUT client monitors in your VMs can connect to it. To achieve this, you'll need to set up proper firewall rules in Proxmox VE, both at the Datacenter and PVE node levels of the PVE web console.

  1. First, get the IP of your, at this point, sole Debian VM. You can see it in the VM's Summary screen, in the Status block.

    VM's IP at Summary screen

    This VM only has one external IP assigned, by your router or gateway, to the only network card that it's currently enabled in the VM.

  2. Lets give an Alias to the VM's IP. Go to Datacenter > Firewall > Alias and click on Add.

    Datacenter Firewall Alias screen

  3. You'll see the following form.

    New IP Alias form

    Let's fill it in a meaningful manner:

    • Name: careful with this field, not any string will be considered valid here.

      • Use only alphanumerical characters plus the '_' symbol.
      • Use only letters from the english alphabet.

      BEWARE!
      This behavior is not explained in the Proxmox VE official documentation.

    • IP/CIDR: this value can be just a single IP (like 1.2.3.4) or a network range (as 1.2.3.0/24). For this case, you'll just input the VM's main IP here (the one assigned to net0).

    • Comment: you can use this field to put more specific info about the alias.

    New IP Alias form filled

  4. After clicking on Add, you'll see how the new alias appears listed.

    New IP Alias listed

  5. Next, create an IP set to hold all the IPs related to the Kubernetes cluster you'll build in a later guide. Go to the Firewall at the Datacenter level, open the IPSet screen and click on Create.

    Datacenter Firewall IPSet screen

  6. A simple form will raise.

    IPSet creation form

    Let's give the new IP set a meaningful Name such as k3s_nodes_net0_ips.

    IPSet creation form filled

    After filling the form, just click on OK to create the new IP set.

  7. The IPSet screen will now show the new IP set on its left side. Select the new k3s_nodes_net0_ips IP set, then click on the now enabled Add button over the right side.

    IP set created

  8. You'll meet a form where to input the IP to add to the IP set.

    Editor for adding new IP to IP set

    This form has three fields.

    • IP/CIDR: here you can type an IP, or just choose an aliased IP from the unfolded list.

    • nomatch: enabling this option explicitly excludes from the set the IP put in the IP/CIDR field.

    • Comment: any string you might want to type here, although try to make it meaningful.

    Choose your previously aliased IP so the form looks like below.

    New aliased IP filled in

  9. Click on Create and you'll see your aliased IP added to the set on the IP/CIDR list on the right.

    Aliased IP added to IP set

  10. Next step is to create a Security Group. A security group is just a way of grouping together firewall rules, making it easier to handle them later. So, go to Datacenter > Firewall > Security Group in your PVE web console. Then, click on Create.

    Datacenter Firewall Security Group screen

  11. A simple form will raise.

    Security Group editor

    This group will hold the firewall rules related to the NUT port, so let's give it a Name like nut_port_accept_in.

    BEWARE!
    Careful with the Name field here, as it happened with the IP alias name field before, you can only use alphanumerical english characters plus the '_' symbol.

    Security Group form filled

  12. Click on the Create button, and see how your new security group appears on the left half of the Security Group page. Then, select the new group and click on Add on the right half.

    New security group added to list

  13. In the editor that appears, you're going to define a firewall rule to allow access to the NUT 3493 port from the IP set you defined a few steps before.

    Security Group firewall rule editor

    The explanation for each field is the following.

    • Direction: this only has two options, either the rule is about INcoming or OUTgoing connections. Choose in to allow incoming connections to the NUT server.

    • Action: three possibilities here, ACCEPT the connection, DROP or REJECT it. In this case, choose ACCEPT.

      BEWARE!
      The difference between DROP and REJECT is that DROP rejects connections silently, whereas REJECT makes the firewall answer back a rejection code. For stopping connection attempts, always use DROP unless you really need to make the rejection noticeable.

    • Enable: as its name implies, ticking this option enables the rule. Leave it off for now.

    • Macro: this is a list of services known by Proxmox VE. If you choose one, some fields of this rule editor will be filled automatically. Leave it empty, since it doesn't have an option for NUT.

    • Protocol: this a list of known net protocols, from which you'll have to choose the right one. For NUT, choose tcp.

    • Source: indicates the IP where the connection comes from. Be mindful that, depending on the value of the Direction field, Source can refer to an external origin (an INcoming connection) or to your own system (an OUTgoing connection). This field also offers a list of IP sets and IP aliases, so you just have to click on which one you want rather than typing IPs on the field.

      List of existing defined IPs in the PVE system

      For the NUT rule, choose the IP set you defined before, the k3s_nodes_net0_ips. This way, any other IP you put into that IP set will be already covered by this firewall rule.

    • Source port: which port on the source can the connection come from. Leave it empty for accepting any port.

    • Destination: is the IP the incoming or outgoing connection wants to reach. This field is of the same kind as Source. For the NUT rule, leave it empty for now (unless you've already defined an alias or IP set for your PVE node's IPs).

    • Dest. port: port in which the connection wants to connect. Specify the NUT port here: 3493 (or the whichever one you have put in your NUT configuration).

    • Comment: any meaningful string you might like to put here.

    • Log level: if you want this rule to log the connections that go through it, change the default nolog value to any other that suits the severity of the event. For NUT, it shouldn't be more than info, usually.

    New rule defined for Security Group

  14. Click on Add, and see the updated Rules list of your Security Group on the right side of the page.

    Security Group Rules list updated

    Notice that the rule is NOT active, since the checkbox on the On column is NOT ticked. Leave it that way for now.

  15. Go to your PVE node, and browse to the now ruleless Firewall section and click on Insert: Security Group.

    Ruleless PVE node Firewall

  16. A form will raise where you'll be able to define a firewall rule that applies a security group.

    Security Group rule editor

    Here's a brief explanation for each field.

    • Security Group: here you must choose a preexisting security group defined at the Datacenter level. In this case, you only have the one you've created before: nut_port_accept_in.

    • Interface: this field only admits the name (not IPs or MACs) of an existing node network interface, or just being left empty. In this case you'll need to put the name of the vmbr0 bridge, since it's the network device through which your Proxmox VE host communicates.

    • Enable: by ticking this on, you enable this rule on the firewall.

    • Comment: any string you may want to put here, like a brief explanation of what the rule does.

    Security Group rule defined

  17. Click on Add and the new rule will appear immediately in the Firewall of your node.

    Added new rule to node Firewall

Configuring the NUT slave client on your Debian VM

The previous section has been all about configuring the server side of your NUT setup. Now you'll deal with the client side of it in your Debian VM. Remember that, in your VM, you only installed the nut-client package of NUT, since you won't need the server components for a pure client configuration.

  1. Open a shell in your Debian VM as mgrsys, then cd to /etc/nut and make a backup of the nut.conf file.

    $ cd /etc/nut/ ; sudo cp nut.conf nut.conf.orig
  2. Edit the nut.conf file and just change the value of the MODE parameter to netclient.

    MODE=netclient
  3. Save the changes to nut.conf, then make a backup of the upsmon.conf file.

    $ sudo cp upsmon.conf upsmon.conf.orig
  4. Edit upsmon.conf as follows.

    • Search for and comment out the active lines of the RBWARNTIME and SHUTDOWNCMD parameters.

      # --------------------------------------------------------------------------
      # RBWARNTIME - replace battery warning time in seconds
      #
      # upsmon will normally warn you about a battery that needs to be replaced
      # every 43200 seconds, which is 12 hours.  It does this by triggering a
      # NOTIFY_REPLBATT which is then handled by the usual notify structure
      # you've defined above.
      #
      # If this number is not to your liking, override it here.
      
      #RBWARNTIME 43200
      # --------------------------------------------------------------------------
      # SHUTDOWNCMD "<command>"
      #
      # upsmon runs this command when the system needs to be brought down.
      #
      # This should work just about everywhere ... if it doesn't, well, change it.
      
      #SHUTDOWNCMD "/sbin/shutdown -h +0"
    • Append the following lines to the upsmon.conf file.

      # --------------------------------------------------------------------------
      # Customized settings
      
      MONITOR [email protected] 1 upsmonclient s3c4R3_Kly3Nt_p4sSw0rD! slave
      SHUTDOWNCMD "logger -t upsmon.conf \"SHUTDOWNCMD calling /sbin/shutdown to shut down system\" ; /sbin/shutdown -h +0"
      NOTIFYCMD /usr/sbin/upssched
      
      NOTIFYMSG ONLINE "UPS %s: On line power."
      NOTIFYMSG ONBATT "UPS %s: On battery."
      NOTIFYMSG LOWBATT "UPS %s: Battery is low."
      NOTIFYMSG REPLBATT "UPS %s: Battery needs to be replaced."
      NOTIFYMSG FSD "UPS %s: Forced shutdown in progress."
      NOTIFYMSG SHUTDOWN "Auto logout and shutdown proceeding."
      NOTIFYMSG COMMOK "UPS %s: Communications (re-)established."
      NOTIFYMSG COMMBAD "UPS %s: Communications lost."
      NOTIFYMSG NOCOMM "UPS %s: Not available."
      NOTIFYMSG NOPARENT "upsmon parent dead, shutdown impossible."
      
      NOTIFYFLAG ONLINE SYSLOG+EXEC
      NOTIFYFLAG ONBATT SYSLOG+EXEC
      NOTIFYFLAG LOWBATT SYSLOG+EXEC
      NOTIFYFLAG REPLBATT SYSLOG
      NOTIFYFLAG FSD SYSLOG
      NOTIFYFLAG SHUTDOWN SYSLOG
      NOTIFYFLAG COMMOK SYSLOG
      NOTIFYFLAG COMMBAD SYSLOG
      NOTIFYFLAG NOCOMM SYSLOG
      NOTIFYFLAG NOPARENT SYSLOG
      
      RBWARNTIME 7200 # 2 hours

      The configuration above is mostly the same as the one you have in the upsmon.conf on your PVE node, save the MONITOR line which defines to what UPS unit to connect and with what user. Notice that in that MONITOR line you should set the user you've defined, previously in this guide, as an unprivileged NUT slave user in your PVE host's upsd.users file. This is much safer than using the fully-privileged administrator user just for monitoring your UPS unit from a VM. Also see how an IP is being used to refer to the UPS unit to monitor, this IP is the NUT server IP. Also, instead of an IP you could also use the NUT server's hostname if your network is able to resolve it to the right IP.

  5. Save the upsmon.conf file, then rename the upssched.conf file to upssched.conf.orig, and create a new empty upssched.conf file.

    $ sudo mv upssched.conf upssched.conf.orig
    $ sudo touch upssched.conf ; sudo chmod 640 upssched.conf ; sudo chgrp nut upssched.conf
  6. Edit the upssched.conf adding the following lines.

    # upssched.conf
    CMDSCRIPT /usr/sbin/upssched-cmd
    PIPEFN /var/run/nut/upssched.pipe
    LOCKFN /var/run/nut/upssched.lock
    
    AT ONLINE [email protected] EXECUTE online
    AT ONBATT [email protected] EXECUTE onbatt
    AT LOWBATT [email protected] EXECUTE lowbatt

    Like with the previous configuration file, here you can see that this file is almost the same upssched.conf file as the one set up previously in your PVE host. The only difference here is that, in the AT declarations, where there was a call to the localhost now you must put the NUT server IP or its hostname.

  7. Save the upssched.conf file. Then create an empty /usr/sbin/upssched-cmd file with the right permissions.

    $ sudo touch /usr/sbin/upssched-cmd ; sudo chmod 755 /usr/sbin/upssched-cmd
  8. Edit the new /usr/sbin/upssched-cmd so it has the following shell script code.

    #!/bin/bash -u
    # upssched-cmd
    logger -i -t upssched-cmd Calling upssched-cmd $1
    
    UPS="[email protected]"
    STATUS=$( upsc $UPS ups.status )
    CHARGE=$( upsc $UPS battery.charge )
    CHMSG="[$STATUS]:$CHARGE%"
    
    case $1 in
        online) MSG="$UPS, $CHMSG - INFO - Power supply is back online." ;;
        onbatt) MSG="$UPS, $CHMSG - WARNING - Power failure. Now running on battery!" ;;
        lowbatt) MSG="$UPS, $CHMSG - WARNING - Low battery. Shutdown now!" ;;
        *) logger -i -t upssched-cmd "ERROR - Bad argument: \"$1\", $CHMSG" 
            exit 1 ;;
    esac
    
    logger -i -t upssched-cmd $MSG
    notify-send-all "$MSG"

    This script is almost the same one you already configured in your PVE host. The only thing that changes is the UPS variable, which now also needs the IP of the NUT server specified.

  9. Save the upssched-cmd file and start the NUT related services already running in your VM.

    $ sudo systemctl restart nut-client.service nut-monitor.service 

Checking the connection between the VM NUT slave client and the PVE node NUT master server

You have everything configured, so it's time you test the NUT connection between your VM and the Proxmox VE node. To do this, execute the upsc command on your VM as follows.

$ sudo upsc [email protected]

And this command should return the following output.

Error: Connection failure: Connection timed out

Yes, there's a problem with the connection. You can also see this error in the /var/log/syslog file in your VM, in which you should find a number of lines like the following one.

Nov 19 19:44:27 debiantpl upsmon[593]: UPS [[email protected]]: connect failed: Connection failure: Connection timed out

The NUT monitor daemon on your VM tries to connect every few seconds to the NUT server on your PVE node, but something is wrong and the connection times out. So, what's missing? Just a small detail, regarding the security group you created in the firewall at the Datacenter level of your Proxmox VE platform. So, go back to the PVE web console, browse to Datacenter > Firewall > Security Group and select the group you created there before, the nut_port_accept_in one.

Rule nut_port_accept_in disabled at Security Group

You only declared one rule in the security group, and I told you to leave it disabled. This way, I've shown you a bit how the relationship between levels within Proxmox VE works regarding firewall rules. Remember that, at your pve node tier, you have created and enabled a firewall rule that used the security group itself; but only the rules enabled in the security group will be also enforced by the firewall. So, to enable access to the NUT port in your PVE host, you must also activate the rule you prepared right for it. To do so, just click on the checkbox the rule has at the On column.

Rule enabled in Security Group

The rule will be enforced by the firewall right away, so you can already check if now you can communicate with the NUT server from your VM.

$ sudo upsc [email protected]
Init SSL without certificate database
battery.charge: 100
battery.charge.low: 80
battery.charge.warning: 50
battery.date: not set
...
ups.vendorid: 051d

Now you should get a proper answer from the NUT server, in this case information about your UPS unit. On the other hand, and maybe a few seconds after you enabled the rule in the firewall, in the /var/log/syslog of your VM you should see the following line once.

Nov 19 19:47:49 debiantpl upsmon[593]: UPS [email protected]: Communications (re-)established.

The NUT monitor in your VM now is properly connected to the NUT server in your PVE node. On the other hand, in the /var/log/syslog file of your PVE host you should also find an output like below.

Nov 19 19:47:49 pve upsd[923]: User [email protected] logged into UPS [apc]

So, the NUT connection on the server side is also verified. Mind you that, on the server side, you won't see the connection failures, since those happened due to the firewall blocking the connection.

As a final note on this matter, browse in your PVE web console to the Datacenter Firewall.

No rules on Datacenter Firewall

As you can see, there are no visible rules whatsoever, although you created the security group at the Datacenter tier, and let's not forget the rules enabled in the iptables running underneath, but those are not show here either. Essentially, from the firewall point of view, the datacenter and node levels are the same logical network "host" zone, although each level have their own firewall configuration page (with different options available, mind you). Just be aware of this particularity of Proxmox VE whenever you have to deal with its firewall.

Testing a Forced ShutDown sequence (FSD) with NUT

To test if your whole NUT setup works as expected, you can provoke a Forced ShutDown (FSD) event with just one NUT command. But first you must understand what's going to happen after provoking the FSD event.

FSD event shutdown sequence

  1. The upsmon master daemon, running in your Proxmox VE host, notices and sets FSD (the forced shutdown flag) to tell all slave systems that it will soon power down the load.

  2. The upsmon slave daemon in your VM sees the FSD event and:

    1. Generate a NOTIFY_SHUTDOWN event.
    2. Wait FINALDELAY seconds, by default 5.
    3. Call its shutdown command specified by SHUTDOWNCMD.
    4. Disconnect from upsd.
  3. The upsmon master daemon waits up to HOSTSYNC seconds (by default 15) for the slaves to disconnect from upsd. If any are connected after this time, upsmon stops waiting and proceeds with the shutdown process.

  4. The upsmon master:

    1. Generates a NOTIFY_SHUTDOWN event.
    2. Waits FINALDELAY seconds, by default 5.
    3. Creates the POWERDOWNFLAG file, by default /etc/killpower.
    4. Calls its shutdown command specified by SHUTDOWNCMD.
  5. On most systems, init takes over, kills your processes, syncs and unmounts some filesystems, and remounts some read-only.

  6. The init process then runs your shutdown script. This checks for the POWERDOWNFLAG, finds it, and tells the UPS driver(s) to power off the load (your UPS unit).

  7. The system, your Proxmox VE host, loses power.

  8. Time passes. The power returns, and your UPS unit switches back on.

There are a few takeaways to consider from this sequence.

  • Timing is crucial: you need to leave enough time for your client (slaves for NUT) systems, like your VM, to shutdown properly. At this point, you only have one VM that does nothing but keeping itself running, so it can shutdown in a few seconds. Something else will be when, in later guides, you build a Kubernetes cluster with two quite service-loaded VMs. You'll need to adjust the value of the HOSTSYNC parameter (declared in the /etc/nut/upsmon.conf file), both in the master NUT server and in the slave systems, to suit it to the longest period of time required to shutdown safely your VMs, while also keeping in mind the limitations of the battery on your UPS unit. Thus, you'll need to test your whole system to measure how much time does your setup requires to shutdown safely.

  • Use of the SHUTDOWNCMD and NOTIFYCMD parameters: you can prepare shell scripts able to do certain actions, like stopping services in a certain order, and then set them in the SHUTDOWNCMD or the NOTIFYCMD parameter (both declared in the /etc/nut/upsmon.conf file). This is exactly what you've seen previously in this guide, when you created the /usr/sbin/upssched-cmd script and set it in the NOTIFYCMD parameter. In fact, you could just improve the upssched-cmd script and associate it with more events.

  • The UPS unit will restart: with this FSD sequence, the last thing the UPS unit will do is restart. Bear this in mind if you have other unrelated devices plugged in the same UPS unit.

Executing the FSD test

  1. First, as mgrsys, open one shell terminal on your Proxmox VE host and another one on your VM, then put them side to side.

  2. On the Proxmox VE host shell, execute the following upsmon command.

    $ sudo upsmon -c fsd
  3. Right after you execute this command, pay attention to the VM shell. You should immediately see the following output.

    Broadcast message from nut@debiantpl (somewhere) (Fri Nov 19 20:00:04 2021):
    
    Executing automatic power-fail shutdown
  4. On the Proxmox VE shell, you'll get the following output.

    Network UPS Tools upsmon 2.7.4
    
    Broadcast message from nut@pve (somewhere) (Fri Nov 19 20:00:09 2021):
    
    Executing automatic power-fail shutdown
  5. You'll lose connection to both terminals, first from the VM, then from the Proxmox VE system.

  6. A few seconds later, you should see (blinking leds) or hear (internal switching) your UPS unit rebooting.

  7. Switch your Proxmox VE server back on, and then also your VM.

  8. Like before, connect with a shell terminal to each system. Then, open the /var/log/syslog file on both terminals.

    $ sudo less /var/log/syslog
  9. On the Proxmox VE system terminal, look for lines like the following ones.

    ...
    Nov 19 20:00:01 pve upsmon[932]: Signal 10: User requested FSD
    Nov 19 20:00:01 pve upsd[923]: Client [email protected] set FSD on UPS [apc]
    ...
    Nov 19 20:00:09 pve upsmon[932]: Executing automatic power-fail shutdown
    Nov 19 20:00:09 pve upsmon[932]: Auto logout and shutdown proceeding.
    ...
    

    Those lines mark the beginning of the FSD sequence in the Proxmox VE host. Some seconds later, you should find this line.

    Nov 19 20:00:14 pve upsmon.conf: SHUTDOWNCMD calling /sbin/shutdown to shut down system

    That line indicates when the shutdown begins in the Proxmox VE server. Below it you'll see the log of VMs and services being stopped, storages being unmounted and so on.

  10. On the VM terminal, search for log lines like the ones next.

    Nov 19 20:00:04 debiantpl upsmon[593]: UPS [email protected]: Forced shutdown in progress.
    Nov 19 20:00:04 debiantpl upsmon[593]: Executing automatic power-fail shutdown
    Nov 19 20:00:04 debiantpl upsmon[593]: Auto logout and shutdown proceeding.

    That's the beginning of the FSD sequence on the VM. Not far below it, you should see this other line.

    Nov 19 20:00:09 debiantpl upsmon.conf: SHUTDOWNCMD calling /sbin/shutdown to shut down system

    It's exactly like in the Proxmox VE server, and also indicates that the shutdown process of the VM truly starts at this point.

With this test, you can check out the validity of your NUT configuration and your associated shell scripts. Also, you may consider using this mechanism to shut down your whole system (VMs included) with just one command, but you must be aware that Proxmox VE also comes with capabilities regarding the automatic starting and stopping of its VMs.

Relevant system paths

Directories on Debian VM

  • /etc/nut
  • /usr/sbin
  • /var/log
  • /var/run/nut

Files on Debian VM

  • /etc/nut/nut.conf
  • /etc/nut/nut.conf.orig
  • /etc/nut/upsmon.conf
  • /etc/nut/upsmon.conf.orig
  • /etc/nut/upssched.conf
  • /etc/nut/upssched.conf.orig
  • /usr/sbin/upssched
  • /usr/sbin/upssched-cmd
  • /var/log/syslog
  • /var/run/nut/upssched.pipe
  • /var/run/nut/upssched.lock

Directories on Proxmox VE host

  • /etc/nut
  • /usr/sbin
  • /var/log
  • /var/run/nut

Files on Proxmox VE host

  • /etc/nut/nut.conf
  • /etc/nut/nut.conf.bkp
  • /etc/nut/ups.conf
  • /etc/nut/ups.conf.bkp
  • /etc/nut/upsd.conf
  • /etc/nut/upsd.conf.bkp
  • /etc/nut/upsd.users
  • /etc/nut/upsd.users.bkp
  • /etc/nut/upsmon.conf
  • /etc/nut/upsmon.conf.bkp
  • /etc/nut/upssched.conf
  • /etc/nut/upssched.conf.orig
  • /usr/sbin/upssched
  • /usr/sbin/upssched-cmd
  • /var/log/syslog
  • /var/run/nut/upssched.pipe
  • /var/run/nut/upssched.lock

References

NUT configuration

Proxmox VE firewall

Navigation

<< Previous (G021. K3s cluster setup 04) | +Table Of Contents+ | Next (G023. K3s cluster setup 06) >>