You must log in or # to comment.
Also it must come from the back region of France. Otherwise it’s just sparkling archive.
Ooh sparkling archive actually sounds really fancy, I’ll start using that
is it pronounced beaucoup?
Yes it frence for raid 0
It still is in France, it’s like color and colour
One thing I emphasize in every training I do is that you do not have backups until you know exactly how long it will take to restore.
That way you can tell your boss it’ll take three times as long and be hailed as a miracle worker, as Scotty intended.
Do people actually do that? Because that would be funny
Yes, because otherwise you never get a break.
I see you’ve never worked in corporate IT
I 100% started doing this years ago and work is far less stressful. Temper their expectations, because it’s rare for any boss or company to treat you fairly and honestly.
it’s a fair argument but it’s also bullshit if you’re following the process and practices that you used when you tested your backup
lots of my job is backups and verification of the backups
Bold of you to assume people/companies test backups more than once.
Case in point: I once got instructed to “enable EBS snapshots” for customer deployments to meet a new backup requirement. Disaster recovery was a completely different feature we only kind of got to a couple years later and afaik, remains manual to this day.
that’s fair and I agree but it’s not a true maxim
it’s a good principal but I hear it a lot so it’s a thing I get annoyed about because it’s directed at me even though I have the receipts and proven record that it’s not a fact
These kinds of relentless posts finally got me to write a script that verifies all my backed up files using md5 checksums.
Backup fire drill tomorrow!
Blake3(blak3?) is where it’s at
?
Edit: I see, it’s a hashing algorithm. I tried a few different algorithms and had some issues with large files, and didn’t really see performance increases over md5, since it seems reading the data from disk was the bottleneck. I didn’t try this one though. I wonder if it would actually be faster.
I did a large scale data rationalization and migration project for a company that is heavily regulated. They can be asked to prove they have this or that document from seven years ago, for no other reason than they should have it. Not having it means big fines and negative press.
Hundreds of Tbs of data got appropriately labelled and migrated, even more got left behind on the old system till it could be decommissioned safely after a period of parallel running.
As part of the decommissioning the data was backed up twice, and I wanted the backup properly tested with some random file restores. Not a full restore, just a few random restores just a proof of life test that the backups worked. I was told that wasn’t a reasonable request and it wasn’t needed as the architect in charge of backups trusted his backup team and he “designed pragmatic solutions”.
I still mean to call in to the regulator in a year or two to trigger a restore request, lets see if a pragmatic solution design is actually the same as performing some basic testing.
What if my backup is just files and there’s nothing to restore?
Like say I take my existing drives, full of totally working media, and duplicate them, use the originals as a backup and the new drives as the active.
Does that count as a backup? No restoration involved.
In the spirit of this thread: no.
Recovering with the backup should put you back to an operational state equivalent to when the backup was taken.
I.e. if you’ve restored some files, but something is still not working then the backup failed its purpose.
E.g. the timestamps on the files might be important, do they need to be stamped with the time of the backup or the time of the restore?
Sure, if my active drives died after this swap, and I had to restore from the old, now backup, drive, I’d be back at the operational state I was at the time of the backup.
That tracks.
It still doesn’t run anything tho. It’s just a drive. It doesn’t house an os or anything, just files that aren’t restricted in any way.
IMHO there is no point backing up an OS drive, just rebuild it*.
Data is the important thing to back up because you usually can’t regenerate it.
* the corollary here is that you’ve backed up the configuration required to rebuild the OS.
I wouldn’t, I keep all of my data separate from my OS drive entirely so I can reformat or install a new OS whenever I feel like… nasty old habit from bootleg windows 7 well beyond its age, when reformatting every 6 months was good hygiene, before I found Linux… but gave me great data management insight.
An untested disaster recovery plan is wishful thinking
ZFS isn’t a backup, but it is a gateway drug
I get it, esp. in a professional environment.
But “Schrödinger’s data” rubs me the wrong way. The point OOP’s making is not a question of whether the data is there or not, it’s a question of whether you can restore a botched system with a few commands and in a realistic amount of time.
Case in point: I (private person, private system) never needed to fully restore, knock on wood. But the data is there - I have (manually) restored single files or directories on a few occasions.
If the data is present but difficult to restore, it’s annoying. You might need to spend a few days fixing stuff.
If the data is gone, it’s devastating and can bankrupt a company. On a personal level it’s the same as having all your photos destroyed in a fire. And backups not containing the right data are very common.
it’s a question of whether you can restore a botched system with a few commands and in a realistic amount of time.
A few years ago my employer was the victim of randsomware. We’re speaking here about a massive network and all sorts of databases and services build on top of those, spanning decades and many different technologies. Basicly several thousand employees and a decade long focus on working digital and automation. Data restoration was not an issue. I haven’t heard of anyone losing data.
However, restarting all the services was not as easy. Many of these depended on each other and there were some circular dependencies that have grown organicily over the years. Took about two months to restore core functionality (mostly SAP and email) and many more months to restore all sorts of support services that were required for normal day-to-day work. Two years after the incident the last applications were back online.
The next day, the novice’s disk crashed. Three days later, the novice was still reinstalling software.
I laugh in NixOS
Okay, fair, upgrading and clean installs take a while. Specially on slow internet connections. But it doesnt really compile much, we have cache.nixos.org with prebuilt binaries for most packages.
Adding another link, perhaps not equivalent:
I use it to backup media. If the backup roots are your drives, restoring is as easy as copy-paste, merge folders and wait.
Edit: actually, you may be able to just reverse the direction of the sync. I’m not sure.
Has anyone actually had failed backup restoration before? It’s been a meme forever, but in my ~15 years of IT, I’ve never seen a backup not restore properly.
Yep. At one place I worked, we did a big off-site disaster recovery exercise every year.
Most of the time it went fine, but there were multiple years where a restore didn’t work due to an issue with one or more tapes. Either the data and/or indexes couldn’t be read, or the tape physically failed during the restore.
Backups aren’t backups unless they’re tested.
I’ve had an IT Career for about as long as you. I’ve had 2 memorable restore failures and got real lucky both times.
The first was a ransomware incident, and the onsite backup was not hit, but it was corrupt. Thankfully, the client had been using a 3-2-1 strategy, and the off-site one was fine.
The second was a situation where a failed update rendered a client’s RDS unbootable. This time, they didn’t have an on-site backup and the off-site one was corrupt. This time I happened to get immensely lucky in that there was no real data on that RDS, so I was able to spin up a fresh one, and install their LOB app and all was good.
We now test that all backups are stable every 6 months.
Absolutely. Used to work at a small MSP. Got ultra unlucky in that we got chosen as the rest case target for a zero day that leveraged our Remote Support tools so our own systems and all of our client systems that were online got hit with ransomware in a very short time frame.
Some clients had local backups to Synology boxes and those worked ok thankfully. However all the rest had backups based on Hyper-V. The other local copy was on a second windows server that also got hit so the local copies didn’t help. They did also have a remote copy which wasn’t encrypted.
So all good right? Just pull the remote backup copy and apply that… Yea every time we had ever used the service before had either been single servers that physically died and took disks along on the death or just file level restores.
Those all worked fine. Still sounds like not a problem right? Nope. We found both that a couple of the larger servers had backups that didn’t actually have everything in spite of being VM images. No idea how their software even was able to do that.
And the worse part was that their data transfer rate was insanely slow. About 10mbps. Not that per server or par client. Nope that was the max export rate across everything. It would have taken literally months to restore everything at that rate.
I hate to say it but yes we did in fact pay the ransom and the. Had to fight for several days going through getting things decrypted. Then going through months of reinstalling fresh copies and/or putting in new servers. Also changing our entire stack at the same time. Shockingly we handled it well enough we lost no clients. Largely because we were able to prove we couldn’t have known ahead of time.
If you read through all that I’ll even say the vendors name. It was StorageCraft. I now have a deep hate for them.
Also one more is that with the old Apple HFS+ filesystem based time machine backups it would sometimes report as a valid self checked backup even if it had corruption. It would do this as long as some self check confirmed that it could fix the corruption during a restore. However if you tried directly browsing through the time machine backups it would have files that couldn’t be read, unless again you did a full system restore with it.
Nearly lost my wife’s semester ending before finding it worked that way.
I can’t confirm it but seems it is fully fixed with APFS and might be one of the reasons they spent the effort to make that transition.
in my ~15 years of IT, I’ve never seen a backup not restore properly.
I remember Outlook backups failing like nothing else during the restore process 25 years ago.
Which was fucked because it would take 2 weeks to rebuild only to find out it didn’t work.
Fun video. Many backup options failed iirc.
Banger of a video! Thanks!
I’ve made mistakes before, and had that panic realization set in. I can only imagine the feeling this guy got once he realized what he just did. Nightmare fuel.
Damn that’s good
I just run a script that runs a bunch of rsync commands. So I guess every run kinda confirms the backup is functional. I have no use for versioned backups, nor could I afford the hard drive space necessary (thanks Sam)
No it is not. Just use cheksum. Like a normal person…
Cool.
Still doesnt mean you can boot from itChatGPT told me to do
sudo sha256sum /dev/sda1 > /dev/sda1
So. Is this wrong? I thought it backs up the data
Yeah we’re going to see much more of this moving forward. Yesterday i installed Linux for a friend and they asked about fixing problems. I told them to always look at the date & compatibility when they search for solutions. They then volunteered: “and I guess I can always ask ChatGPT, it’s pretty good with these things”. I grunted non-committally.
Or that it’s complete.
What?? Why do you presume the data is bootable? I’m confused, isn’t this about restoring data from some backup data? Am I missing something? Is it a sex thing? I can never tell here at Lemmy…
