ZFS or ?????

[hobsoni]

Hi,
I bought a new machine 5 weeks ago, and installed Proxmox VE. It has a 2.5" spinning rust drive, upon which I installed the O/S, and a 1TB NVME SSD which I set up as ZPS and created my VMs upon it. Everything works fine, but. ...

To my knowledge I have had 4 short power outages - all at night, all during thunderstorms. The two I was awake for lasted about 5 to 10 seconds. The Proxmox server is configured to auto start after a power failure. Unfortunately this reboot has failed twice with a message to the effect that it cannot find the ZFS file store while scanning. Booting then halts with a message on the screen (not attached) asking me to enter the root password or Ctrl-D on the keyboard (also not attached).

Due to my knowledge of Debian and Proxmox being a bit slight, I have re-installed each time. Fortunately I have only lost the time, and had the family grumble about no WIFI (pyHole is a VM), because both failures happened the night after I took a backup of every VM before bed.

My question - what is the best thing to do about this? Some ideas I have had.
Configure ZFS to hold less cached data for much less time. How? This will only reduce the risk.
Switch to ext4
Buy a UPS
(More) Googling to find out how to recover the ZFS file system. My efforts so far have failed.

Ideas welcome.
Ian

 

[mach]

If you had four power outages within five weeks, I would say UPS is what you definitely need.

There are drives that have power loss data protection, like these Intel drives but those are expensive enterprise drives. So UPS is a much cheaper solution.

 

[leesteken]

I have had some corruption (which was detected by ZFS happilly!) due to power outage but I'm surprised that you get a message about not being able to find the rpool.

Just checking: does a normal complete shutdown, disconnect power and press power button before reconnecting power and a cold start work for your system? Otherwise you need to increase the timeout that the system will search for the ZFS pools (rootdelay?).

Also: automatic start after power loss might cause the system to start while the power is not actually good yet and it may fail multiple times in a row. This really screams UPS.

 

[guletz]

Hi to all!

USP is only one part of the problem. WIth or without a UPS, most of the time the power spikes will go thru the UPS(medium-range price) without any problem especially during thunderstorms. What you really need Is a good ground circuit, and a power outlet with protection for short-circuit, U > 240 V, U < 180 V(I do not know the term in english, sorry), and the time of protection circuit to cut the power to be <= 2 mili secs(2-4 KJouls energy disipation)

So UPS + power outlet with protection + ground circuit = NO failure server during thunderstorms.

In my region where I live we have thunderstorms like one/month at least. Using this setup, I do not have ANY problems for years.

Good luck / Bafta !

 

[guletz]

Fortunately I have only lost the time, and had the family grumble about no WIFI (pyHole is a VM), because both failures happened the night after I took a backup of every VM before bed

... family grumble(only) ....how lucky you are

You could use another system like a Raspberry 3.141. Is cheap, and is very let say "greeen"! I have a situation like yours at my home, with a PMX server(nothing could be better ...) and a router for wifi/Internet(vlans) and a Raspberry.... After 22.30 my server go to sleep, and only Rasperry is my primary DNS, and the router is secondary.

NOTE: 3.141... = pi

Good luck/Bafta !

 

[hobsoni]

Just checking: does a normal complete shutdown, disconnect power and press power button before reconnecting power and a cold start work for your system?

Yes. This works perfectly.

 

[kenneth104]

You should be in rescue mode, it's not zfs problem, because you have only one system disk, a 2.5" spinning rust drive

So, you can input another same capacity as the same model (can work differently) and reinstall pve use software RAID1 mode
But one nvme on ZFS, it is easy to cause data loss ..If your data isn't important, it's ok