As to best practices go RAID 1 or 10 or Single if you don't need raid. If you can duplicate your metadata 3 times (RAID1(0)c3) Do a srub every month or two, doesn't need to be often and after that do a btrfs balance. That's pretty much all. Tips: Use reflink copies whenever possible (cp --reflink always) It will create copy without physically copying the data. Only the new metadata will be created, similar to sym/hard links. It just doesn't come with any of the drawbacks. It's like dedpulication and can be used like that. ;)
@@duser yeah it's really useful for dealing with VMs, cloning them with that flag is pretty nice and safes not only time but space as well. Good thing is that latest gnutils set the --reflink auto by default. So whenever you copy stuff it will get reflinked. Nautilus on Gnome also does the same
I'm late to the party, but here are a few things: 1. When taking a snapshot of a location that has a snapshot, are you not adding that existing snapshot into the new snapshot? Or is BTRFS smart enough to exclude snapshots? 2. Yes, please make more BTRFS videos 3. Please make BTRFS related videos with TimeShift, BackInTime, and Snapper as the focus 4. Please make a BTRFS video focusing on btrfs send | btrfs receive
My top btrfs configuration options on Arch: Create a subvolume at /home, /tmp, /var/log, /var/cache, /var/lib/flatpak, /var/lib/libvirt, /srv, /root, /opt, /usr/local and of course /.snapshots, also ideally you want to replace the root subvolume so I create another one and mount it at /. For mount options: noatime,compress=zstd,discard=async,space_cache=v2,ssd_spread. Disable CoW for /var/lib/libvirt for better performance. And most importantly, even if you take snapshots of your root subvolume, you won't be able to boot, so I have a pacman hook that triggers every time I upgrade my kernel, which runs rsync -a --delete /boot /.bootbackup. Small tip is to also delete the subvolid from fstab (it's not needed, subvol is enough) so you can easily restore a subvolume.
Yeah good recommends! But for me those would be too many subvols to keep track of and snapshot. I just have @ @home and libvirt images of course. I really don't like the naming but that's what you need for Timeshift........ Also yes you really shouldn't use subvolids in fstab or anywhere. Just renaming the subvols to rollback is a must.
The most optimal defaults differ form usecase to usecase. However it would be nice for fedora to step up their btrfs game, as it comes by default, the current one is an extremely vanilla setup and definitely less than ideal.
As usual, a very well researched and well-presented video from you. Good balanced overview of BTRFS, with the pros and cons, and where to looks for updates on that. It is great that a modern file system is being actively developed, and glad you highlighted it. Thanks Jay.
My Synology is using BTRFS for about 4 years now in a SHR raid. Its working like a charm and even disk replacement and raid rebuild, went flawless. Stable, stellar performance and deduplication is an awesome feature.
Bro, I’m a little high right now so my sense of time is a little warped and it makes me pay much closer attention to your voice tones, and let me tell you… you have gotten a shit load better my brother. Much props on the high quality effort on the script and prep work you did. You really took the time to explain nuance. I have bookmarked this to rewatch while not high lol
Excellent video Jay. Thank you very much. I have looked at BTRFS for a while, but never dove into it. Going to try it in a VM tonight. Please more videos on the subject. Also have you done videos on "backup best practices?" I would very much enjoy watching that. Thanks and Cheers from Canada!
It's only really RAID 5/6 (stripping) that is not completely stable with BTRFS. Actually, it's implementation of RAID1 is one of the best things about BTRFS for desktop computers in my opinion. Having installed an OS on two disks using RAID0, I was able to on-the-fly, without rebooting, in a matter of seconds, switch to RAID1. Something even the mighty zfs can't do for now (changing vdev structure).
This! RAID 0 and RAID 1 are perfectly fine and work even better than md-raid. It's just RAID 5/6 that are incomplete (although errors are extremely rare with proper use - similar guarantees as md-raid on Linux kernels prior to 4.4).
If RAID 5/6 had never been released, the BTRFS story would be really simple: “It’s a very reliable, flexible, and full feature modern file system for the desktop, and and anywhere else you don’t need big pools of disks.” As a bit of a ZFS zealot, I have to say BTRFS is really good.
"Something even the mighty zfs can't do for now (changing vdev structure)" Nonsense. ZFS can easily add/remove disks from stripes, and add mirrors to those disks--i.e., do exactly what you described that you did. But parity RAID vdevs can't (currently) be changed; that's been Coming Soon™ for a number of years.
@@LucS0042 Well, there's no such thing as a "two disk stripe vdev"; that would be two single-disk vdevs striped together. But with that caveat, it seems a strange thing to want to do, but yes, ZFS can do it.
A really good video about btrfs. btrfs is a great FS. I use it since 2012. Never had any problems with btrfs. N8ever used Raid5/6 - only the so called "RAID1". Maybe that is why I never run into any problems...) One of the best features is to do read only snapshots of our home directory and use them for making backups via send / receive to another btrfs device. (It took me some time to understand the construct of send / receive - but finally it works :)
BTRFS is awesome and it's much lighter than ZFS. I worked as a storage engineer for ~7 years and that's what I've learned during tons of tests and implementations - you DON'T need ZFS and BTRFS is really highly usable these days.
I used BTRFS for @, @home, and @data subvol. I do dual boot Ubuntu with windows. With BTRFS it can share file directly with WinBTRFS. My laptop small SSD can fit much of data with it's compression. Some game like DotA 2 compressed at almost 50% ratio. I install all my program at BTRFS, also set my fav folder like download, document to @data subvol. Data deduplication also work in Windows. That is very much better integration than old school ext4.
As far as I could tell from every other source, the BTRFS 'raid likes' are all fine and commonly used as defaults EXCEPT for raid5/6 implementations. But please do look at multiple closer-to-official accounts of btrfs raid profiles
Thanks - I learned a bunch We have BTRFS running on a Synology NAS but EXT3 / EXT4 / ZFS everywhere else. It was good to hear a bit on how BTRFS works.
Great video! I'm always interested in filesystems and ways how to manage it. It would be awesome to have a video on what are the optimal ways of managing a btrfs filesystems checking for errors, repairing, etc. I have btrfs on my main Linux machine and also on my NAS. I'm always on the lookout for best practices on how to secure my data.. especially when it comes to the filesystems I'm using which is btrfs. 😅
This video helped me understand BTRFS better. Great work! ...also Snapper is great. I set it up following a different TH-cam video. It's awesome for people who want snapshots to be automated. It's in the RPM repos. (I'm also using Fedora.)
Love the content 👍🏼! Yes, kindly provide a refresh on BTRFS and it's applications such as for virtualization (proxmox,etc,.), improvements, practical uses and such. Apparently, some of the major tech giants use it in their environment and it's natively supported in the Linux kernel.
I found btrfs due to Fedora and Garuda. I love how Garuda implemented snapper, was very helpful when I was more experimental and overall just give me peace of mind. With Steam recently supporting CoW, would it be worth it to have my game partition as btrfs?
I write too long stuff. TL;DR: Yes. I use it now for few years for hundreds of games spanning over 3.5TB, btrfs just works. I use @games subvolume, mounted at /mnt/games. I use and recommend Steam flatpak, fully sandboxed (all access to my files revoked), gave access to /mnt/games:rw (read write permissions from flatpak), just works. And it can’t mess with my personal files at all. btrfs is the successor of ext4, it basically can do over 99% of what ext4 can do, but much more at same, little worse, or even better performance. Of course they are extremely different fs, but for end user, I can’t say anything other than it just works. When paired with ECC RAM. But if not having ECC RAM, the other fs suffer the same, but they don’t tell you. Btrfs on the other hand knows e.g. when your file integrity is broken. But btrfs trusts the CPU and the RAM, so it will tell you your disk is broken or that cosmic microwave background radiation hit your NAND cell and broke some data on the disk, it won’t tell you data read is correct, but the one in the RAM is faulty, it does not know that, it thinks errors detected are from the disk. It assumes your CPU and RAM do not lie. When using btrfs raid 1 or 10 (with data copy) it periodically checks file integrity automatically, cause it hashes or checksums each file (always, that’s how it detects errors), if it detects mismatch, it will automatically check the other disk and repair broken data automatically, cause it knows if something is broken, and in RAID 1 case has a fully functioning copy. So awesome for system and documents and so on. If you don’t care about games integrity you can just ignore that feature, but it is still a good fs. You can use snapshotting, e.g. when modding a game, when an update broke your game mods, you can then look into snapshot and restore stuff or whatever. Only limitation is your imagination. It is better than ext4, allows resizing, and whole lot more features, comparable to zfs,.. so of course it is a good option. It is also newer than zfs, in my view it is better than zfs, but it depends. E.g. BSD NAS should use zfs. But for desktop, or server, always btrfs. But still, ext4 is good filesystem as well, as well as xfs, they are good choice as well. For games it does not matter that much. I recommend for system and home, two NVMe SSD, e.g. Micron 7400 Pro with power loss protection, using dm-crypt / luks on them, on top of that btrfs (in RAID 1 fully copy mode), and ECC RAM. Then maybe unencrypted additional SSD, e.g. FireCuda 530, btrfs for games. ECC RAM is no bottleneck in gaming. I just ‚downgraded‘ speed, from 64kMiB 3.6GT/s cl16-19-19-19 to 128kMiB 3.2GT/s cl22-22-22 ECC RAM - literally no difference in FPS (I play at 4k). If you have stutter in CSGO, right click, properties, compatibility, choose Steam Linux Runtime, working in Flatpak as well, they made it compatible, sandboxing works with native performance.
10:12 No you absolutely should use RAID with BTRFS, just not R5 or 6. RAID 1 and 10 are much superior to mdraid or hw raid. Just striped RAID setups suffer from write holes if you're not careful.
@@squishy-tomato fair, but I just wanted make it clear that RAID1(0) is indeed stable and totally usable. I already have some corruption from time to time, because of a bad controller. BTRFS does it's job great and corrects those flipped bits just fine. And yes I should definitely swap the controller, but cannot right now...
I was wandering what happened BTRFS .. it in fact has been around for decades and every now and again it raises its head and comes to the fore front with new features. I remember it back in SMS days when I worked there.. where is was played around in UNIX on Solaris. I have always liked the BTRFS concept and idea of use and I an happy to see it once again come to the fore front as commercially viable filesystem. Thank you for your introduction and viewing. I personally would like to see more about it and where it is going currently.
I would elaborate more about the benefits, purpose, and compelling reasons why an average user might need to use BTRS. A person has to be motivated enough to use or even try it. Because managing it, especially without the possibility of using GUI, looks pretty complex to the average home user.
He hinted GUI end use it when he mentioned TimeShift but he did not go into it. I agree with you that the average Linux beginner (myself for example) is not going to want to modify fstab files manually or run all these command lines.
Would love to learn more about what Fedora 36 does or doesn't do out of the box to protect your data. It's kinda important to know whether scrubbing and automatic snapshots are your responsibility or not.
btrfs goes back many years ago. It was first used as a file manager for an accounting package. It has had an up and down history. One of the issues was once you used one application with btrfs you can not use another app using btrfs. It was a very popular file system many years ago and has reinvented itself and it's popularity thanks to Linux.
I use it for years on my very old laptop. Very satisfied, snapshots are great. And you can do things you can't with other fs even we have to be carefull
Thanks Sensei ! 25:23 if I understand correctly, you're saying that : "despite BTRFS subvolumes being stretchable, without its quotas feature, the system becomes unusable when every bit of subvolume freespace becomes redistributed/saturated (e.g because of a logfile out of control). And since as of 2024 the quotas feature is still not ready, BTRFS on LVM appears mandatory/convenient for : hot-resizing/hot-bay-expanding/hot-bay-removal/insertion in a LVM rig. BTRFS itself seems enough for partitioning/isolating into subvolumes, but without LVM, users have to know beforehands the allocated storage sizes." So now I wonder why I saw nobody on YT putting up a guide on such way of installing (e.g for a Debian install + switch to busybox in the process) : efi + boot + PV, then on PV, a LV for root and a LV for home at least. BTRFS with subvolumes on LV root, BTRFS or EXT4 for LV home. Snapper integration able to snapshot only few subs from LV root, as well as LV home. (Optionally, LUKS sandwiched on LVM and under BTRFS would be the cherry on the top. So far, only skai nyght made it this way on YT. But him and EF (both Arch guys) encrypt swap in a way that prevents hibernation, so unconvenient for nomad laptops...) Maybe I'm missing some content, i don't know, otherwise feel free to hand out the idea to Tom if you already scheduled bunch of other content already ! :)
It would be awesome with a new btrfs where you advice on some efficient backup systems, how to interact with external drives, NAS and possibly even syncing across clients. Maybe one practice for server backup and one for desktop/laptop.
First, sorry for my poor English, i'm slightly dyslectic... The only reason I still have 2 Windows computers, is a program called RollbackRx. I have heard/read about BTRFS, and wondered if the works the same way as Rollback Rx And your GREAT video, shows me, it is almost the same idea and same functions, though all from the terminal. Thanks for showing some of this functions. -------------------------- I have for several years used RollbackRx. for Windows I bought 5 Pro licenses years ago and over time updatet to new edtions. As I wrote I have only 2 Windows computers today, and Rollback are the main reason. But I have had an eye on BTRFS for some time, but has been unsecure, since the sometimes negative critics. With your video here, I can se that most of the reputation is mainly a userproblem. (no critics to BTRFS from me, it is complicated, but not impossible.)
Good video. I find it much much less confusing and error prone to use the subvolume name than subvolid in fstab, and that's what I would recommend. FWIW, timeshift also works seamlessly that way with no need to manually change fstab.
Also don't forget about the usefulness of reflink copies! It will make a deduplicated copy of a file, meaning that it will only copy the metadata of the file and link the data to the already present blocks. If you modify normal CoW will kick in and copy only the modified data. This is why you can copy multi gig isos on Fedora (within a mountpoint) in microseconds. Try it out with Nautilus ;) This really helped me restore movies from snapshots that I had accidentally deleted. You just need to mount the root subvolume to keep the copy within a mountpoint.
LVM is easier to implement then btrfs. It is probably easier to use btrfs, but LVM is easier to implement, and then you just add ordinary file systems on top of LVM. But btrfs is combination of LVM and a modern fs. But it is actually another implementation based on ZFS, which is better then btrfs, as it is older and design by people that actually knows how to run huge servers. That said, btrfs features are not bad, and a bit lighter then ZFS. So all features in btrfs is in ZFS, but not everything in ZFS is implemented in btrfs (like RAID etc).
LVM and btrfs are not for the same purpose. The first is a system for managing physical or logical volumes and disks but sencond is a file system. We can user btrfs in out LVM logical volumes.
Fantastic video. Please make a video on how to convert from existing ext4 install to a BTRFS instance. This would save so much time from reinstalling and configuring some linux systems. If that is possible!
Check "man btrfs-convert" for documentation. It can convert an existing ext4 filesystem in-place into btrfs (without losing any data). If you decide that you didn't like btrfs, you can even roll-back to original ext4 (and get back the same contents of ext4, as it was before initial conversion) ... however rollback only works until you have run btrfs-balance for the first time. You should preferably make a backup and boot from USB stick, when doing the conversion.
Please do more Videos on fedora and btrfs. Especially Backup Techniques in Case everything including the boot drive fails, Best Practices using Deja dup or other Backup tools -thank you!
Regarding btrfs RAID, the RAID1, RAID0, RAID1C3 (three copies), RAID1C4 (four copies), and RAID10 (mirror on stripe) configurations are all considered stable. The ones known to be unstable are the RAID5/6 strategies.
I've been using btrfs on Debian since bullseye was released and it's great. I have separate root, home, and snapshot subvolumes and can boot home and root snapshots with dmenu scripts.
I finally made it through this whole video after several attempts. I think I'm ready to make the switch and probably wont ever use any other advanced features for a desktop machine. I just was interested in using it on my Linux Mint desktop to have full drive compression with the simple fstab mount option for my NVME drive to squash down my steam library. Ark alone is 400GB on my drive and I did test btrfs out for a few minutes and it compressed down to 50% which is a huge saving of space. Like you said in the video I can use timeshift like I have already been but I will just tell it to use btrfs mode.
I've used BtrFS and honestly, I still don't like it. I like ZFS much more and I really wish they'd resolve the GPL vs CDDL issue at some point. ZFS is still more stable IMO and more reliable. Just not my cup of tea mate.
Btrfs WAS unstable hence the reputation it got and why no one wanted to use it. With ZFS around though, I don't really get why we would even use btrfs as the former has been tested and is production ready since the last decade, a thing btrfs can't claim yet. It is a nice project, but again re-inventing the wheel when there was a solution already (ZFS).
so BTRFS raid1 and raid10 are stable, so the work-around we've been doing is to make *metadata* raid1 or raid1c3 and have the data as raid5 or raid6 which solved the potential corruption issue.
Thank you, it's very useful! But for me (as a btrfs newbe) would be very important to see how to create subvolumes (the process of installation OS with btrfs). I'm still using ext4 ( because ,in my case, I know how to create ext4 partitions: / on SSD + /home on HDD ). In case with btrfs still can't understand it (
The vulgar pronunciation "Butter FS" is what the rubes call it. The reason Btrfs is called "B-tree FS" is because it employs a B-tree file directory structure. The "B" stands for "balance", because it provides a self-balancing binary search tree (BST). Although, B-trees aren't the only self-balancing BST around, they are commonly used. For example, Apple's APFS (HFS++) also uses B-trees to store most of it's volume metadata. Personally, I'm more excited to see XFS (the nameless file system) adopting the features for snapshots, CoW and deduplication. Hopefully, by then we won't have to listen to people mispronouncing Btrfs anymore. Maybe they'll even finally give "X" (for eXperimental) its name. AGFS, perhaps? Naw... That might confuse the Apple rubes.
Can I be THAT guy ? Why doesn't btrfs integrate with df and du ? Not having to learn a slew of new commands for each fs would be, oh, so neat. Also, how is performance, when using subvolumes and quotas and such ? Is it just a couple percent slower in heavy duty (big listings, logs writing, file retrievals by a server etc all at the same time) ? Or it's more than that ?
At 10.25 you mention about instability of btrfs. Probably true (I ve read many articles about it being inconsistent - short of) but some companies like Terramaster (one came to my mind right now) are using btrfs by default for their Nas devices.
Some history. . . BTRFS was developed by Oracle (not RH)" before they purchased Sun Microsystems who developed ZFS for Solaris UNIX. Oracle was developing BTRFS to be a performant filesystem on their fork of Linux that would support an Oracle DB environment. BTRFS was always behind in terms of features and stability compared to ZFS of which has long been stable for 10-12 years. ZFS was introduced in Solaris 10. I figured after the Sun purchase, they'd adopt ZFS and ditch BTRFS. It looks like Oracle has given the BTRFS project over to be open-sourced under an Apache 2.0 license. When Oracle controlled the project, back when, it was under a pseudo proprietary license (best my memory serves me). While an astute administrator has a workable and tested backup and recovery plan, having a stable and performant filesystem is equally important. I would prefer to rely on ZFS as a part of that plan than to plan for when BTRFS will fail or slowly rot data (which usually goes unnoticed), especially in a virtualized environment. ZFS provides the ability to take snapshots and able to move pools from one host to another, though a bit clunky when compared to truly clustered filesystems for Linux (e.g. Ceph, Gluster).
yes, thank you for the great video. If it were up to me, I would kindly request a more in-depth look at the mentioned feature of btrfs to extend storage by adding multiple physical drives (e.g. you have your /home folder on a separate drive, as you should, but then you want to add a second drive also for the /home folder... how would that look like in terms of btrfs?) thanks, and awesome video
The problem with btrfs is a bad architecture. Basically it stores the data belonging to the not existing volume object with the first physical disk of the volume. That is basically a beginner's fault and one of the reasons Red Hat stopped using btrfs. With this type of mixing objects, it will always be difficult to maintain a reliable system. Personally I only use btrfs in simple configurations. I have used btrfs with Peppermint 10 in Raid-0 on a Pentium 4 HT with IDE HDDs and I was happy with the results. The system booted in ~45 seconds. The free storage was very confusing and basically BS. I have used it too in my VMs like Fedora, but nowadays I install Fedora on ext4. That is not the fault of btrfs, but of the way I use VMs. Once per week I zero out the not used space to save space on my Host running OpenZFS (lz4 compressed). Zeroing out a compressed btrfs or OpenZFS on a compressed virtual disk of a VM takes close to an infinitive time. That is why I install my VMs always on ext4 :) Personally I prefer OpenZFS, the system has originally been designed by Sun Micro-system and it has a great architecture without the btrfs architectural blunder. Also the command line interface of ZFS is superior to that of btrfs. I have the feeling that ZFS is blocked by some companies for inclusion in the Linux kernel using the excuse of license issues. These license issues are just an excuse to block ZFS, since Ubuntu has integrated ZFS in Ubuntu since 16.04 and nobody sued Canonical. The same is true for many server systems like Proxmox; TrueNAS etc.
I've use btrfs in the past without quotas and had absolutely no problems with it but I wasn't using quotas. I think in the future I am going to stick with ext4 until they get the quota system worked out, that's a feature I would like to use but not going to play with experimental stuff. The snapshot feature was awesome though.
I have ran btrfs on a raid1 style for a while. It had been reliable and easy to work with, but I have noticed increase and decrease in performance when I go to new kernels.
Hey Jay, thanks for the video been waiting for this one since your chat about it on Lawrence Systems. I successfully installed arch today with btrfs, systemd boot, and luks. :D I was having problem with the kernel but once I downgraded to LTS everything was fine. I'm not too sure about the additonal flags in arch.conf for systemd-boot but it's working.
btrfs has been included in fedora for the last 23 releases (introduced with fedora 13 in 2010) but if you are still saying it isn't feature complete, i'll stay with ext4 a bit longer
Honestly I don't even know why this is brought up all day, every day, but BTRFS doesn't eat your data (anymore)... I use it in a RAID10 some time ago (online) converted from a RAID1. It was run in an extremely unsafe maner. I yanked out drives, had a m.2 to mpcie to 2 Sata adapter, that was losses put in place while sometimes punching those drives out of the array and I'm still here to tell you that I was always able to read my data. It easily recovered all of that mistreatment. So it just is sad to me that this false claim still persist. Even after doing horrible things to my array, btrfs always gave me the option to at the very least mount the array read only. It might not be as perfect as ZFS, but it's damn stable if you ask me.
To properly use Timeshift you need to install Fedora with the Rootsubvolume name of @ and the name of home as @home. Otherwise ts doesn't work. This can be done afterwards as well without any problems or recovery media. Mount the btrfs partition with the subvolid of 5 to mnt (mount -o subvolid=5 /dev/sda3 /mnt) Then go into that dir and move root to @ and home to @home. (cd /mnt; mv root @; mv home @home) Your running system won't even notice, check it out with mount (mount | grep " / ")! Now you'll need to change fstab to reflect the change as well. In particular just change the names from root and home to @ and @home. At last change your current active grub boot entry from subvol=root to subvol=@. Reboot and e voila, you did it. That's also how you would restore a snapshot on a running system, just without changing fstab and grub.
@@LampJustin what exactly doesn't work with "sane" names for my subvolumes? I have set it up with the default names as fedora uses them and Timeshift seemed to work. Or well, it didn't work well because I needed 4 subvolumes to be handled, therefore I switched to snapper
@@matthiasbendewald1803 I am not sure what you're trying to say, but for whatever reason TimeShift only supports doing snapshots for the @ and @home subvolume. Fedora instead names to root volume root (-> @) and the home volume home (-> @home). That's why I renamed them like stated before. I Infact also have more subvols but as expected TimeShift will completely disregard them.
I have good experience with btrfs in simple configurations. I have used it in VMs with e.g Fedora and it works fine. I moved with my wife to her Dominican Republic and the Pentium 4 is still an used CPU here. So twice I installed Peppermint on a Pentium 4 HT with two IDE HDDs using btrfs in Raid-0. The system booted in 40-45 seconds, so I was very happy with btrfs. However as retired architect I have problems with the architecture of btrfs. E.g it is missing the volume and virtual disk objects, so it stores some of that data with the first physical disk of that "volume". In my opinion it is an architectural beginner's error and one of the reasons, that it will remain difficult to maintain btrfs for the more complex configurations. I also don't like the mixing of sub-volume; snapshot and mount point names. A snapshot is NOT the same as a sub-volume! Personally I use and prefer OpenZFS, a file system originally designed by Sun Microsystems in 2005. ZFS has a superior architecture and an elegant CLI. It has 2 additional objects essential for the more complex configurations; the vdev (virtual device) and the datapool (volume). It keeps dataset (sub-volume) name and mount point folder name separated and a snapshot is NOT equal to a dataset. Snapshot storage (the history) is read-only, so you can't change the history! The last 10 years before retirement I worked in Brussels on the boundary of architecture/standardization and politics, so I developed some intuition for politics. So I have the feeling, that commercial interests of some companies, that rule the Linux Foundation, block the inclusion of OpenZFS in Linux. The excuse is the potential license problems, but Canonical integrated ZFS since Ubuntu 16.04 without being sued and the same is true for many commercial server products like Proxmox; TrueNas etc. Personally I will use btrfs only as OS filesystem in simple hardware configurations like single disk; Raid-0 and Raid-1. Here I can profit from the built in compression, less storage but more important less IO operations. Also snapshots and the integration in Timeshift are reasons to use btrfs there. I will use OpenZFS always for my data storage, since it is more elegant and has a proven reliability. In the recent past I also used OpenZFS for OS storage, but I changed with 22.04, because I have the feeling that Canonical stopped all development on booting from ZFS. Currently since 22.04 I boot the Host OS (a minimal install of Ubuntu 22.04 LTS) from a 16GB ext4 partition instead of reserving a full disk for OpenZFS. I consider to move the OS to a 16GB btrfs partition to save space, iops and to to use timeshift :)
You will need ecc-memory for ZFS, or so i read on the internet, because allegedly there is no inherent repair mechanism, unlike in ext4. So a small error on writing can lead to data corruption . I always wonder how that works when you have a power outage... You might get an error too, which might not be correctable by ECC...
@@johnp9890 There is no filesystem in the world, that can protect you from memory errors, for that you always need ECC memory. That is the reason, that for servers they always recommend using ECC memory and because ZFS is mainly used by servers, they often recommend it for ZFS too. The repair mechanism in ext4 or ntfs is inferior to OpenZFS or btrfs. Ext4 (or ntfs) are only able to repair inconsistencies in the disk administration by using their journal. Ext4 or ntfs are incapable to correct internal file errors caused by bitrot or crashes during rewriting the file. If you end up with half the new file and half the old file, there is a fair chance that the application can't read that file anymore. Ext4 or ntfs cannot correct errors in the 10,000 different internal file formats in use, a system crash during rewriting e.g a mp3 file can cause problems during playing that corrupted file. OpenZFS and btrfs are transactional file systems based on Copy on Write, so the file will be either the new version or the old version before the crash. In all cases the internal file format will be correct and you can still play e.g that mp3 file. You did not understand an essential part of the video, so I advice you to watch it again, staring at 4:00 or google for the Copy on Write mechanism.
@@bertnijhof5413 ok. But if you do not have ECC memory installed i read that ZFS might lose a bigger problem if a crash does occur, because allegedly it has less repair mechanisms then ext4. Or so some sources on the internet say....
@@johnp9890 there is nothing in ZFS that makes it inherently dependent on ECC memory. But a computer with ECC memory will be able to protect your data better than one without it, no matter what file system is used. So yes, for absolute maximum data integrity, use ECC memory. But even without ECC memory, ZFS is the safest PC file system that I am aware of. BTRFS would be the only other contender. But I’m a file system enthusiast and not an expert so I could have missed something. I think the “must use ECC memory myth comes from two places within the FreeNAS (now TrueNAS ) forums. That “knowledge” then spread until it became gospel that you had to use ECC memory with ZFS. First, as a file system built for “big iron” and focused on data integrity, it was natural that the systems it first shipped on had ECC memory. When it became open source and moved into the PC space, it appealed to people willing to sacrifice some performance for the extra data integrity it offered. With that focus, the only “proper” way to deploy ZFS was with ECC memory because that gave the greatest possible data integrity. So that was the standard recommend for anyone inquiring about what hardware to use. So slowly “it’s even better with” morphed into “you have use” ECC memory because that is how to make the most of ZFS. Second, in the FreeNAS forums there was a very active contributor who was fairly knowledgeable about hardware selection and various aspects of setting up your ZFS storage for your intended use case. He wrote an essay about how ECC memory was essential because otherwise one memory error could cause cascading failure because it would caused a mismatch with the data and metadata on disk, with one failure leading to another, to another, to another, and so. According to his essay, without ECC memory, your ZFS storage pool was effectively a ticking time bomb, just waiting for the first bit flip in memory to bring the whole house of cards crashing down. The problem is it wasn’t an accurate assessment of how ZFS works and the consequences of a memory error. To the extend that a memory error will affect what is stored on disk, ZFS is either less vulnerable or no more vulnerable than any other file system. The other issue raised by some people against ZFS is that it doesn’t offer the tools to repair a corrupted file system that are present for other file systems. The response from ZFS experts, of which I am not one, is that the types of problems those tools repair don’t appear on a ZFS system because it is designed not to have them. The types of problems that can occur on a ZFS system are so severe that they would be just as unrecoverable as on any other file system. One other myth about ZFS is that you need to have at least one gigabyte of memory for each terabyte of storage. ZFS caches as much of your data as it can, so the more memory that you give it, the greater the likelihood that your data can be retrieved from RAM rather than disk. RAM is much faster than even the fastest SSDs, so yes, more memory is a good thing for ZFS. How much you’ll benefit from more RAM will depend greatly on the workload, type of data your storing, and the layout data sets, disks, and vdevs that make up your storage pool.
@@williamp6800 To support your last paragraphs with a practical example: On my backup server I use FreeBSD 13.1 with OpenZFS. It runs on a Pentium 4 HT (1C2T; 3.0GHz); 1.5GB of DDR (400MHz) and 4 HDDs in total 1.21TB (2x IDE 3.5" and 2x SATA-1 2.5"). Normally it uses around 1GB of memory receiving the backup, but I must be honest OpenZFS bypasses the memory cache (L1ARC) for the backup. However I can also use the system as desktop without problems using e.g Firefox 101 :) The advantage: my investment was $0 and the disadvantage: the backup takes 1 to 2 hours, because it runs at 200Mbps instead of 1 Gbps, due to a ~95% load on one of the CPU threads. That 19 year old CPU is getting somewhat slow for these modern times :) :(
10:10 - RAID 0 and 1 work fine. Higher RAID levels are the problem. And RAID 0 and 1 do an excellent job of read (and in the case of 0) write striping. To me, they have seemed far faster than using, say, LVM for RAID1. 29:00 - btrf subvol list will list _all_ subvolumes on the btrfs filesystem that the specified directory exists on. AFAIK, there is no good way to list a subset of snapshots. As far as 'eating your data'... I really, really, really hate how filesystems are handled by people. One tiny problem with someone's bizarre edge case is enough to put everybody off using the filesystem for a decade. Any filesystem that's based on something completely new is going to have some rough edges for awhile. btrfs has been very, very stable for me, and the overall design is actually better for stability if your hard-drive doesn't lie to your OS because of COW. I've been using btrfs as my primary filesystem since not long after it made it into the kernel as a default option. I think something like 2010 or so. Using repeated read-only snapshots along with btrfs send and btrfs recv has been a fantastic boon for high-quality backups.
24:39 THANK YOU for calling a GiB a gibibyte and not a gigabyte. As someone who came from Windows I'm sick of Windows telling me KB when it actually means KiB, MB when it actually means MiB, and GB when it actually means GiB.
I use both Linux Mint and Ubuntu - the improvements to Timeshift enable reliable backups - only criticism is inability to store in a remote location - I can move the archive after as a workaround. So I will continue with ext4 for now but thanks for this.
Some of my friends and colleagues use Btrfs. This is mainly on simple configurations, such as on laptops and they are happy with it. With regard to Raid 5 and Raid 6, it is clear that there are problems with that. The video even advises against using Raid. Search for the channel "2.5 admins" in youtube and listen to episode 3 (03: Apache’s Btr than you think) from minute 16:30. There are also other episodes about btrfs (episode 41, 58) I just check the Btrfs wiki of Debian and this is still the case. (the second warning of "Other Warnings") This makes me stick with ext4, xfs, ...
Great video!!! Really helpfull! I am definitely going to give btrfs a try. Loved the sub-volume feature. Could you please update the btrfs feature status link. There is an frightening "OBSOLETE CONTENT" sign on top of the wiki page. 😨 Many Thanks!!!!!
Good info/vid! What about migrating an EXT4 to BTRFS? In-place utilities or copy/add/delete/adjust from one to the other. I guess the question might also be BRTRS to EXT4? Never seen that topic or topics addressed but I haven't researched to any extent -- link to existing process? I remember doing these file system changes on IBM propertary OS but I didn't operate in the Unix/Linux OS environment. Blown long 3 day holiday weekend process counting full backup, execution, backup to new file system schema.
32:20 to revert back to a snapshot, please don't modify the fstab. Instead mount the rootsubvolume with the id 5 to mount for example and move the home subvol to home-back and then do the reverse for the created snapshot. That way the name of the old snapshot is now home and will be mount at boot like normal. Also you should not fear doing that as when you move the old home subvolume the running system will just chug along. You can check it by taking a look at mount after doing the move ;)
I was going to as something along the lines of "... so if you switch the home to the snapshot , does that mean that the space needed to make home is now effectively 'useless' or 'wasted memory' space or does it get merged or overwritten with the home change --or am I thinking about this wrong?
I am not sure what you're saying, but if you go back to a snapshot that you took without deleting the original subvolume that you switched away from, then yes the difference between the snapshot-subvolume and the original subvolume gets wasted but since it's only the difference and every snapshot only holds the difference it's not much generally. But yes if you're sure that you don't need the org anymore then go ahead and delete it afterwards
@@LampJustin yeah, i did not ask the question well but, that is essentially what i meant- I saw how Jay explained the differences later towards the end of the video . Thanks for your reply
I've only watched half of the video but will continue tomorrow. I just wanned to thank you so much for making such in deepth videos for numbnuts like me. I don't know why I have such hard time grasping how Linux is built, but I wished I would have migrated to Linux from Windows a long time ago, then it probably would have been a lot easier. 🙏👌 I created my btrfs partitions while installing EndeavourOS but it seems like they are partitions and not subvolumes because my home partition and root doesn't share the same size. Can I reconfigure this afterwards or do I need to do it before I install EndeavourOS next time? I had devided a 960GB NVMe into 512mb fat32, 40Gb for / and the rest for home. Would be greatful if you give me a smallexplanation. Thx alot for a great Channel!
Please do make more videos on btrfs! especially on things like snapper, best practices, tricks with the file system that may be useful.
As to best practices go RAID 1 or 10 or Single if you don't need raid. If you can duplicate your metadata 3 times (RAID1(0)c3) Do a srub every month or two, doesn't need to be often and after that do a btrfs balance. That's pretty much all.
Tips: Use reflink copies whenever possible (cp --reflink always) It will create copy without physically copying the data. Only the new metadata will be created, similar to sym/hard links. It just doesn't come with any of the drawbacks. It's like dedpulication and can be used like that. ;)
@@LampJustin Thank you! I heard about reflinks but never really have an application for them at the moment. Oh and that raid setup 100%!
@@duser yeah it's really useful for dealing with VMs, cloning them with that flag is pretty nice and safes not only time but space as well. Good thing is that latest gnutils set the --reflink auto by default. So whenever you copy stuff it will get reflinked. Nautilus on Gnome also does the same
th-cam.com/users/shortsPF77M2V0UYs?feature=share
I can't believe you released a video about btrfs at the exact moment I wanted to know more about it!
lol same, I literally searched it yesterday
I can't believe it's not butter
same... I am planning to migrate my ext4 gentoo to btrfs since a while but been lazy to research bout btrfs. but now Jay solved it
Mmm, toast. Now with butter!
I'm late to the party, but here are a few things:
1. When taking a snapshot of a location that has a snapshot, are you not adding that existing snapshot into the new snapshot? Or is BTRFS smart enough to exclude snapshots?
2. Yes, please make more BTRFS videos
3. Please make BTRFS related videos with TimeShift, BackInTime, and Snapper as the focus
4. Please make a BTRFS video focusing on btrfs send | btrfs receive
1. BTRFS excludes own subvolumes. And snapshots are a special form of a subvolume.
My top btrfs configuration options on Arch: Create a subvolume at /home, /tmp, /var/log, /var/cache, /var/lib/flatpak, /var/lib/libvirt, /srv, /root, /opt, /usr/local and of course /.snapshots, also ideally you want to replace the root subvolume so I create another one and mount it at /. For mount options: noatime,compress=zstd,discard=async,space_cache=v2,ssd_spread. Disable CoW for /var/lib/libvirt for better performance. And most importantly, even if you take snapshots of your root subvolume, you won't be able to boot, so I have a pacman hook that triggers every time I upgrade my kernel, which runs rsync -a --delete /boot /.bootbackup. Small tip is to also delete the subvolid from fstab (it's not needed, subvol is enough) so you can easily restore a subvolume.
Yeah good recommends! But for me those would be too many subvols to keep track of and snapshot. I just have @ @home and libvirt images of course. I really don't like the naming but that's what you need for Timeshift........
Also yes you really shouldn't use subvolids in fstab or anywhere. Just renaming the subvols to rollback is a must.
The most optimal defaults differ form usecase to usecase. However it would be nice for fedora to step up their btrfs game, as it comes by default, the current one is an extremely vanilla setup and definitely less than ideal.
When it comes to btrfs, you are producing the best content available. Great stuff.
As usual, a very well researched and well-presented video from you. Good balanced overview of BTRFS, with the pros and cons, and where to looks for updates on that. It is great that a modern file system is being actively developed, and glad you highlighted it. Thanks Jay.
I'd certainly love to see more btrfs content. Thanks for all the great work!
I’ll consider doing more
My Synology is using BTRFS for about 4 years now in a SHR raid. Its working like a charm and even disk replacement and raid rebuild, went flawless. Stable, stellar performance and deduplication is an awesome feature.
Bro, I’m a little high right now so my sense of time is a little warped and it makes me pay much closer attention to your voice tones, and let me tell you… you have gotten a shit load better my brother.
Much props on the high quality effort on the script and prep work you did.
You really took the time to explain nuance.
I have bookmarked this to rewatch while not high lol
Great video! Thank you! And yes, I would like to see more advanced concepts of btrfs in future videos.
I am all for more BTRFS videos. I have been learning a lot from your channel. Thank you sir.
Excellent video Jay. Thank you very much. I have looked at BTRFS for a while, but never dove into it. Going to try it in a VM tonight. Please more videos on the subject. Also have you done videos on "backup best practices?" I would very much enjoy watching that. Thanks and Cheers from Canada!
It's only really RAID 5/6 (stripping) that is not completely stable with BTRFS. Actually, it's implementation of RAID1 is one of the best things about BTRFS for desktop computers in my opinion. Having installed an OS on two disks using RAID0, I was able to on-the-fly, without rebooting, in a matter of seconds, switch to RAID1. Something even the mighty zfs can't do for now (changing vdev structure).
This!
RAID 0 and RAID 1 are perfectly fine and work even better than md-raid.
It's just RAID 5/6 that are incomplete (although errors are extremely rare with proper use - similar guarantees as md-raid on Linux kernels prior to 4.4).
If RAID 5/6 had never been released, the BTRFS story would be really simple:
“It’s a very reliable, flexible, and full feature modern file system for the desktop, and and anywhere else you don’t need big pools of disks.”
As a bit of a ZFS zealot, I have to say BTRFS is really good.
"Something even the mighty zfs can't do for now (changing vdev structure)" Nonsense. ZFS can easily add/remove disks from stripes, and add mirrors to those disks--i.e., do exactly what you described that you did. But parity RAID vdevs can't (currently) be changed; that's been Coming Soon™ for a number of years.
@@danbrown586 are you saying ZFS can take a two disk stripe vdev and convert it to mirror vdev on the fly and back to stripe?
@@LucS0042 Well, there's no such thing as a "two disk stripe vdev"; that would be two single-disk vdevs striped together. But with that caveat, it seems a strange thing to want to do, but yes, ZFS can do it.
A lot of arch distros use BTRFS by default, that's how I found out about it. The snapshot feature has saved me from quite a few mishaps.
A really good video about btrfs.
btrfs is a great FS. I use it since 2012. Never had any problems with btrfs. N8ever used Raid5/6 - only the so called "RAID1". Maybe that is why I never run into any problems...) One of the best features is to do read only snapshots of our home directory and use them for making backups via send / receive to another btrfs device. (It took me some time to understand the construct of send / receive - but finally it works :)
BTRFS is awesome and it's much lighter than ZFS. I worked as a storage engineer for ~7 years and that's what I've learned during tons of tests and implementations - you DON'T need ZFS and BTRFS is really highly usable these days.
I used BTRFS for @, @home, and @data subvol. I do dual boot Ubuntu with windows. With BTRFS it can share file directly with WinBTRFS. My laptop small SSD can fit much of data with it's compression. Some game like DotA 2 compressed at almost 50% ratio. I install all my program at BTRFS, also set my fav folder like download, document to @data subvol. Data deduplication also work in Windows. That is very much better integration than old school ext4.
@@Julian-sj5tr compressing Dota 2 doesn't hurt performance? (fps)
@@TheGeoreyexactly my thought
Thoroughly enjoyed this video. Thanks for making content like this, Jay. I would love to see more content on the advanced features of btrfs.
thanks for the great video, btu I am curious what is the difference between ZFS and Btrfs, which one do you prefer?
ZFS, Btrfs losses data if you use anything but single drive, mirror, or stripe mode.
As far as I could tell from every other source, the BTRFS 'raid likes' are all fine and commonly used as defaults EXCEPT for raid5/6 implementations. But please do look at multiple closer-to-official accounts of btrfs raid profiles
Anything other than pure RAID 1 or RAID 0 will eat data and will probably never be implemented correctly.
Thank you for the information on BTRFS. I'm a new Linux user and i'm watching\learning as much as I can about all the various aspects of Linux.
Thanks - I learned a bunch
We have BTRFS running on a Synology NAS but EXT3 / EXT4 / ZFS everywhere else.
It was good to hear a bit on how BTRFS works.
Yes, please provide additional content on btrfs. Thanks!
Great video! I'm always interested in filesystems and ways how to manage it. It would be awesome to have a video on what are the optimal ways of managing a btrfs filesystems checking for errors, repairing, etc. I have btrfs on my main Linux machine and also on my NAS. I'm always on the lookout for best practices on how to secure my data.. especially when it comes to the filesystems I'm using which is btrfs. 😅
This video helped me understand BTRFS better. Great work!
...also Snapper is great. I set it up following a different TH-cam video.
It's awesome for people who want snapshots to be automated. It's in the RPM repos. (I'm also using Fedora.)
Love the content 👍🏼! Yes, kindly provide a refresh on BTRFS and it's applications such as for virtualization (proxmox,etc,.), improvements, practical uses and such. Apparently, some of the major tech giants use it in their environment and it's natively supported in the Linux kernel.
I found btrfs due to Fedora and Garuda. I love how Garuda implemented snapper, was very helpful when I was more experimental and overall just give me peace of mind. With Steam recently supporting CoW, would it be worth it to have my game partition as btrfs?
I write too long stuff.
TL;DR: Yes.
I use it now for few years for hundreds of games spanning over 3.5TB, btrfs just works.
I use @games subvolume, mounted at /mnt/games.
I use and recommend Steam flatpak, fully sandboxed (all access to my files revoked), gave access to /mnt/games:rw (read write permissions from flatpak), just works.
And it can’t mess with my personal files at all.
btrfs is the successor of ext4, it basically can do over 99% of what ext4 can do, but much more at same, little worse, or even better performance.
Of course they are extremely different fs, but for end user, I can’t say anything other than it just works.
When paired with ECC RAM.
But if not having ECC RAM, the other fs suffer the same, but they don’t tell you.
Btrfs on the other hand knows e.g. when your file integrity is broken. But btrfs trusts the CPU and the RAM, so it will tell you your disk is broken or that cosmic microwave background radiation hit your NAND cell and broke some data on the disk, it won’t tell you data read is correct, but the one in the RAM is faulty, it does not know that, it thinks errors detected are from the disk.
It assumes your CPU and RAM do not lie.
When using btrfs raid 1 or 10 (with data copy) it periodically checks file integrity automatically, cause it hashes or checksums each file (always, that’s how it detects errors), if it detects mismatch, it will automatically check the other disk and repair broken data automatically, cause it knows if something is broken, and in RAID 1 case has a fully functioning copy.
So awesome for system and documents and so on.
If you don’t care about games integrity you can just ignore that feature, but it is still a good fs.
You can use snapshotting, e.g. when modding a game, when an update broke your game mods, you can then look into snapshot and restore stuff or whatever.
Only limitation is your imagination.
It is better than ext4, allows resizing, and whole lot more features, comparable to zfs,.. so of course it is a good option.
It is also newer than zfs, in my view it is better than zfs, but it depends. E.g. BSD NAS should use zfs.
But for desktop, or server, always btrfs. But still, ext4 is good filesystem as well, as well as xfs, they are good choice as well.
For games it does not matter that much.
I recommend for system and home, two NVMe SSD, e.g. Micron 7400 Pro with power loss protection, using dm-crypt / luks on them, on top of that btrfs (in RAID 1 fully copy mode), and ECC RAM. Then maybe unencrypted additional SSD, e.g. FireCuda 530, btrfs for games.
ECC RAM is no bottleneck in gaming. I just ‚downgraded‘ speed, from 64kMiB 3.6GT/s cl16-19-19-19 to 128kMiB 3.2GT/s cl22-22-22 ECC RAM - literally no difference in FPS (I play at 4k).
If you have stutter in CSGO, right click, properties, compatibility, choose Steam Linux Runtime, working in Flatpak as well, they made it compatible, sandboxing works with native performance.
10:12 No you absolutely should use RAID with BTRFS, just not R5 or 6. RAID 1 and 10 are much superior to mdraid or hw raid. Just striped RAID setups suffer from write holes if you're not careful.
@@squishy-tomato fair, but I just wanted make it clear that RAID1(0) is indeed stable and totally usable. I already have some corruption from time to time, because of a bad controller. BTRFS does it's job great and corrects those flipped bits just fine. And yes I should definitely swap the controller, but cannot right now...
Finally, someone who managed to explain subvolumes in a way even my feeble mind could comprehend.
I was wandering what happened BTRFS .. it in fact has been around for decades and every now and again it raises its head and comes to the fore front with new features. I remember it back in SMS days when I worked there.. where is was played around in UNIX on Solaris. I have always liked the BTRFS concept and idea of use and I an happy to see it once again come to the fore front as commercially viable filesystem. Thank you for your introduction and viewing. I personally would like to see more about it and where it is going currently.
I would elaborate more about the benefits, purpose, and compelling reasons why an average user might need to use BTRS. A person has to be motivated enough to use or even try it. Because managing it, especially without the possibility of using GUI, looks pretty complex to the average home user.
He hinted GUI end use it when he mentioned TimeShift but he did not go into it. I agree with you that the average Linux beginner (myself for example) is not going to want to modify fstab files manually or run all these command lines.
I think Suse has developed a GUI.
Would love to learn more about what Fedora 36 does or doesn't do out of the box to protect your data. It's kinda important to know whether scrubbing and automatic snapshots are your responsibility or not.
Butter FS!!!!! Where is the Toasted Bread?????? Served with an FS on the side?
btrfs goes back many years ago. It was first used as a file manager for an accounting package. It has had an up and down history. One of the issues was once you used one application with btrfs you can not use another app using btrfs. It was a very popular file system many years ago and has reinvented itself and it's popularity thanks to Linux.
I use it for years on my very old laptop. Very satisfied, snapshots are great. And you can do things you can't with other fs even we have to be carefull
Yes btrfs is not raid
Thanks Sensei ! 25:23 if I understand correctly, you're saying that : "despite BTRFS subvolumes being stretchable, without its quotas feature, the system becomes unusable when every bit of subvolume freespace becomes redistributed/saturated (e.g because of a logfile out of control). And since as of 2024 the quotas feature is still not ready, BTRFS on LVM appears mandatory/convenient for : hot-resizing/hot-bay-expanding/hot-bay-removal/insertion in a LVM rig.
BTRFS itself seems enough for partitioning/isolating into subvolumes, but without LVM, users have to know beforehands the allocated storage sizes."
So now I wonder why I saw nobody on YT putting up a guide on such way of installing (e.g for a Debian install + switch to busybox in the process) :
efi + boot + PV,
then on PV, a LV for root and a LV for home at least.
BTRFS with subvolumes on LV root, BTRFS or EXT4 for LV home.
Snapper integration able to snapshot only few subs from LV root, as well as LV home.
(Optionally, LUKS sandwiched on LVM and under BTRFS would be the cherry on the top. So far, only skai nyght made it this way on YT. But him and EF (both Arch guys) encrypt swap in a way that prevents hibernation, so unconvenient for nomad laptops...)
Maybe I'm missing some content, i don't know,
otherwise feel free to hand out the idea to Tom if you already scheduled bunch of other content already ! :)
It would be awesome with a new btrfs where you advice on some efficient backup systems, how to interact with external drives, NAS and possibly even syncing across clients. Maybe one practice for server backup and one for desktop/laptop.
First, sorry for my poor English, i'm slightly dyslectic...
The only reason I still have 2 Windows computers, is a program called RollbackRx.
I have heard/read about BTRFS, and wondered if the works the same way as Rollback Rx
And your GREAT video, shows me, it is almost the same idea and same functions, though all from the terminal.
Thanks for showing some of this functions.
--------------------------
I have for several years used RollbackRx. for Windows
I bought 5 Pro licenses years ago and over time updatet to new edtions.
As I wrote I have only 2 Windows computers today, and Rollback are the main reason.
But I have had an eye on BTRFS for some time, but has been unsecure, since the sometimes negative critics.
With your video here, I can se that most of the reputation is mainly a userproblem.
(no critics to BTRFS from me, it is complicated, but not impossible.)
Good video. I find it much much less confusing and error prone to use the subvolume name than subvolid in fstab, and that's what I would recommend.
FWIW, timeshift also works seamlessly that way with no need to manually change fstab.
Also don't forget about the usefulness of reflink copies! It will make a deduplicated copy of a file, meaning that it will only copy the metadata of the file and link the data to the already present blocks. If you modify normal CoW will kick in and copy only the modified data. This is why you can copy multi gig isos on Fedora (within a mountpoint) in microseconds. Try it out with Nautilus ;) This really helped me restore movies from snapshots that I had accidentally deleted. You just need to mount the root subvolume to keep the copy within a mountpoint.
It sounds like btrfs is easier to implement that LVM. It would be nice to see a detailed comparison with use cases for both.
LVM is easier to implement then btrfs. It is probably easier to use btrfs, but LVM is easier to implement, and then you just add ordinary file systems on top of LVM.
But btrfs is combination of LVM and a modern fs. But it is actually another implementation based on ZFS, which is better then btrfs, as it is older and design by people that actually knows how to run huge servers. That said, btrfs features are not bad, and a bit lighter then ZFS. So all features in btrfs is in ZFS, but not everything in ZFS is implemented in btrfs (like RAID etc).
@@AndersJackson there is RAID in btrfs
LVM and btrfs are not for the same purpose. The first is a system for managing physical or logical volumes and disks but sencond is a file system. We can user btrfs in out LVM logical volumes.
Fantastic video. Please make a video on how to convert from existing ext4 install to a BTRFS instance. This would save so much time from reinstalling and configuring some linux systems. If that is possible!
Check "man btrfs-convert" for documentation.
It can convert an existing ext4 filesystem in-place into btrfs (without losing any data).
If you decide that you didn't like btrfs, you can even roll-back to original ext4 (and get back the same contents of ext4, as it was before initial conversion) ... however rollback only works until you have run btrfs-balance for the first time.
You should preferably make a backup and boot from USB stick, when doing the conversion.
Please do more Videos on fedora and btrfs. Especially Backup Techniques in Case everything including the boot drive fails, Best Practices using Deja dup or other Backup tools -thank you!
Regarding btrfs RAID, the RAID1, RAID0, RAID1C3 (three copies), RAID1C4 (four copies), and RAID10 (mirror on stripe) configurations are all considered stable. The ones known to be unstable are the RAID5/6 strategies.
Спасибо за урок! Всё что нужно по btrfs. Не чего лишнего, только самое нужное. Отличное видео.
I've been using btrfs on Debian since bullseye was released and it's great. I have separate root, home, and snapshot subvolumes and can boot home and root snapshots with dmenu scripts.
My Synology NAS uses butter-fs. Keep telling us more Jay. Many thanks laird Bill.
Is there any validity in btrfs reducing the lifespan of SSDs? Great video! Thanks
enjoyed the btrfs demonstration. fedora desktops default to btrfs. i had the server version defaulted to xfs
I finally made it through this whole video after several attempts. I think I'm ready to make the switch and probably wont ever use any other advanced features for a desktop machine. I just was interested in using it on my Linux Mint desktop to have full drive compression with the simple fstab mount option for my NVME drive to squash down my steam library. Ark alone is 400GB on my drive and I did test btrfs out for a few minutes and it compressed down to 50% which is a huge saving of space. Like you said in the video I can use timeshift like I have already been but I will just tell it to use btrfs mode.
butterfs comes from CoW's ;)
story checks out haha
Good explanation BTRFS, but I am sticking with EXT4 and LVM. They do all that I need. The snapshot feature is interesting.
I've used BtrFS and honestly, I still don't like it. I like ZFS much more and I really wish they'd resolve the GPL vs CDDL issue at some point. ZFS is still more stable IMO and more reliable.
Just not my cup of tea mate.
Btrfs WAS unstable hence the reputation it got and why no one wanted to use it. With ZFS around though, I don't really get why we would even use btrfs as the former has been tested and is production ready since the last decade, a thing btrfs can't claim yet. It is a nice project, but again re-inventing the wheel when there was a solution already (ZFS).
In love with the meme usage
Please make more videos on btrfs. Thanks
so BTRFS raid1 and raid10 are stable, so the work-around we've been doing is to make *metadata* raid1 or raid1c3 and have the data as raid5 or raid6 which solved the potential corruption issue.
Thank you, it's very useful! But for me (as a btrfs newbe) would be very important to see how to create subvolumes (the process of installation OS with btrfs). I'm still using ext4 ( because ,in my case, I know how to create ext4 partitions: / on SSD + /home on HDD ). In case with btrfs still can't understand it (
This video tells about creating subvolume during installation.
I don't know if this helps.
th-cam.com/play/PLwQD7rvZhbK-L6JlQwqOSP3DvMUmvbPvj.html
The vulgar pronunciation "Butter FS" is what the rubes call it. The reason Btrfs is called "B-tree FS" is because it employs a B-tree file directory structure. The "B" stands for "balance", because it provides a self-balancing binary search tree (BST). Although, B-trees aren't the only self-balancing BST around, they are commonly used. For example, Apple's APFS (HFS++) also uses B-trees to store most of it's volume metadata. Personally, I'm more excited to see XFS (the nameless file system) adopting the features for snapshots, CoW and deduplication. Hopefully, by then we won't have to listen to people mispronouncing Btrfs anymore. Maybe they'll even finally give "X" (for eXperimental) its name. AGFS, perhaps? Naw... That might confuse the Apple rubes.
Can I be THAT guy ? Why doesn't btrfs integrate with df and du ? Not having to learn a slew of new commands for each fs would be, oh, so neat.
Also, how is performance, when using subvolumes and quotas and such ? Is it just a couple percent slower in heavy duty (big listings, logs writing, file retrievals by a server etc all at the same time) ? Or it's more than that ?
Definitely want to see more about this topic.
At 10.25 you mention about instability of btrfs. Probably true (I ve read many articles about it being inconsistent - short of) but some companies like Terramaster (one came to my mind right now) are using btrfs by default for their Nas devices.
You can use blivet GUI tool to manage few operations of btrfs. It is handy when creating or deleting btrfs subvolume.
Excellent tutorial. I Would like to know more about the possibilities for more complex fs solutions implemented with btrfs. Thank you!
Some history. . . BTRFS was developed by Oracle (not RH)" before they purchased Sun Microsystems who developed ZFS for Solaris UNIX. Oracle was developing BTRFS to be a performant filesystem on their fork of Linux that would support an Oracle DB environment. BTRFS was always behind in terms of features and stability compared to ZFS of which has long been stable for 10-12 years. ZFS was introduced in Solaris 10. I figured after the Sun purchase, they'd adopt ZFS and ditch BTRFS. It looks like Oracle has given the BTRFS project over to be open-sourced under an Apache 2.0 license. When Oracle controlled the project, back when, it was under a pseudo proprietary license (best my memory serves me).
While an astute administrator has a workable and tested backup and recovery plan, having a stable and performant filesystem is equally important. I would prefer to rely on ZFS as a part of that plan than to plan for when BTRFS will fail or slowly rot data (which usually goes unnoticed), especially in a virtualized environment. ZFS provides the ability to take snapshots and able to move pools from one host to another, though a bit clunky when compared to truly clustered filesystems for Linux (e.g. Ceph, Gluster).
Yes, please more BTRFS vids!
yes, thank you for the great video. If it were up to me, I would kindly request a more in-depth look at the mentioned feature of btrfs to extend storage by adding multiple physical drives (e.g. you have your /home folder on a separate drive, as you should, but then you want to add a second drive also for the /home folder... how would that look like in terms of btrfs?) thanks, and awesome video
Cheers from Beazil. Thank you for this useful video
The best tutorial I have ever seem!! Congratulations!!!!
This was so great and so informative. Thank you very much!
Thank you. Please create more content about btrfs.
I hope to do exactly that :)
The problem with btrfs is a bad architecture. Basically it stores the data belonging to the not existing volume object with the first physical disk of the volume. That is basically a beginner's fault and one of the reasons Red Hat stopped using btrfs. With this type of mixing objects, it will always be difficult to maintain a reliable system. Personally I only use btrfs in simple configurations.
I have used btrfs with Peppermint 10 in Raid-0 on a Pentium 4 HT with IDE HDDs and I was happy with the results. The system booted in ~45 seconds. The free storage was very confusing and basically BS.
I have used it too in my VMs like Fedora, but nowadays I install Fedora on ext4. That is not the fault of btrfs, but of the way I use VMs. Once per week I zero out the not used space to save space on my Host running OpenZFS (lz4 compressed). Zeroing out a compressed btrfs or OpenZFS on a compressed virtual disk of a VM takes close to an infinitive time. That is why I install my VMs always on ext4 :)
Personally I prefer OpenZFS, the system has originally been designed by Sun Micro-system and it has a great architecture without the btrfs architectural blunder. Also the command line interface of ZFS is superior to that of btrfs. I have the feeling that ZFS is blocked by some companies for inclusion in the Linux kernel using the excuse of license issues. These license issues are just an excuse to block ZFS, since Ubuntu has integrated ZFS in Ubuntu since 16.04 and nobody sued Canonical. The same is true for many server systems like Proxmox; TrueNAS etc.
This is really great and very helpful. Thank you for sharing.
I've use btrfs in the past without quotas and had absolutely no problems with it but I wasn't using quotas. I think in the future I am going to stick with ext4 until they get the quota system worked out, that's a feature I would like to use but not going to play with experimental stuff. The snapshot feature was awesome though.
I have ran btrfs on a raid1 style for a while. It had been reliable and easy to work with, but I have noticed increase and decrease in performance when I go to new kernels.
Hey Jay, thanks for the video been waiting for this one since your chat about it on Lawrence Systems. I successfully installed arch today with btrfs, systemd boot, and luks. :D I was having problem with the kernel but once I downgraded to LTS everything was fine. I'm not too sure about the additonal flags in arch.conf for systemd-boot but it's working.
yep, this is great stuff, and a lot of updates have happened to the btrfs for a year ago :-)
btrfs has been included in fedora for the last 23 releases (introduced with fedora 13 in 2010) but if you are still saying it isn't feature complete, i'll stay with ext4 a bit longer
Great video! I really like to see more videos about btrfs.
Honestly I don't even know why this is brought up all day, every day, but BTRFS doesn't eat your data (anymore)... I use it in a RAID10 some time ago (online) converted from a RAID1. It was run in an extremely unsafe maner. I yanked out drives, had a m.2 to mpcie to 2 Sata adapter, that was losses put in place while sometimes punching those drives out of the array and I'm still here to tell you that I was always able to read my data. It easily recovered all of that mistreatment. So it just is sad to me that this false claim still persist. Even after doing horrible things to my array, btrfs always gave me the option to at the very least mount the array read only. It might not be as perfect as ZFS, but it's damn stable if you ask me.
I still avoid Btrfs as I had veeeeery slow and messy experiences with it 😱
It is slow for desktop use.
@Astrocat 3D Yes. Works great. Only avoid smr drives, I think it applies not only for synology but all platforms...
Great video! I would like to see an advanced video on how to create a full system backup & restore on Fedora 36, using btrfs/timeshift or snapper.
To properly use Timeshift you need to install Fedora with the Rootsubvolume name of @ and the name of home as @home. Otherwise ts doesn't work. This can be done afterwards as well without any problems or recovery media. Mount the btrfs partition with the subvolid of 5 to mnt (mount -o subvolid=5 /dev/sda3 /mnt) Then go into that dir and move root to @ and home to @home. (cd /mnt; mv root @; mv home @home) Your running system won't even notice, check it out with mount (mount | grep " / ")! Now you'll need to change fstab to reflect the change as well. In particular just change the names from root and home to @ and @home. At last change your current active grub boot entry from subvol=root to subvol=@. Reboot and e voila, you did it. That's also how you would restore a snapshot on a running system, just without changing fstab and grub.
@@LampJustin what exactly doesn't work with "sane" names for my subvolumes? I have set it up with the default names as fedora uses them and Timeshift seemed to work. Or well, it didn't work well because I needed 4 subvolumes to be handled, therefore I switched to snapper
@@matthiasbendewald1803 I am not sure what you're trying to say, but for whatever reason TimeShift only supports doing snapshots for the @ and @home subvolume. Fedora instead names to root volume root (-> @) and the home volume home (-> @home). That's why I renamed them like stated before. I Infact also have more subvols but as expected TimeShift will completely disregard them.
@@LampJustin okay then I must have mixed things up. Maybe I had used that rsync option in Timeshift? Whatever, thanks for the answer!
it's neat when it works...
I have good experience with btrfs in simple configurations. I have used it in VMs with e.g Fedora and it works fine. I moved with my wife to her Dominican Republic and the Pentium 4 is still an used CPU here. So twice I installed Peppermint on a Pentium 4 HT with two IDE HDDs using btrfs in Raid-0. The system booted in 40-45 seconds, so I was very happy with btrfs.
However as retired architect I have problems with the architecture of btrfs. E.g it is missing the volume and virtual disk objects, so it stores some of that data with the first physical disk of that "volume". In my opinion it is an architectural beginner's error and one of the reasons, that it will remain difficult to maintain btrfs for the more complex configurations. I also don't like the mixing of sub-volume; snapshot and mount point names. A snapshot is NOT the same as a sub-volume!
Personally I use and prefer OpenZFS, a file system originally designed by Sun Microsystems in 2005. ZFS has a superior architecture and an elegant CLI. It has 2 additional objects essential for the more complex configurations; the vdev (virtual device) and the datapool (volume). It keeps dataset (sub-volume) name and mount point folder name separated and a snapshot is NOT equal to a dataset. Snapshot storage (the history) is read-only, so you can't change the history!
The last 10 years before retirement I worked in Brussels on the boundary of architecture/standardization and politics, so I developed some intuition for politics. So I have the feeling, that commercial interests of some companies, that rule the Linux Foundation, block the inclusion of OpenZFS in Linux. The excuse is the potential license problems, but Canonical integrated ZFS since Ubuntu 16.04 without being sued and the same is true for many commercial server products like Proxmox; TrueNas etc.
Personally I will use btrfs only as OS filesystem in simple hardware configurations like single disk; Raid-0 and Raid-1. Here I can profit from the built in compression, less storage but more important less IO operations. Also snapshots and the integration in Timeshift are reasons to use btrfs there. I will use OpenZFS always for my data storage, since it is more elegant and has a proven reliability.
In the recent past I also used OpenZFS for OS storage, but I changed with 22.04, because I have the feeling that Canonical stopped all development on booting from ZFS. Currently since 22.04 I boot the Host OS (a minimal install of Ubuntu 22.04 LTS) from a 16GB ext4 partition instead of reserving a full disk for OpenZFS. I consider to move the OS to a 16GB btrfs partition to save space, iops and to to use timeshift :)
You will need ecc-memory for ZFS, or so i read on the internet, because allegedly there is no inherent repair mechanism, unlike in ext4. So a small error on writing can lead to data corruption . I always wonder how that works when you have a power outage... You might get an error too, which might not be correctable by ECC...
@@johnp9890 There is no filesystem in the world, that can protect you from memory errors, for that you always need ECC memory. That is the reason, that for servers they always recommend using ECC memory and because ZFS is mainly used by servers, they often recommend it for ZFS too.
The repair mechanism in ext4 or ntfs is inferior to OpenZFS or btrfs. Ext4 (or ntfs) are only able to repair inconsistencies in the disk administration by using their journal. Ext4 or ntfs are incapable to correct internal file errors caused by bitrot or crashes during rewriting the file. If you end up with half the new file and half the old file, there is a fair chance that the application can't read that file anymore. Ext4 or ntfs cannot correct errors in the 10,000 different internal file formats in use, a system crash during rewriting e.g a mp3 file can cause problems during playing that corrupted file.
OpenZFS and btrfs are transactional file systems based on Copy on Write, so the file will be either the new version or the old version before the crash. In all cases the internal file format will be correct and you can still play e.g that mp3 file.
You did not understand an essential part of the video, so I advice you to watch it again, staring at 4:00 or google for the Copy on Write mechanism.
@@bertnijhof5413 ok. But if you do not have ECC memory installed i read that ZFS might lose a bigger problem if a crash does occur, because allegedly it has less repair mechanisms then ext4. Or so some sources on the internet say....
@@johnp9890 there is nothing in ZFS that makes it inherently dependent on ECC memory. But a computer with ECC memory will be able to protect your data better than one without it, no matter what file system is used.
So yes, for absolute maximum data integrity, use ECC memory. But even without ECC memory, ZFS is the safest PC file system that I am aware of. BTRFS would be the only other contender. But I’m a file system enthusiast and not an expert so I could have missed something.
I think the “must use ECC memory myth comes from two places within the FreeNAS (now TrueNAS ) forums. That “knowledge” then spread until it became gospel that you had to use ECC memory with ZFS.
First, as a file system built for “big iron” and focused on data integrity, it was natural that the systems it first shipped on had ECC memory. When it became open source and moved into the PC space, it appealed to people willing to sacrifice some performance for the extra data integrity it offered. With that focus, the only “proper” way to deploy ZFS was with ECC memory because that gave the greatest possible data integrity. So that was the standard recommend for anyone inquiring about what hardware to use.
So slowly “it’s even better with” morphed into “you have use” ECC memory because that is how to make the most of ZFS.
Second, in the FreeNAS forums there was a very active contributor who was fairly knowledgeable about hardware selection and various aspects of setting up your ZFS storage for your intended use case. He wrote an essay about how ECC memory was essential because otherwise one memory error could cause cascading failure because it would caused a mismatch with the data and metadata on disk, with one failure leading to another, to another, to another, and so.
According to his essay, without ECC memory, your ZFS storage pool was effectively a ticking time bomb, just waiting for the first bit flip in memory to bring the whole house of cards crashing down. The problem is it wasn’t an accurate assessment of how ZFS works and the consequences of a memory error. To the extend that a memory error will affect what is stored on disk, ZFS is either less vulnerable or no more vulnerable than any other file system.
The other issue raised by some people against ZFS is that it doesn’t offer the tools to repair a corrupted file system that are present for other file systems. The response from ZFS experts, of which I am not one, is that the types of problems those tools repair don’t appear on a ZFS system because it is designed not to have them. The types of problems that can occur on a ZFS system are so severe that they would be just as unrecoverable as on any other file system.
One other myth about ZFS is that you need to have at least one gigabyte of memory for each terabyte of storage. ZFS caches as much of your data as it can, so the more memory that you give it, the greater the likelihood that your data can be retrieved from RAM rather than disk. RAM is much faster than even the fastest SSDs, so yes, more memory is a good thing for ZFS.
How much you’ll benefit from more RAM will depend greatly on the workload, type of data your storing, and the layout data sets, disks, and vdevs that make up your storage pool.
@@williamp6800 To support your last paragraphs with a practical example: On my backup server I use FreeBSD 13.1 with OpenZFS. It runs on a Pentium 4 HT (1C2T; 3.0GHz); 1.5GB of DDR (400MHz) and 4 HDDs in total 1.21TB (2x IDE 3.5" and 2x SATA-1 2.5"). Normally it uses around 1GB of memory receiving the backup, but I must be honest OpenZFS bypasses the memory cache (L1ARC) for the backup. However I can also use the system as desktop without problems using e.g Firefox 101 :)
The advantage: my investment was $0 and the disadvantage: the backup takes 1 to 2 hours, because it runs at 200Mbps instead of 1 Gbps, due to a ~95% load on one of the CPU threads. That 19 year old CPU is getting somewhat slow for these modern times :) :(
Very well presented. Thanks for your hard work.
More btrfs content please!
10:10 - RAID 0 and 1 work fine. Higher RAID levels are the problem. And RAID 0 and 1 do an excellent job of read (and in the case of 0) write striping. To me, they have seemed far faster than using, say, LVM for RAID1.
29:00 - btrf subvol list will list _all_ subvolumes on the btrfs filesystem that the specified directory exists on. AFAIK, there is no good way to list a subset of snapshots.
As far as 'eating your data'... I really, really, really hate how filesystems are handled by people. One tiny problem with someone's bizarre edge case is enough to put everybody off using the filesystem for a decade. Any filesystem that's based on something completely new is going to have some rough edges for awhile.
btrfs has been very, very stable for me, and the overall design is actually better for stability if your hard-drive doesn't lie to your OS because of COW.
I've been using btrfs as my primary filesystem since not long after it made it into the kernel as a default option. I think something like 2010 or so.
Using repeated read-only snapshots along with btrfs send and btrfs recv has been a fantastic boon for high-quality backups.
24:39 THANK YOU for calling a GiB a gibibyte and not a gigabyte. As someone who came from Windows I'm sick of Windows telling me KB when it actually means KiB, MB when it actually means MiB, and GB when it actually means GiB.
I use both Linux Mint and Ubuntu - the improvements to Timeshift enable reliable backups - only criticism is inability to store in a remote location - I can move the archive after as a workaround. So I will continue with ext4 for now but thanks for this.
Some of my friends and colleagues use Btrfs. This is mainly on simple configurations, such as on laptops and they are happy with it.
With regard to Raid 5 and Raid 6, it is clear that there are problems with that. The video even advises against using Raid. Search for the channel "2.5 admins" in youtube and listen to episode 3 (03: Apache’s Btr than you think) from minute 16:30. There are also other episodes about btrfs (episode 41, 58)
I just check the Btrfs wiki of Debian and this is still the case. (the second warning of "Other Warnings")
This makes me stick with ext4, xfs, ...
Great content as always!
Huge thank you 🙏
A very very nice video, TNX👏👏👏
Great video!!! Really helpfull!
I am definitely going to give btrfs a try. Loved the sub-volume feature.
Could you please update the btrfs feature status link. There is an frightening "OBSOLETE CONTENT" sign on top of the wiki page.
😨
Many Thanks!!!!!
Top, Jay! Thanks for your explanation! :-)
Thanks for the vid as always.
Pls mkvid about migrating existing ext4 root and home partition to btrfs.
Given that btrfs is like ext4 but better, I'm of the opinion that it should be called better fs, because it's just better, not butter.
Good info/vid! What about migrating an EXT4 to BTRFS? In-place utilities or copy/add/delete/adjust from one to the other. I guess the question might also be BRTRS to EXT4?
Never seen that topic or topics addressed but I haven't researched to any extent -- link to existing process?
I remember doing these file system changes on IBM propertary OS but I didn't operate in the Unix/Linux OS environment. Blown long 3 day holiday weekend process counting full backup, execution, backup to new file system schema.
32:20 to revert back to a snapshot, please don't modify the fstab. Instead mount the rootsubvolume with the id 5 to mount for example and move the home subvol to home-back and then do the reverse for the created snapshot. That way the name of the old snapshot is now home and will be mount at boot like normal. Also you should not fear doing that as when you move the old home subvolume the running system will just chug along. You can check it by taking a look at mount after doing the move ;)
I was going to as something along the lines of "... so if you switch the home to the snapshot , does that mean that the space needed to make home is now effectively 'useless' or 'wasted memory' space or does it get merged or overwritten with the home change --or am I thinking about this wrong?
I am not sure what you're saying, but if you go back to a snapshot that you took without deleting the original subvolume that you switched away from, then yes the difference between the snapshot-subvolume and the original subvolume gets wasted but since it's only the difference and every snapshot only holds the difference it's not much generally. But yes if you're sure that you don't need the org anymore then go ahead and delete it afterwards
@@LampJustin yeah, i did not ask the question well but, that is essentially what i meant- I saw how Jay explained the differences later towards the end of the video . Thanks for your reply
@@TroyYounts you're welcome
Simple use RAID 1 on LVM and do BTRFS over LVM.
I've only watched half of the video but will continue tomorrow. I just wanned to thank you so much for making such in deepth videos for numbnuts like me. I don't know why I have such hard time grasping how Linux is built, but I wished I would have migrated to Linux from Windows a long time ago, then it probably would have been a lot easier. 🙏👌
I created my btrfs partitions while installing EndeavourOS but it seems like they are partitions and not subvolumes because my home partition and root doesn't share the same size. Can I reconfigure this afterwards or do I need to do it before I install EndeavourOS next time? I had devided a 960GB NVMe into 512mb fat32, 40Gb for / and the rest for home.
Would be greatful if you give me a smallexplanation. Thx alot for a great Channel!
Looks like pop os 22.10 will switch to btrfs as default too. They're working on tools for creating and restoring snapshots.