@martin @daftaupe
Tried btrfs à long time ago. Snapshots Can be great for OS Drive. But for data storage nothing Can Beat mdadm+lvm2. Big problem with btrfs : you can't know how much space is still available. (At the Time i checked it at least.)

@hoper @martin You can use btrfs filesystem df for example to know how much space is available (or used). or btrfs filesystem du to get details about a particular file.

@daftaupe @martin
Seems that things change, but according to the FAQ, there are still some issues with this matter (know how much Space left and use it)
btrfs.wiki.kernel.org/index.ph
Il still think that mixing réplication (raid) and file system is a very bad idea.

@hoper @martin Well I'm using it and liking it. I never liked lvm2 and mdadm. That's the beauty of having different alternatives.

@daftaupe @martin
With how many disk ? Did you already Lost a drive or two ? WhatsApp happens then ?
Already loose 2 drives on a raid5 with mdadm and manage to get back all data. (One drive was really out of order, the other one was just suddenly disconected)

@hoper @martin Good for you but I make backups, so I'm not afraid of losing disks :D which is really easy using btrfs send/receive feature.

@daftaupe @martin
Backups are Always needed of course. But it's a "last ressort solution" because when you restore you loose data... Maybe a day of work if you backup every night. So, be able to loose disks without any impact (notre even the need to umount the filesystem) is important.

@daftaupe
You said you have these magic send/receive , did you already them somewhere? Or could you them maybe? 😬🙊
@hoper

@martin @hoper There's nothing magical :) it's basically having 2 btrfs partitionned drives, making a snapshot of a subvolume and sending that read-only snapshot from one to the other disk. That prevent us from the use a of a tool like rsync as there's an option to send only the increment when you already have subvolumes on the other drive. Based on : btrfs.wiki.kernel.org/index.ph

@daftaupe @martin
So... Just to be sure... Sending data to another drive on the same server is what you call a "backup" ? In this case, it's not only about tools that we disagree :)

@hoper @martin I did mention another disk or drive, it's up to you to think that it's in the same server.

@hoper obviously the (incremental) snapshot should be sent to a backup-machine... 🤓
sending it to another disk in the same server sound like poor-man's raid - and as we know, has better raid options 💾

@martin
In m'y mind, a snapshot is not a clone. It's just a sort of "diff" of two devices. Something that you can't separate from it's source.
So I dont understand how you can have a filesystem somewhere, and a snapshot of it elsewhere that can be usefull. A copy, yes. Easy. But a snapshot ? Guess there is plenty of things I dont understand about btrfs.
Maybe the first reason for me to not like it :o)

@daftaupe @martin @hoper

it's not magic it's just

[speaks 14 magical words and all of the shadows in the city disappear]

@hoper
That entry is indeed good to know! And per-file levels sound like an interesting concept. Thanks for sharing 🍫
However, you're right that things makes a rather impression... Bit -like maybe..? 🤔
@daftaupe

@martin @daftaupe
A "systemd like solution" LOL. Exactly ! As long at it's work as intended, perfect thing. But when you want to change, add or remove a thing you can't because it's far too complex tu use and really understand.
Mdadm and lvm2, both are doing a only thing but they did it well. Can't make differents raid level per file, yes. But at raid level you can do anything (reshaping, and so on)

@hoper @martin Well maybe snapshots are not your use case (RAID is not mine). Keep your tools if you think they're good, but stop trying to convince people not to use new tools that might fit their needs.

@hoper @daftaupe @martin why would you think mixing FS/RAID is a bad idea? I'd argue it is a very good idea :)

@fink @daftaupe @martin
This break the "layer" model. i have 3 layer : raid, volume management, and fs (Can Swap the two first). If one of them is upgraded or replace by another, no problem. This also respect the unix philosophy : do a single thing, but do it well. Computing had always used layer model (network...). Because it's good and reliable. And for me, keeping my data safe is the most important thing.

@hoper @fink @martin Talking about data safety I wonder why Synology is recommending btrfs (they probably want to take the risk of putting their customers data in danger), synology.com/en-us/dsm/Btrfs. I think you might be interested about the paragraph related to data corruption. By the way which fs are you using that provides the same features ?

@daftaupe @fink @martin
Just checked this link, thanks.
Metadata miroring is a FS feature. For exemple ext4 make many copy of the superbloc. Guess that one day ext(5?) will offer more. In fact, I don't feel the need to have multiple copy of metadata because :

hardware problems (badblock) can't be an issue anymore with raidX

Logical problem (silent corruption) is also an issue already solved by mdadm with raid6. (callded "self healing" on synology link)

For last feature, snapshots >LVM.

Ok, your stack is the best, end of the story.

By the way, talking about links, what about his one ?
btrfs.wiki.kernel.org/index.ph
"The parity RAID code has multiple serious data-loss bugs in it. It should not be used for anything other than testing purposes."

Or this one :
phoronix.com/scan.php?page=new
Saying that very serious bug about raid have been corrected in linux 4.16. (Sorry, I'm still using a 4.9 kernel, and I think this is not so old...)

It's good to see people making news things, and other in a hurry to test it...

Me ? I will wait :)

@fink @daftaupe @martin
So yes, btrfs have more functionality. Some of them can be really great for OS drive (said it before). Allow rollback after an upgrade, snapshot per file and so on. But for the data... I don't need more functionality, I need something 100% stable, repairable on worst scénario, and wich make me feel secure. Because I can tell exactly how my files are stored, exactly where on the drives etc.

@fink @daftaupe @martin
Last thing about the layer model. Sometimes you don't need all layers. For example you don't always need a file system. Think about swap, or about oracle or other database software that will work better if you just give them raw devices, because they have their own integrated "file system". Can you make a "raw volume" (without any fs or metadata or whatever on it) with btrfs ?

Sign in to participate in the conversation
Binfalsetodon.

mstdn.binfalse.de is one server in the network