Upgrading - following the pfSense docs 'Installing and Upgrading' includes having a fall back plan

Tzvia

@guardian Good questions. I would first verify that when I do a backup of settings, that UPDATE/SYSTEM UPDATE is set to PREVIOUS STABLE. You can then also verify that the software versions of anything specific to 2.45p1 are still available in PACKAGE MANAGER. Provided they are listed, applying your backup to a fresh install of 2..45p1 should set it this way and it should be able to install all the packages when it reboots...

fabrizior

@guardian
Have any recommendations on a usb-bootable tool for step 4, preferably to make a bootable backup copy. not well-versed in FreeBSD SysAdmin.

guardian

@fabrizior said in Upgrading - following the pfSense docs 'Installing and Upgrading' includes having a fall back plan:

@guardian
Have any recommendations on a usb-bootable tool for step 4, preferably to make a bootable backup copy. not well-versed in FreeBSD SysAdmin.

I was hoping someone who is BSD experienced would comment.

IIUC as long as I am going to image the whole disk and the backup and restore are done from another running system (i.e. boot from a USB disk or second drive) almost any imaging method will do All we are doing is blindly copying bytes. If it is necessary to restore, then the whole image must be restored to the same disk, or one with identical geometry.

My plan was to use Foxclone. I installed a copy of Linux Mint and Foxclone on an old 320GB USB hard drive. I have about 300GB free to copy drive images. That should be more than enough. I was thinking that I could even do something as simple as:
Backup:

dd if=/dev/sdx bs=4096 of=backupdir/pfsense.img status=progress

Restore:

dd if=backupdir/pfsense.img bs=4096 of=/dev/sdx status=progress

Can someone confirm if my understanding is correct or not and/or a better tool to use.

fabrizior

@guardian
How’s this sound as a rough starting point?
I don’t have a dev system to try this on right now... VM infra is busy with other projects ATM.

For the purposes of switch-over / fail-back with upgrades, and to maintain RTO/RPO SLA objectives against drive faults (not catastrophic system-wide HW/electrical faults) I’m considering:

Add a second internal drive (same size)
Could try this with an approp-sized removable flash drive for off-line / off-site backup safety vs ease of switchover/fail-back, depending on SLAs, I suppose.
partitioning new drive appropriately and make it bootable (values may vary, only an pasted example)
Q: can anyone help validate details with a pfsense env?

gpart destroy -F adaX
gpart create -s GPT adaX
gpart add -b 40 -s 128 -t freebsd-boot adaX
gpart add -s 1880 -t efi adaX
gpart add -s 4G -t freebsd-swap -l SWAP adaX
gpart add -t freebsd-ufs -l RECOVER adaX

gpart bootcode -b /boot/pmbr -p /boot/gptboot -i 1 adaX
gpart bootcode -p /boot/boot1.efifat -i 2 adaX
gpart set -a bootme -i 4 adaX

newfs -ntEU /dev/adaXp4
tunefs -a enable /dev/adaXp4

periodically cloning the current boot drive to the other. (with additional script logic to set source/dest parameters and manage the mount point, etc...)

use sysutils/clone to make a perfect clone of your system’s startup disk as follows:

mount -o noatime /dev/gpt/RECOVER /mnt
clone -c rwoff / /mnt

After the file system cloning has finished, you only would need to edit /mnt/etc/fstab:

# Device            Mountpoint      FStype  Options Dump    Pass#
/dev/gpt/RECOVER    /               ufs     rw      1       1
/dev/gpt/SWAP       none            swap    sw      0       0
proc                /proc           procfs  rw      0       0

clone(1) can be used to keep file systems synchronized, and you would use the very clone command together with the -s flag for keeping your recovery disk updated. In this mode only changed files would be copied and synchronization needs much less time than the initial full cloning process.

Alternatives? Is there a better way, still w/o 3rd-party software? dd doesn’t have an incremental sync mode like clone(1) does... while block-leve copies might be technically faster, dd is an all-or-nothing operation. clone seems like it would have some advantages after the first time.

Any other (perhaps pfSense-specific) partitioning/boot details to account for?

credit:
found the idea and example code from an 6-Aug-20 reply by @obsigna on the freebsd forum.

Patch

Given Negate policy to remove access to old binaries as soon as a new versions is released the only safe strategy I can see is for all users to individually maintain their own archive of any version they use. So

When wanting to upgrade pfSense by any method, first download and archive an iso in case you need it at a later date (note you can not reliably do this later as Netgate will pull the binary before you try to upgrade the next time).
Upgrade pfSense by whatever method you prefer
If the upgrade fails for you, use the archived version (the one you archived during a prior upgrade

Not efficient for a community but at least it ensures a reliable upgrade fall back from a known good source.

guardian

@patch said in Upgrading - following the pfSense docs 'Installing and Upgrading' includes having a fall back plan:

Given Negate policy to remove access to old binaries as soon as a new versions is released the only safe strategy I can see is for all users to individually maintain their own archive of any version they use. So

When wanting to upgrade pfSense by any method, first download and archive an iso in case you need it at a later date (note you can not reliably do this later as Netgate will pull the binary before you try to upgrade the next time).

Upgrade pfSense by whatever method you prefer

If the upgrade fails for you, use the archived version (the one you archived during a prior upgrade

Not efficient for a community but at least it ensures a reliable upgrade fall back from a known good source.

Good for the base version, but what about the plug-ins etc.?

fabrizior

@patch Any thoughts on an appropriate process to make a bootable clone of the boot disk?

Thought being: if an upgrade fails, change the boot device to the backup disk and reboot to the clone... back online almost immediately. then clone the booted backup back over the failed upgrade image to restore DR capability before attempting upgrade again at next maintenance window...

Anyone have practical experience with the partitioning/cloning a pfSense drive?

Would this need to be done offline?
If attempted online, any pid/lock/runtime state files that need to be excluded or reset on the clone?

Patch

First line upgrade strategy is for the upgrade to work as expected.

A fall back plan for me needs to be much more reliable than the upgrade path, hence the desire to return to a reference state (original install media & backed up configuration file).

Dynamic snapshots are useful if you have been using them regularly so have validated their functionality. Relying on them for a once off fall back plan is too risky for me, or if validated, too time consuming for something I intend not to use.

Dynamic snapshots do make far more sense if you run pfSense on a virtual machine (but then you are in the "using them regularly" group).

fabrizior

@patch Thanks. That standard process is probably a good fit for many. [edit] Though, one might think the most “reliable” fallback plan would be a direct clone of the (presumably stable) system immediately prior to upgrade. Having to reinstall and reapply a config reliant on specific prior versions of packages that may or may not be available...

From my point of view, we could come up with a quicker, less maintenance-intensive process from an RTO metric / SLA perspective (when relevant) without resorting to clustered hardware.

The recovery time in case of upgrade failure when the process is to re-install the base image, restore the config, validate all the packages (hopefully at same version as pre-upgrade), etc... isn’t the quickest process and somewhat manually labor intensive (in comparison.)

Executing a scripted incremental update of a cloned disk (especially if done regularly, and if could be done online (?)) just-in-time prior to the upgrade maintenance window would allow for a faulted upgrade recovery time no longer than what it takes to reboot.
Much reduced recovery effort seems valuable enough to continue investigating how this could be done.

Only downsides seems to be the cloning process discovery/analysis and the cost of a second disk.

itpp21

Reading this with interest, I'm looking at an usb stick recover method and the possibilities of adding packages from a running environment to this stick. As recovery(installation) by stick is already supported we might expand this by incorporating packages.
One question is will the usb stick auto install or do you still need console access to get it running? if so can you force auto install so it becomes a matter of insert stick and boot?
(yes you can yell at me) but under windows you can extract the wim file and modify it to your liking, repackage the wim and at boottime push out additional packages without a key pressed to add on a fresh installation.

Gertjan

@itpp21 said in Upgrading - following the pfSense docs 'Installing and Upgrading' includes having a fall back plan:

(yes you can yell at me)

Noop.
Just 'gime' 1 % of 1 % of the Microsoft develop budget and you'll have that option ready for tomorrow.
Btw : with MS conditions of course : no documentation what so ever, and a real "lets test it in the field and see what telemetry brings home".

itpp21

I don't get the reply, if the stick allows pfSense installation (which it does) then why can't it be modified to auto install and why can't you add stuff? Its just an 'image' bits and bytes. I remember the ubuntu iso's which when you don't do anything it times out of some default options and then deploys.

Gertjan

You can modify the ISO before you use it.
What's stopping you ?

( ok, you need to know how, what to place where, how things work, have plenty of time to test etc )

itpp21

@gertjan said in Upgrading - following the pfSense docs 'Installing and Upgrading' includes having a fall back plan:

( ok, you need to know how, what to place where, how things work, have plenty of time to test etc )

Exactly, I've ordered some sticks for this, will get back when I get anywhere, maybe others find this idea interesting to assist in the how, where and what.

itpp21

The easy solution:
Create a pfSense VM with virtualbox
Load packages and settings from your bare metal device
Verify the config works as bare metal.
Shut the VM down

Restore:
Use vboxmanage to convert vdi to vhd
Use rufus to load the vhd on the ram device (taken out of the bare device mounted via usb)
Place storage back, done. (and tested)

I'm gonna test the reverse (take bare metal in to a VM) but have no doubts this will work.

nb. cloning the storage device as backup also works.